CA2337079C - A search system and method for retrieval of data, and the use thereof in a search engine - Google Patents
A search system and method for retrieval of data, and the use thereof in a search engine Download PDFInfo
- Publication number
- CA2337079C CA2337079C CA002337079A CA2337079A CA2337079C CA 2337079 C CA2337079 C CA 2337079C CA 002337079 A CA002337079 A CA 002337079A CA 2337079 A CA2337079 A CA 2337079A CA 2337079 C CA2337079 C CA 2337079C
- Authority
- CA
- Canada
- Prior art keywords
- words
- text
- query
- word
- sequences
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
- Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10S707/00—Data processing: database and file management or data structures
- Y10S707/99931—Database or file accessing
- Y10S707/99933—Query processing, i.e. searching
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
- Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10S707/00—Data processing: database and file management or data structures
- Y10S707/99931—Database or file accessing
- Y10S707/99933—Query processing, i.e. searching
- Y10S707/99935—Query augmenting and refining, e.g. inexact access
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
- Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10S707/00—Data processing: database and file management or data structures
- Y10S707/99931—Database or file accessing
- Y10S707/99933—Query processing, i.e. searching
- Y10S707/99936—Pattern matching access
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
- Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10S707/00—Data processing: database and file management or data structures
- Y10S707/99931—Database or file accessing
- Y10S707/99937—Sorting
Abstract
A search system for information retrieval comprises a data structure in the form of a non-evenly spaced sparse suffix tree for storing suffixes of words and/or symbols, or sequences thereof, in a text T, a metric M comprising combining edit distance metrics for an approximate degree of matching respectively between words and/or symbols, or between sequences thereof, in the text T and a query Q, the latter distance metric including weighting cost functions for edit operations which transform a sequence S of the text into a sequence P
of the query Q, and search algorithms for determining the degree of matching respectively between words and/or symbols, or between sequences thereof, in respectively the text T and the query Q, such that information R is retrieved with a specified degree of matching with the query Q.
Optionally the search system also comprises algorithms for determining exact matching such that information R may be retrieved with an exact degree of matching with the query Q. A method in the search system comprises generating the data structure as a word-spaced sparse suffix tree, storing sequence information of the words in the text T in the generated suffix tree, generating a combined edit distant metric for words or sequences thereof in the text T and a query word q or sequences thereof in the query Q and including word-weighting cost functions for the sequence-transforming edit operations, and determining the degree of matching between retrieved information R and a query Q. - Use in an approximate search engine.
of the query Q, and search algorithms for determining the degree of matching respectively between words and/or symbols, or between sequences thereof, in respectively the text T and the query Q, such that information R is retrieved with a specified degree of matching with the query Q.
Optionally the search system also comprises algorithms for determining exact matching such that information R may be retrieved with an exact degree of matching with the query Q. A method in the search system comprises generating the data structure as a word-spaced sparse suffix tree, storing sequence information of the words in the text T in the generated suffix tree, generating a combined edit distant metric for words or sequences thereof in the text T and a query word q or sequences thereof in the query Q and including word-weighting cost functions for the sequence-transforming edit operations, and determining the degree of matching between retrieved information R and a query Q. - Use in an approximate search engine.
Description
CA 02337079 2001-Ol-06 . . t ( ( t t t ( t ~ ( ~ t t v . ~ n . . ' t 1 f < 1 . t 1 t t t t f ! t L t ( ( E
t f ( P ; 1 f t ( . C ( t ( ( t t ( ( t t t C f f f '( CE ft 1f' ~i~i f(' f' ft' E'f A search system and method for retrieval of data, and the use thereof in a search engine The present invention concerns a search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words and/or symbols and sequences thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R, comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric Mwhich measures the degree of matching between the query Q and retrieved information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords kw; a method in a search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words .and symbols and sequences thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric Mwhich measures the degree of matching between the query Q and retrieved. information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords krv, wherein the information in the text is divided into words and word sequences, the words being substrings of the entire text separated by word boundary terms and forming a sequence of symbols, and wherein each word is structured as a sequence of symbols.
The invention also concern the use of the search system.
A tremendous amount of information in various fields of human knowledge is collected and stored in computer memory systems. As the coiilputer memory systems increasingly are linked in public available data communication networks, there has been an increasing effort to develop systems and methods for searching and retrieving information for public or personal use. Present search methods for data have, however, limitations that t~t~Nt3ED SHE
CA 02337079 2001-Ol-06 t ~ t f t t t f' t t t t t t I
a r r t l . ~' to E f i f t t fJr l E E t i ~ t g i , ~r f t t t r f t t l f f' t ' f t t f C % f t F C l i t f E ~ f f' E C f' t t l f seriously reduce the possibility of efficiently retrieving and using information stored in this manner.
Information may be stored in the form of different data types, and in the context of information search and retrieval it will be useful to discern 5 between dynamic data and static data. Dynamic data is data that change often and continuously, so that the set of valid data varies all the time, while static data only changes very seldom or never at all. For instance will economic data, such as stock values, or meteorological data be subject to very quick changes and hence dynamic. On the other hand archival storage of books and documents are usually permanent and static data. The concept the volatility of the data relates to how long the information is valid. The volatility of data has some bearing upon how the information should be searched and retrieved. Large volumes of data require some structure in order to facilitate searching, but the time cost of building such structures must not be higher than the time the data is valid. The cost of building a structure is dependent on the data volume and hence the building of data structures for searching the information should take both the data volume and the volatility into consideration. The information collected are stored in databases and these may be structured or unstructured. Moreover, the databases may contain several types of documents, including compound documents which contain images, video, sound and formatted or annotated text. Particularly strucfiured databases are usually furnished with indexes in order to facilitate searching and retrieving the data. The growth of the World Wide Web (WWW) offers a steadily growing collection of compound and hyperlinked documents. A great many of these are not collected in structured databases and no indexes facilitating rapid searching are available. However, the need for searching documents in the World Wide Web is obvious and as a result a number of so-called~search engines has been developed, enabling searching at Ieast parts of the information in the World Wide Web.
With a search engine it is commonly understood one or more tools for searching and retrieving information. In addition to the search system proper, a search engine also contains an index, for instance comprising test from a large number of uniform resource locators (URLs ). Examples of such search engines are Alta Vista, HotBot with Inktomy technology, Infoseek, Excite and Yahoo. All these offer facilities for performing search and retrieval of information in the World Wide Web. However, their speed and efficiency do ~
CA 02337079 2001-Ol-06 t . , r t f . t 1 S t t f t ( f ~ t t f t r ~ E r r z r , . r r t 1 t i f f t ~ f E I t t C 1 i f ' 1 ! Y C ! t E . i .f r i t f & F .f E Y t C t E F f L f t t f r , ' by no means match the huge amount of information available on the World Wide Web and hence the search and retrieval efficiency of these search engines leaves rnuch.to be desired.
Searching a large collection of text documents can usually be done with several query types. The most common query type is matching and variants of ' this. By specifying a keyword or set of keywords that has to be present in the queried information the search system retrieves all documents that fulfils this requirement. The basic search method is based on so-called single keyword matching. The keywordp is searched for and all documents containing this word shall be retrieved. It is also possible to search for a keyword prefix p~
and all documents where this prefix is present in any keyword in the documents, will be retrieved. Instead of searching with keywords, the search is sometimes based on so-called exact phrase matching, where the search uses several single keywords in particular sequence. As well-known by persons skilled in the art, the exact matching of keyword phrases in many search systems may be done with the use of Boolean operators, for instance based on operators such as AND, OR, and NOT which allow a filtering of the information; e.g. using an AND phrase results in that all documents containing the two leeywords linked by the AND operator will be returned.
Also a NEAR operator has been used for returning just the documents with the keywords matching and located "near" to each other in the document text.
In many structured database the documents contained in the database have been annotated, e.g. provided with fields vcThich denote certain parts or types of information in the document. This allows the search for matches in only parts of the documents and is useful when the type of queried information is known in advance.
When searching in text documents the data are structured and most likely present in some natural language, like English, Norwegian etc. When searching for documents with a certain context it is possible to apply proximity metrics for matching keywords or phrases that match the query approximately. Allowing errors in keywords and phrases are common method for proximity, using a thesaurus is another common method. A proximity search requires only that there shall be a partial match between the information retrieved and the query. International published application W096/00945 titled "Variable length data sequence matching method and apparatus" (Doringer & al.) which has been assigned to International AH~~'~~ S~ C
CA 02337079 2001-Ol-06 _ t F t S f i _ h t . t r r f r r t . t r a4 t t t ~ r t f f r t t c c t f a r a d t r F i t ' E f 4 t f F ( f t C ( F I C ~ r i t t r f f f f C' ~ f t f f t f E t c Business Machines, Corp., discloses the building, maintenance and use of a database with a trie-like structure for storing entries and retrieving at least a partial.match, preferably the longest partial match or all partial matches of a search argument (input key) from the entries.
In order to further illuminate the general prior art mention can be made of international published patent application W092/15954 (KimbaIL & al., assigned to Red Brick System, U.S.A.) and US patent no. 5 627 74~ (Baker & al., assigned to Lucent Technologies, Inc., U.S.A.),'both disclosing data structures in the form of suffix trees for searching/matching in a square matrix. Neither of these two publications disclose anything beyond a regular suffix tree, except for the use of a linked list during matching and do not teach or suggest approaches to limit the search space when searching for approximate matches. However, such approaches would be most desirable when applying data structures based on suffix trees to searching, particularly for approximate matches in extremely large document collections, such as may be found on the World Wide Web.
The main object of the present invention is thus to provide a search system and a method for fast and efficient search and retrieval of information in . large volumes of data. Particularly it is an object of the present invention to provide a search system suited for implementing search engines for searching of information systems with distributed large volume data storage, for instance Internet. It is to be understood that the search system according to the invention by no means shall be limited to searching and retrieving information stored in the form of alphanumeric symbols, but equally well may be applied to searching and retrieving information stored in the form of digitalized images and graphic symbols, as the word text used herein also may interpreted as images when these are represented wholly or partly as sets of symbols. It is also to be understood that the search system according to the invention can be implemented as software written in a suitable high-level language on commercially available computer systems, but it may also be implemented in the form of a dedicated processor device for searching and retrieving information of the aforementioned lcind.
The above-mentioned objects and advantages are realized according to the invention with a search system which is characterized in that the data structure comprises a tree structure in the form of a non-evenly spaced sparse ~t~~N~D S~H~~~
CA 02337079 2001-Ol-06 , . r . " ' ~ c ~ t c i i ~ r t r , C t t s s r t , t t i f f r t ; t a t f ~ f t i t i f i f a ; i r' f F i f ( f ( t t C t ( t f i f f < . i' t ~ ~C ; f ; f t i Y C < i t suffix tree ST(T) for storing suffixes of words and/or symbols, and sequences thereof in the text T, that the metric M comprises a combination of an edit distance metric for an approximate degree of matching between words and/or symbols in the text T and a query Q and an edit distance metric for an approximate degree of matching between sequences S of words and/or symbols in the text T and a query sequence l' of words and/or symbols in the query Q, the latter edit distance metric including weighting cost functions for edit operations which transform a sequence S' of words and/or symbols in the text T into the sequence P of words and/or symbols in the query Q, the weighting taking place with a value proportional to a change in the length of sequence upon a transformation or dependent on the size of the words and/or symbols in sequences to be matched, that the implemented search algorithms comprise a first algorithm for determining the degree of matching between words and/or symbols in the suffix tree representation of respectively the text T and a query Q, and a second algorithm for determining the degree of matching between sequences of words and/or symbols iri the suffix tree representation of respectively the text T and the query Q, said first and/or second algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words or sequences of symbols or combinations thereof, such that information R is retrieved on the b asis of query Q with a specified degree of matching between the former and the latter, and that the search algorithms optionally also comprise a third algorithm for determining exact matching between words and/or symbols in the suffix tree representation of respectively the text T and the query Q
and/or a fourth algorithm for determining exact matching between sequences of words and/or symbols in the suffix tree representation of respectively the text T and the query Q, said third and/or fourth algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words, or sequences of symbols or combinations thereof, such that information R is retrieved on the basis of the query Q with an exact matching between the former and the latter.
In an advantageous embodiment of the search system according to the invention the suffix tree ST(T) is a word-spaced sparse suffix tree SSTWS(T), comprising only a subset of the suffixes in the text T.
Preferably is then the word space sparse suffix tree SST~,S(T) a keyword-spaced sparse suffix tree SST~,S(T).
AA~~fD Si-~E~
°
CA 02337079 2001-Ol-06 a t i q f - f t f f f ( a f f i t . (. t f n 1 ' G fj~ t F f f ~ t i f fl f f f f f t f t t i t F ( t t 6 : f t i F f f i t f t f c f F ~ i f 6 f f t E i F F Q t ~: f !~
In further advantageous embodiments of the search system according to the invention the first-algorithm for detecting the degree of keyword matching in a keyword-spaced sparse suffix tree SSTkWS(T) is implemented as disclosed by dependent claim 4, the second algorithm for determining the degree of sequence matching in a keyword-spaced sparse suffix tree SSTkWS(T) implemented as disclosed by dependent claim 5, whereby a subroutine of the second algorithm preferably is implemented as disclosed by dependent claim 6, the third algorithm for determining an exact keyword matching in a keyword-spaced sparse suffix tree SSTkWS(T) implemented as disclosed by dependent claim 7, and finally the fourth algorithm for determining an exact keyword sequence matching in a keyword-spaced sparse suffix tree SSTkWs implemented as disclosed by dependent claim 8.
The above-mentioned objects and advantages are also realized according to the invention with a method which is characterized by generating the data structure as a word-spaced sparse suffix tree SSTWS(T) of a text T for representing all the suffixes starting at a word separator symbol in the text T, storing sequence information of the words in the text T in the word-spaced sparse suffix tree SSTWS(T), generating a combined edit distance metric M
comprising an edit distance metric D(s, q) for words s in the text T and a query word q in a query Q and a word-size dependent edit distance metric DWS(S,P) for sequences S of words in the text T and a sequence P of words q in the query Q, the edit distance metric DWS(S,P) being the minimum sum of costs for edit operations transforming the sequence S into the sequence P, the minimum sum of costs being the minimum sum of cost functions for each edit operation weighted by a parameter proportional to the change in the total length of the sequence S or by the ratio of the current word length and average word length in the sequences S;P, and determining the degree of matching between words s, q by calculating the edit distance. D(s, q) between the words s of the retrieved informatiomR and the word q of a query Q, or in case the words s,q are more than k errors from each other, determining the degree of matching between the word sequences SR, PO of retrieved information R and a query Q respectively by calculating the edit distance DWS(SR,Pg) for all matches.
Advantageously the method according to the invention additionally comprises weighting an edit operation which changes a word s into word q with a parameter for the proximity between the characters of the words s; q, r~~9~Nt~D SHE~~
' CA 02337079 2001-Ol-06 E i' i' ( f ( t t ' C ( t 1 t l f f ~'',~ ( C a E~ t t t ( /F i t t p t F I t t t i t a C C.
f C C C ( f Y f C ( t t (' C C P t ( i C f f f C C ~ C f C t thus taking the similarity of the words s; q in regard when determining the cost of the edit operation in question. .
In an advantageous embodiment of the method according to the invention the number of matches is limited by calculating the edit distance by calculating the edit distance DWS(SR,Po) for restricted number of words in the query word sequence Pg.
In another advantageous embodiment of the method according to the invention the edit distance D(s, c~ between word s and a word q is defined recursively and the edit distance D(s, q) calculated by means of a dynamic programming procedure, and the edit distance DN,S(S,P) between sequences S
and a sequence P is correspondingly recursively defined and the edit distance D,~,s(S,P) calculated by means of a dynamic programming procedure.
According to the invention the above-mentioned obj ects and advantages are also realized with the use of the search system according to the invention in an approximate search engine The search system and the method according to the invention shall now be discussed in greater detail in the following with reference to the accompanying drawing figures, of which fig. 1 shows an example of a suffix tree, fig. 2 examples of word-spaced sparse suffix trees. as used with the present invention, fig. 3 an example of a so-called PATRICIA trie as known in prior art, fig. 4 a further example of a word-spaced sparse suffix tree as used with the present invention, fig. 5 an example of explicitly stored word sequence information as used with the present invention, fig. 6 a leaf node structure as used with the present invention, and fig. 7 schematically the structure of a search engine with the search system according to the present invention.
The search system according to the invention consists essentially of three parts, namely the data structure, the metrics for approximate matching and ~t~Eh~f3~D ShiE~
, CA 02337079 2001-Ol-06 ~ ~ _ , ! ( z, t r ( r C t' f r r ! r ' ~,' ' ~rc rt r't .rf~.t t r s « ~ t t r E s.
F f C C f J' ( f F E i C
E t E f' f C t C v F f' t 1 E f f f C t rt C C E C : f F f C i the search algorithm. When full text retrieval is the target, as essentially will be the case with the search system according to the present invention, then the entire data set which shall be retrievable, will be stored in a data structure which supports a high query performance.
The basic concepts underlying the present invention shall first be discussed in some detail. Stored information in the forn of text T is divided into words s and word sequences S. Words are substrings of the entire text separated by word boundary terms. The set of word boundary terms is denoted BTWOra. A
common set of word boundary terms could be the set {'; ': ; 'It; 'I~ ; '10 ;
'. ;
;' ; '?') where \t denotes a tab character, \n denotes a linefeed character and \0 denotes an end-.of document indicator. In connection with the following description of the present invention it will be useful with some definitions concerning strings and sequences.
Definition l: String A string is a sequence of symbols taken from an alphabet, such as the ASCII
characters. Then the length of a string is the number of instances of symbols or characters comprising the string, and is denoted ~x~. If x has the length m the string may also be written as xjxZ...x~...x"t, where x; represents the ith symbol in the string. ' A substring of x is a string given by a contiguous group of symbols within x.
Thus, a substring may be obtained from x by deleting one or more characters from the beginning or the end of the string.
Definition 2: Substrin~, suffix and prefix A substring of x is a string x/ = x;xl+ j ...x~ for some 1< i < j < n. The string x; =xln = x~...x,~ is a suffix of string x and the string x~ =xl = x1 x2...xj is a prefix of string x.
Also the notion of a word sequence will be used.
Definition 3: Word sequence -A word sequence is a sequence of separated, consecutive words. A word sequence S' = sl,s~, ...,sn consists of h single words (or strings) s~,s2, up to sn.
Word sequences are delimited by sequence boundary terms. The set sequence boundary terms are denoted BTseq. A common set of sequence boundary terms could be the set {'0\'~, where \0 indicates an end-of document marker.
~a,~~3 S'~~'~
°
~ CA 02337079 2001-Ol-06 i i r F t f r i t c t f f t i f I f C ~ f t d ~ t t i . I t t ~ t i t f ~ d t 4 ' t t l C d F' i a t d s t C E ( f f t f L C t C' C
i t 1 C F C t i i f' f ! i [ f ( i 'f The concept approximate word matching can be described as follows.
Given a string s = s js2... s,~ and a query term q = qlq2... q"~. Then the task is to . find all occurrences of q in s that is at most k errors away from the original query term q, A proximity metric determines how to calculate the errors between q and a potential match s1... s~.
A common metric for approximate word matching is the Levenstein distance or edit distance (V.I. Levenstein, "Binary codes capable of correcting deletions, insertions, and reversals", (Russian) Doklady Akademii nauk SSSR, VoI. 163, No. 4, pp. 845-8 (1965); also Cybernetics and Control Theory, Vol. 10, No. 8, pp. 707-10, (1966)). This metric is defined as the minimum number of edit operations needed to transform one string into another. An edit operation is given by any rewrite rule, for instance:
~ (a~8), deletion ~ (spa), insertion ~ (a~b), change Let p and nz be two words of size i and j,~respectively. Then D(i,j) denotes the edit distance between the ith prefix of p and the jth prefix of m. The edit distance can then recursively be defined as:
D(i,0) = D(0, i) = i D(i-l, j)+1 Dpi, j) = min D(i, j -1~+1 (1) D(i-l, j -1)+~7~i, j) where a~i, j ) = 0 if p~ = n~ j else 1 It is also possible to define an approximate matching on the level of words in a word sequence and this can be described as follows.
Given a text T consisting of the ~ words w~,wa...w" where each of the words is a string of characters. A sequence pattern P consists of the ~r~ words p1, p~, ..., p",. The sequence pattern P is said to have an approximate occurrence in T if the sequence p~, p2, ..., pm differs with at most k errors from a sequence w~, Wl+la ..., YV j for some ij, such that 1 < i <_ j <_ ~z.
Again, a proximity metric determines how to calculate the number of errors between the. two sequences.
~~~tVD~'D SHEET
' CA 02337079 2001-Ol-06 . ' t r i r r c a i , r r r c ~ f < r r c f i ~r r f i ~ f t t r r f r a r c r s r r f r f S i r F i t f F F
S ~ t f F r r f' ~ t F f ( t f r. i r A text that shall be retrieved in a search system must be indexed in a manner which facilitates searching the data. Consequently the data structure is a .
kernel data structure of the search system according to the present invention and is based on so-called suffix trees and particularly a sparse suffix tree.
These two kinds of structures shall be defined in the following. A suffix tree S(T) is a tree representation of all possible suffixes in the text T. All unary nodes in a suffix tree S(T) are concatenated with its child to create a compact vari ant.
Fig. 1 shows the suffix tree for the text T= "structure".
Even more particularly the present invention is based on sparse suffix trees.
These were introduced by J. Karkka.inen ~ E. LTkkonen, in "Sparse Suffix Trees", Proceedings of the Second Annual International Computing and Combinatorics Conference (COCOON '96), Springer Verlag, pp.219-230, which again was based on ideas published by D.R. Morrison, "PATRICIA -Practical Algorithm To Retrieve Information Coded in Alphanumeric", Journal of the AC1VI, 15, pp. 514-534 (1968). A sparse suffix tree is defined as follows.
Definition 4: Sparse suffix tree A sparse suffix tree SST(T) of the text T is a suffix tree, containing only a subset of the suffixes present in the suffix tree ST(T) of the text.
When using the search system according to the present invention searching for entire words, advantageously a non-evenly spaced sparse suffzx tree may be created by storing suffixes starting at word boundaries only. The concept words-spaced sparse suffix tree is defined as follows.
Definition 5: Word-spaced sparse suffix tree A word-spaced sparse suffix tree SSTws(T) of a text T is a sparse suffix tree SST(T) containing only the suffixes starting at a word separator character in the text.
Fig. ~ S110WS two .examples of word-spaced sparse suffix trees. Parts of the suffixes have been omitted to enhance the readability. The word-spaced sparse suffix tree for T = "to be the best" is the left structure, and T = "to malce the only major modification" is the right structure in fig. 2.
~A~ED
' CA 02337079 2001-Ol-06 I , ~ 4 I ~ t t r r f f f f t f t ( t c f r ' t t f f t t f 4 ( t ( 7~1~i4 ! P ~f 4 t E f: ( l C ( 4 f ; t f.
f f t ( t t l' f t C ( 4 7 ( ( f' ( f f t t' ( t t f l ( t f t In the search system of the present invention the text is naturally divided into words which are stored independently in the word-spaced sparse suffix tree.
As the atomic search term for searching is the word itself, advantageously each suffix will be terminated at the end of the word. This reduces the sparse suffix tree to a so-called PATRICIA trie (Morrison, op.cit.). A trie as defined in the literature is a rooted tree with the properties that each node, except the root, contains a symbol of the alphabet and that no two children of the same node contain the same symbol. It should be noted that the word trie derives from the word "retrieval" and hence indicates that the trie is a tree structure suitable for retrieval of data. A PATRICIA trie is defined as a keyword-spaced sparse suffix tree (KWS tree) where the suffixes stored in the leaf nodes are limited by keyword delimiters. An example of a PATRICIA trie for the set of keywords ~"avoid", "abuse", "be", "become", "breathe", "say"~ is shown in fig. 3. The structure used in the search system of the present invention differs from the PATRICIA trie because the search system explicitly stores sequence information of the words. Reducing the suffix length requires that the representation of the leaf node is changed.
Pointers to the original text are replaced by the suffix string itself. A
suffix length reduction of this kind is shown in .fig. 4 for the same two strings as shown in fig. 2. In other words fig. 4 shows word-spaced sparse suffix trees with suffixes cut off at word boundaries. The word-spaced sparse suffix tree for T= "to be the best" is shown at left and the word-spaced sparse suffix tree for T = "to make the only maj or modification" is shown at right in the figure. A leaf node will contain a list of all positions where the word represented by the leaf node occurs.
Instead of using the implicit sequence of information found in the original text, the present invention explicitly stores sequence information in the word-spaced sparse suffix tree. This is done, by using pointers between the leaf nodes that represent consecutive words in the original text. As at least all the occurrences of the word represented by a particular leaf node are available, a pointer must be added to the next consecutive leaf.
A leaf node contains only the suffix of the word it represents, so when traversing the sequence pointers in the occurrence list only the suffixes of each of the consecutive words are revealed. This is handled by storing the entire word in the leaf node instead of just the suffix and thus also data structure of the invention differs from the PATRICIA trie in this respect. The a~f~.G'~ u~~~~
CA 02337079 2001-Ol-06 r t a r i r r t L t T f c r V i t r ( i i r , , , r r r!2, r , r , r ~ r ' t r r r i a a r i s s r c r a i c r r . r s. t i r c s r r r r a c ~ t r r r r data structure for explicitly stored word sequence information with an occurrence list with pointers to the next consecutive word and to its occurrence is shown in fig. 5.
The search system according to the present invention uses a PATRICIA trie for organizing the occurrence list (Morrison, op.cit.). The PATRICIA trie enables the search system to access the list of all consecutive words matching the string p2 in a time O(~p2~), where ~p2~ of course is the length of p2. By using a PATRICIA trie to organize the list of occurrences, a completely defined tree structure is obtained for storing words from a text and maintaining the sequence information. A typical leaf node, with both a PATRICIA trie for the organized occurrence list and the extra unsorted list of occurrences, is shown in fig. 6. As an example the memory requirement for an occurrence list as used in the search system of the present invention, a database with about 742358 documents has a total of 333 856 744 words and a lexicon of 53 8 244 distinct words. The total size of the database is 2054.52 MB. The average word length is thus 6.45 bytes. A sparse suffix tree will use 8 bytes for each internal node, using 32 bit pointers. It is assumed that an average of 3 internal nodes is used for each word. The leaf node would then require 6.45 bytes for storing the entire word plus 32 bits for a pointer to an occurrence list. A total of 34.45 bytes/word gives a total size of 18.108 MB.
In addition the occurrence list has the size of 4 bytes per entry and 12 bytes if the full version is to be used. Hence the total memory requirement of the occurrence list varies from 1273 MB to 3820 MB. The data structure using a sparse suffix tree will have a size between 60% to 200% of the original text.
This is comparable with the requirements of an inverted file, but the sparse suffix tree as used in the search system according to the invention provides much faster searching, enables approximate matching and makes sequence matching easy to perform.
In approximate searching, a metric is used to give an error measure of a possible match. The search system according to the present invention employs several metrics, and particularly a unique combination of metrics.
These metrics along with the combined metric shall be discussed in the following.
An edit distance metric as defined above allows the operations deletion, insertion and change which intuitively apply to words as well as characters.
Afi~i~D SHE~~
' ,,~~,, CA 02337079 2001-Ol-06 , . , .. f f : f V f t : t t I t ( 1 f -a t~f. t, f t t f t' , t d y t . d :ljt t t t d f t t ~ Y Y S f t f ( t a t t ~ f f t S f ( f F " 9 t t f f f f t f. t ~ f . f C t f f : f t t Common errors in matching phrases are missing, extra or changed words.
Hence the edit distance metric as previously defined shall be adapted and extended in order to apply to the approximate word sequence snatching problem. Edit operations for sequences are defined below.
Definition 6: Edit operations for sequences For transforming one sequence S of words into another sequence P of words, the edit operations allowed on the word in the sequences may be written according to the following rewrite rules: ' ~ (ass), deletion of word a from the sequence ~ (spa), insertion of word a into the sequence ~ (a~b), change of word a into word b ~ (ab-~ba), transposition of adjacent words a and b.
Instead of characters as atoms, the search system according to the invention applies the edit operations to words which then should be regarded as the operational atoms. .
A cost function cealr(x~Y) is a constant which is defined as 1 delete _ 1 insert edit ~'~' ~ .~~ 2 1 transpose a(x, y) change where a(x,y) is defined as ~y~= 0 x=y ( a x 1 else By using the edit operations as defined above the edit distance for sequences can now be defined.
Definition 7: Edit distance for sequences The edit distance metric for sequences defines the distance DSe9(S,P) between the sequence S = sl,s,~, ...,sn and the sequence P = pl,p~, ...,p,n as the minimum sum of cost c(x-~y) for the sequence of edit operations transforming the sequence S into the sequence P.
The search system according to the present invention enhances the edit distances metric for sequences to weight the cost of the edit operations by the ~~a~tD 5~1~
CA 02337079 2001-Ol-06 -t t , t t ; t. t t r t t d -n , ~ ~ ~ ~ a ~ a t Y t f ~ t i f ! f ~
F t t f . t C~~~'t i t C C - t t P ( t t t f t t d t t f d t F f i t ~ t t i i d F t F 1 t i t f i t t f' t E d 2 t d size of the words operated. upon.
Definition 8: Word size-dependent edit distance for sequences _ The word size dependent edit distance for sequences is defined as the minimum sum of costs for the editing operations needed to transform one sequence into the other. The cost functions axe dependent on the word size of its operands.
In the search system according to the invention a definition of cost functions is given by the equations Chtsert (E ~ a) _ l al .
Z
Cdelete (a ~ ~) _ l al l (4) ~trpnspose (ab -~ ba) =1 max(Ila~-Ibll,1) Cchnnge (a ~ b~ = l where Z denotes the average length of a word in the two sequences being compared. The cost of each edit operation is weighted by a size proportional to the change in the total length of the sequence or by the ratio of the current word length and the average word length in the sequences considered.
Now the distance metric reflects the assumption of some relation between the word length and how important the word is to the semantic context of the word sequence. Furthermore the search system according to the invention employs proximity at the character level when the change edit operation (a~b) is used. Replacing a word a by another word b should be related to the similarity between these two words. The new cost function fox the change edit operation hence is given as:
max a-b ,l Cchange (a ~ b~ - C'3aPProx (a~ b~ ~~ I I ~~ (5) l where a~PPrax (a~ bJ - 1.7(a, b) Where D(a, b) is the normalized edit distance measuring function for words, 0 means full similarity, 1 means no similarity.
~s~~
CA 02337079 2001-Ol-06 r r ( i , t : c W r ' . ~ a r t r ( C ( 'Y , a t r r t ( ( t r ( L'S ( a ( t ( ( ( t ( f t 1 t ( < F 't t ( ' ( t 1 t t t F . t ( i ( F ( i ( t i f' i i i t E f E t ( E
The search system according to the invention combines the edit distance metric for sequences with the cost functions as given by formulas (4), (5) and (6), with an edit distance metric for words as given by formula (1). This means that sequence edit operations are only used when the words being matched are more than k errors away from each other.
The algorithms used in the search system according to the invention perform efficient searching of the described structures. Matches are found according to the metrics as given above.
Approximate word matching in a word-spaced sparse suffix tree is done by combining the calculation of the edit distance matrix and a traversal of the suffix tree. An algorithm for this is written in pseudo-code and given in table I.
This algorithm is adapted from a trie-matching algorithm as proposed by H.
Shang 8~ T.H. Merrettal, "Tries for Approximate String Matching", IEEE
Transactions on Knowledge and Data Engineering, Vol.S, No. 4, pp. 540-547 (1996). The expected worst case running time of the algorithm is O(k~E~k) according to Shang ~ Merrettal (op.cit.).
Approximate word sequence matching requires the calculation of the word sequence edit distance for all possible matches. However, the number of possible matches can be limited by starting the calculation of the edit distance only on the possible words. The cost of deleting a word from the sequences determines the number of possible start words. If the accumulated cost of deleting the i first words in a query sequence PQ rises above a given error threshold, the candidate sequence starting with the ith word of the query cannot possibly be a match. Therefore for a query sequence PQ of i words, at most i possible start words will be tried. Since there are no baclcpointers in the sequence structure of the tree, it will not be ensured that all possible matches are obtained. Adding backpointers would solve this problem. The algorithm for approximate word-sequence matching as used in the search system according to the present invention, is given in pseudo-code in table II below. This algorithm tries to match the first keyword with pl,p2...
sequentially, testing all possible start positions.
Aid .~wEf~
' CA 02337079 2001-Ol-06 , v ~ t f f ~ t a f 1 t t F t ~ i f t S ~ f t ,. .. _ ~ .. . i ~ t i t t L6ttt t f t Y f F t , t S t i t S ( t E t ' t t t t S t t t s Y 3'ttF .r . if t: Y Lt St In the ApproxSequenceMatch algorithm in table II the App~oxMateIzRest function is defined by the algorithm in table III below. This function matches the remaining sequence, using an initial error value.
> ' CA 02337079 2001-Ol-06 t t t f i t t t t r t r t ' ~ ' ~ ' ' ~ c f . i F ' F f a t ( f t J 1 ~ ~~ ( t ~ i a i S 7 i f C t f i I t y . r t C i F f i f F f S F
° F F A i ( S F t i f t i t t t t Table I
FindApproximate (root,p,k) node E- root;
i E- 1;
nodes ~- Children (node) ; // A stack of nodes for all v E nodes do if IsZeaf (v) then for j E- i to length (Suffix (node) ) do w~ F- Suffix (node) ~_i;
if w~ - '$' then // '$' is a stopchar output wl.,_~ ;
return;
if EditDist (i) - co then break;
else //Inte rnal node .7. ~ 1-~1;
wi ~- label ( v) if EditDist (i) - oo then , break;
nodes E- Children ( v) U nodes;
// end for EditDistance(j) // Calculates jth row for i~ 1 to length ( P) do if pi = w~ then 8 ~- 0 else 8 ~ 1:
Cl = D~i-lr j~ ,+ Cins (mj) ;
C2. = D~i,j-1~ + Cdel (pi) i Cg = D~1-l, j-1~ -f- Cchange (pir mj) i Cq = D~'1-~, j-2~ -1- Ctranspose(pirmj-1) i D~1, j~ ~ Cfraction (~ ~-Z ) ' I~1.1.11 ( C1, C~, C3, Cq ) ;
a.f D[i, j~ > k return oo; // No distance below k return D(i, j~
CA 02337079 2001-Ol-06 t t t a t f t: f , . s E r f t 1 f t f [ f < ~ f f . ' E f ( t~ ~.' t t t f~ C f f t f E C . S t t F i t E C t f E C t f t ' ( t f 5 f t t 6 . v l l i E Y F C E f v Table II
- ApproxSequenceMatch ED (root, P( pl,,pa.~--.Pmt ~k) m~.~p) matches startError ~ 0 startTndex E- 1 while startError _< k ~R startlndex S m do StartNOde E- .FlndExact (,pStartlndex~ ~
list ~ UnorderedOccurrence.List (startNodeJ ;
for all v~E lists do if ApproxMatchRest2 (v,P,k,startError) then matches ~ U v;
startError ~- startError + Cdel (,jJstartlndex~
startlndex E- startlndex + 1;
t~.~ y ~ ~_;~s a =3 ~
~ ~ CA 02337079 2001-Ol-06 t F i t t F r F r r c c <
F, t, t t ~ i t ~ f a c F L f r t il~ s~ f < . t F r c f t t t f t a r t ' f f ~ F t t f ( t ( F f T
t c t ! t ( f i i f F 3 f 7 ! ! t Table III
ApproxMa tch.Rest ( a o P, K, s tartError) error E- startError;
lastError ~ startError;
column ~ O ;
node ~ u;
for a ~ p2 to p~p~ do node ~ NextOccurrence(node);
word ~ Keyword(node);
IastError ~- error;
error ~- startError + EditDistance (column);
if error > k AND IastError > k then return false;
return true;
Edit.Distance(j) // Calculates jth row for iE- 1 to length ( P) do if pi = w~ then a ~ 0 else r7 E- 1;
Cl = D[1-l, ]' + Cins (mj) i C~ = D[1, j-1~ + Cdel (pi) i Cg =~ D[2-l, j-1~ + C~hange(~pirmj) r Cq = D[1-~, j-2~ + Ctranspose(pirmj-1) i D[.1, j~ ~ Cfraction (,~l-Z ) ' Illlll ( C1, C~, C3, Cq ) ;
if D[i, j] >- k return oo; // No distance below k return D[i, j~
~;~~~r~Q s~~
CA 02337079 2001-Ol-06 a ~ . _ t f r a t c ; s r a t i t , ' i~i t f t f ' a t c t ~ t z t : f t ' F S f f a ~ t f F t t' 1 f ! S
S t ' d I f ~ t f t t : < Z F 1 t f t C f t f ~ f' O t a S r F f t t The algorithms in tables II and III are written in the same pseudo-code as the algorithm in table I.
The Fi~cclExact function used to find the leaf node matching the first word in the sequence performs a simple traversal of the tree and its running time is O~pj~ where pl denotes the first word in a query sequence PQ. Calculating the edit distance can be done in ~P~2 time using straightforward dynamic programming or in O(k) time (where k denotes the error threshold) using improved versions of the calculation algorithm, see E. Ukkonen, "Finding Approximate Patterns in Strings", Journal of Algorithms, vol. 6, pp. 132-137 IO (1985).
If ~~a~~(p;) denotes the total sum of the number of occurrences of each word pj in the word sequence, then the worst case running time is O(k~~co~C(pz)).
Finally the implementation of a search engine based on the search system according to the invention shall briefly be discussed. Particularly a search engine based on the search system according to the invention is implemented as an approximate search engine (ASE) and is intended as a search engine for indexing large document collections and providing algorithms for exact and approximate searching of these document collections. ASE shall provide a data structure for storing large texts or collection of documents. It is to be understood that the data structure may be generated from documents which contain additional information, such as images, video, sound, and the text may be formatted and/or annotated. The data structure is identical to the word-spaced sparse suffix tree as discussed above and it is, of course, to be understood that the words is the keywords of the search system, hence the word-spaced sparse suffix tree may instead be termed a keyword-spaced sparse suffix tree (KWS tree). The ASE shall contain algorithms for indexing documents in the KWS tree. These algorithms, of course, do not form a part of the search system according to the present invention, but they are well-known to persons skilled in the art and described in the literature, see for instance J. Karkkainen & E. ZJkkonen (op.cit.) and D.R. Morrison (op. cit) .
The search system according to the invention and as used in the ASE
employs algorithms both for exact and approximate matching of a pattern in a KWS tree: The algorithms given above in table I and table II are used for approximate word and word sequence snatching with the non-uniform edit ~~~D 5:~~
CA 02337079 2001-Ol-06 , . ! I . .
a i S n . . a t . , t .. . ' ~i . < , i ~~lCti t [
t i . i F [ [ t f t k I d ~ . ( [ d < ! f t ; .
t f i F t f S ( t F ; ~ d' t i S t distance as a metric. Finding an exact match of keyword p with length ~a in a .
KWS tree is known in the art and easily implemented as a simple traversal of the tree structure. An appropriate algorithm for exact keyword matching written in pseudo-code is given in table IV. The search system according to the invention also shall be able to support algorithms for exact keyword sequence matching. Algorithms for exact keyword sequence matching are known in the art and easily implemented as e.g. shown in pseudo-code in table V below. The algorithm given here will find the exact match of the first keyword, if any. Then it will for all occurrences of the first keyword check if the second keyword matches the second keyword of the query. If so, the MatchRest procedure in table V is used to determine if the occurrence of the two first keywords are matching in the entire sequence. For approximate keyword matching in a KWS tree the search system implements the algorithm in table I above. For approximate lceyword sequence snatching the search system implements the algorithm in table II above, matching a first keyword sequentially with p,,p~... and testing all possible start positions, applying the ApproxMatchRest function as given in table III to match a sequence starting at a particular position and handle the initial error value.
Finally, the ASE shall need a simple front end which gives the user control of indexing and querying the document collection. The front end should also be able to furnish statistics of the document collection and provide both a networlc interface for remote access, e.g. via WWW, and a local server user interface.
The ASE with the search system according to the invention should be general in a manner that allows for the adding new indexing and searching algorithms easily. Also, storing extra information about each document or keyword shall be possible to implement in an easy manner. Particularly the front end should be independent of the data structure and the search algorithms, such that internal changes in these has no effect on the design of the former.
The use of the search system according to the invention the ASE should be designed to have as low memory overhead as possible in the data structure.
Also,,searching should be designed to be as fast as possible. However, there will usually be a trade-off between these two factors.
~~~~3 ~i~~~
i CA 02337079 2001-Ol-06 f 1 t 3 1 .( i v F r ! s t ., . . ' , . t . ' ~ ; t' ;.~
t ~ , 9 i r Jy Y t , R t t t G t 5 c i t t t a t t , C r t t E f t t ' f f f t r F ! F t f' f a C r t t f .
Table IV
FindExact (root,p) _ i~ 1;
node ~- Find Child (root,pi) ;
while node AND i _< length (p) do if IsLeaf (node) AND Suffix (node) - Pi...Pm then return node;
i ~ i+1 node ~- FindChild (node, pi);
return NIL;
Table V
MatchSequenceExact (P., root) ma tches ~ Q~;
v ~ FindExact (pl,root) ;
if ~ P ~ > 1 then if v ~ NIL then list ~ UnorderedOccurrenceList (v);
for all a E List do if NextKeyword(u) - p2 then if MatchRest (p3...pm, u) then matches ~ matches U Occurrence (u) ;
return matches;
MatchRest (P, u) node ~
u;
for v ~ p1 to p~P~ do ~
node ~ NextOccurrence(node);
word ~ Keyword(node);
if v ~ word then return false;
a~~~~~~.~ yAt~~~, °
, CA 02337079 2001-Ol-06 i r c r t ; r f r t r r r r s' t f r a r i a r r r t t t t - r t23 , trt i~ t i ( i: f C° r f f t t C a f f f C ( ( t f v f ( f ( f t i f f t f f f t f C . t f C C t' r f To sum up, an ASE with a search system according to the invention shall comprise four major modules.
1. Document indexing module DIM for indexing documents in the KWS tree structure. This module should also contain all extensions to support several document types.
t f ( P ; 1 f t ( . C ( t ( ( t t ( ( t t t C f f f '( CE ft 1f' ~i~i f(' f' ft' E'f A search system and method for retrieval of data, and the use thereof in a search engine The present invention concerns a search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words and/or symbols and sequences thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R, comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric Mwhich measures the degree of matching between the query Q and retrieved information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords kw; a method in a search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words .and symbols and sequences thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric Mwhich measures the degree of matching between the query Q and retrieved. information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords krv, wherein the information in the text is divided into words and word sequences, the words being substrings of the entire text separated by word boundary terms and forming a sequence of symbols, and wherein each word is structured as a sequence of symbols.
The invention also concern the use of the search system.
A tremendous amount of information in various fields of human knowledge is collected and stored in computer memory systems. As the coiilputer memory systems increasingly are linked in public available data communication networks, there has been an increasing effort to develop systems and methods for searching and retrieving information for public or personal use. Present search methods for data have, however, limitations that t~t~Nt3ED SHE
CA 02337079 2001-Ol-06 t ~ t f t t t f' t t t t t t I
a r r t l . ~' to E f i f t t fJr l E E t i ~ t g i , ~r f t t t r f t t l f f' t ' f t t f C % f t F C l i t f E ~ f f' E C f' t t l f seriously reduce the possibility of efficiently retrieving and using information stored in this manner.
Information may be stored in the form of different data types, and in the context of information search and retrieval it will be useful to discern 5 between dynamic data and static data. Dynamic data is data that change often and continuously, so that the set of valid data varies all the time, while static data only changes very seldom or never at all. For instance will economic data, such as stock values, or meteorological data be subject to very quick changes and hence dynamic. On the other hand archival storage of books and documents are usually permanent and static data. The concept the volatility of the data relates to how long the information is valid. The volatility of data has some bearing upon how the information should be searched and retrieved. Large volumes of data require some structure in order to facilitate searching, but the time cost of building such structures must not be higher than the time the data is valid. The cost of building a structure is dependent on the data volume and hence the building of data structures for searching the information should take both the data volume and the volatility into consideration. The information collected are stored in databases and these may be structured or unstructured. Moreover, the databases may contain several types of documents, including compound documents which contain images, video, sound and formatted or annotated text. Particularly strucfiured databases are usually furnished with indexes in order to facilitate searching and retrieving the data. The growth of the World Wide Web (WWW) offers a steadily growing collection of compound and hyperlinked documents. A great many of these are not collected in structured databases and no indexes facilitating rapid searching are available. However, the need for searching documents in the World Wide Web is obvious and as a result a number of so-called~search engines has been developed, enabling searching at Ieast parts of the information in the World Wide Web.
With a search engine it is commonly understood one or more tools for searching and retrieving information. In addition to the search system proper, a search engine also contains an index, for instance comprising test from a large number of uniform resource locators (URLs ). Examples of such search engines are Alta Vista, HotBot with Inktomy technology, Infoseek, Excite and Yahoo. All these offer facilities for performing search and retrieval of information in the World Wide Web. However, their speed and efficiency do ~
CA 02337079 2001-Ol-06 t . , r t f . t 1 S t t f t ( f ~ t t f t r ~ E r r z r , . r r t 1 t i f f t ~ f E I t t C 1 i f ' 1 ! Y C ! t E . i .f r i t f & F .f E Y t C t E F f L f t t f r , ' by no means match the huge amount of information available on the World Wide Web and hence the search and retrieval efficiency of these search engines leaves rnuch.to be desired.
Searching a large collection of text documents can usually be done with several query types. The most common query type is matching and variants of ' this. By specifying a keyword or set of keywords that has to be present in the queried information the search system retrieves all documents that fulfils this requirement. The basic search method is based on so-called single keyword matching. The keywordp is searched for and all documents containing this word shall be retrieved. It is also possible to search for a keyword prefix p~
and all documents where this prefix is present in any keyword in the documents, will be retrieved. Instead of searching with keywords, the search is sometimes based on so-called exact phrase matching, where the search uses several single keywords in particular sequence. As well-known by persons skilled in the art, the exact matching of keyword phrases in many search systems may be done with the use of Boolean operators, for instance based on operators such as AND, OR, and NOT which allow a filtering of the information; e.g. using an AND phrase results in that all documents containing the two leeywords linked by the AND operator will be returned.
Also a NEAR operator has been used for returning just the documents with the keywords matching and located "near" to each other in the document text.
In many structured database the documents contained in the database have been annotated, e.g. provided with fields vcThich denote certain parts or types of information in the document. This allows the search for matches in only parts of the documents and is useful when the type of queried information is known in advance.
When searching in text documents the data are structured and most likely present in some natural language, like English, Norwegian etc. When searching for documents with a certain context it is possible to apply proximity metrics for matching keywords or phrases that match the query approximately. Allowing errors in keywords and phrases are common method for proximity, using a thesaurus is another common method. A proximity search requires only that there shall be a partial match between the information retrieved and the query. International published application W096/00945 titled "Variable length data sequence matching method and apparatus" (Doringer & al.) which has been assigned to International AH~~'~~ S~ C
CA 02337079 2001-Ol-06 _ t F t S f i _ h t . t r r f r r t . t r a4 t t t ~ r t f f r t t c c t f a r a d t r F i t ' E f 4 t f F ( f t C ( F I C ~ r i t t r f f f f C' ~ f t f f t f E t c Business Machines, Corp., discloses the building, maintenance and use of a database with a trie-like structure for storing entries and retrieving at least a partial.match, preferably the longest partial match or all partial matches of a search argument (input key) from the entries.
In order to further illuminate the general prior art mention can be made of international published patent application W092/15954 (KimbaIL & al., assigned to Red Brick System, U.S.A.) and US patent no. 5 627 74~ (Baker & al., assigned to Lucent Technologies, Inc., U.S.A.),'both disclosing data structures in the form of suffix trees for searching/matching in a square matrix. Neither of these two publications disclose anything beyond a regular suffix tree, except for the use of a linked list during matching and do not teach or suggest approaches to limit the search space when searching for approximate matches. However, such approaches would be most desirable when applying data structures based on suffix trees to searching, particularly for approximate matches in extremely large document collections, such as may be found on the World Wide Web.
The main object of the present invention is thus to provide a search system and a method for fast and efficient search and retrieval of information in . large volumes of data. Particularly it is an object of the present invention to provide a search system suited for implementing search engines for searching of information systems with distributed large volume data storage, for instance Internet. It is to be understood that the search system according to the invention by no means shall be limited to searching and retrieving information stored in the form of alphanumeric symbols, but equally well may be applied to searching and retrieving information stored in the form of digitalized images and graphic symbols, as the word text used herein also may interpreted as images when these are represented wholly or partly as sets of symbols. It is also to be understood that the search system according to the invention can be implemented as software written in a suitable high-level language on commercially available computer systems, but it may also be implemented in the form of a dedicated processor device for searching and retrieving information of the aforementioned lcind.
The above-mentioned objects and advantages are realized according to the invention with a search system which is characterized in that the data structure comprises a tree structure in the form of a non-evenly spaced sparse ~t~~N~D S~H~~~
CA 02337079 2001-Ol-06 , . r . " ' ~ c ~ t c i i ~ r t r , C t t s s r t , t t i f f r t ; t a t f ~ f t i t i f i f a ; i r' f F i f ( f ( t t C t ( t f i f f < . i' t ~ ~C ; f ; f t i Y C < i t suffix tree ST(T) for storing suffixes of words and/or symbols, and sequences thereof in the text T, that the metric M comprises a combination of an edit distance metric for an approximate degree of matching between words and/or symbols in the text T and a query Q and an edit distance metric for an approximate degree of matching between sequences S of words and/or symbols in the text T and a query sequence l' of words and/or symbols in the query Q, the latter edit distance metric including weighting cost functions for edit operations which transform a sequence S' of words and/or symbols in the text T into the sequence P of words and/or symbols in the query Q, the weighting taking place with a value proportional to a change in the length of sequence upon a transformation or dependent on the size of the words and/or symbols in sequences to be matched, that the implemented search algorithms comprise a first algorithm for determining the degree of matching between words and/or symbols in the suffix tree representation of respectively the text T and a query Q, and a second algorithm for determining the degree of matching between sequences of words and/or symbols iri the suffix tree representation of respectively the text T and the query Q, said first and/or second algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words or sequences of symbols or combinations thereof, such that information R is retrieved on the b asis of query Q with a specified degree of matching between the former and the latter, and that the search algorithms optionally also comprise a third algorithm for determining exact matching between words and/or symbols in the suffix tree representation of respectively the text T and the query Q
and/or a fourth algorithm for determining exact matching between sequences of words and/or symbols in the suffix tree representation of respectively the text T and the query Q, said third and/or fourth algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words, or sequences of symbols or combinations thereof, such that information R is retrieved on the basis of the query Q with an exact matching between the former and the latter.
In an advantageous embodiment of the search system according to the invention the suffix tree ST(T) is a word-spaced sparse suffix tree SSTWS(T), comprising only a subset of the suffixes in the text T.
Preferably is then the word space sparse suffix tree SST~,S(T) a keyword-spaced sparse suffix tree SST~,S(T).
AA~~fD Si-~E~
°
CA 02337079 2001-Ol-06 a t i q f - f t f f f ( a f f i t . (. t f n 1 ' G fj~ t F f f ~ t i f fl f f f f f t f t t i t F ( t t 6 : f t i F f f i t f t f c f F ~ i f 6 f f t E i F F Q t ~: f !~
In further advantageous embodiments of the search system according to the invention the first-algorithm for detecting the degree of keyword matching in a keyword-spaced sparse suffix tree SSTkWS(T) is implemented as disclosed by dependent claim 4, the second algorithm for determining the degree of sequence matching in a keyword-spaced sparse suffix tree SSTkWS(T) implemented as disclosed by dependent claim 5, whereby a subroutine of the second algorithm preferably is implemented as disclosed by dependent claim 6, the third algorithm for determining an exact keyword matching in a keyword-spaced sparse suffix tree SSTkWS(T) implemented as disclosed by dependent claim 7, and finally the fourth algorithm for determining an exact keyword sequence matching in a keyword-spaced sparse suffix tree SSTkWs implemented as disclosed by dependent claim 8.
The above-mentioned objects and advantages are also realized according to the invention with a method which is characterized by generating the data structure as a word-spaced sparse suffix tree SSTWS(T) of a text T for representing all the suffixes starting at a word separator symbol in the text T, storing sequence information of the words in the text T in the word-spaced sparse suffix tree SSTWS(T), generating a combined edit distance metric M
comprising an edit distance metric D(s, q) for words s in the text T and a query word q in a query Q and a word-size dependent edit distance metric DWS(S,P) for sequences S of words in the text T and a sequence P of words q in the query Q, the edit distance metric DWS(S,P) being the minimum sum of costs for edit operations transforming the sequence S into the sequence P, the minimum sum of costs being the minimum sum of cost functions for each edit operation weighted by a parameter proportional to the change in the total length of the sequence S or by the ratio of the current word length and average word length in the sequences S;P, and determining the degree of matching between words s, q by calculating the edit distance. D(s, q) between the words s of the retrieved informatiomR and the word q of a query Q, or in case the words s,q are more than k errors from each other, determining the degree of matching between the word sequences SR, PO of retrieved information R and a query Q respectively by calculating the edit distance DWS(SR,Pg) for all matches.
Advantageously the method according to the invention additionally comprises weighting an edit operation which changes a word s into word q with a parameter for the proximity between the characters of the words s; q, r~~9~Nt~D SHE~~
' CA 02337079 2001-Ol-06 E i' i' ( f ( t t ' C ( t 1 t l f f ~'',~ ( C a E~ t t t ( /F i t t p t F I t t t i t a C C.
f C C C ( f Y f C ( t t (' C C P t ( i C f f f C C ~ C f C t thus taking the similarity of the words s; q in regard when determining the cost of the edit operation in question. .
In an advantageous embodiment of the method according to the invention the number of matches is limited by calculating the edit distance by calculating the edit distance DWS(SR,Po) for restricted number of words in the query word sequence Pg.
In another advantageous embodiment of the method according to the invention the edit distance D(s, c~ between word s and a word q is defined recursively and the edit distance D(s, q) calculated by means of a dynamic programming procedure, and the edit distance DN,S(S,P) between sequences S
and a sequence P is correspondingly recursively defined and the edit distance D,~,s(S,P) calculated by means of a dynamic programming procedure.
According to the invention the above-mentioned obj ects and advantages are also realized with the use of the search system according to the invention in an approximate search engine The search system and the method according to the invention shall now be discussed in greater detail in the following with reference to the accompanying drawing figures, of which fig. 1 shows an example of a suffix tree, fig. 2 examples of word-spaced sparse suffix trees. as used with the present invention, fig. 3 an example of a so-called PATRICIA trie as known in prior art, fig. 4 a further example of a word-spaced sparse suffix tree as used with the present invention, fig. 5 an example of explicitly stored word sequence information as used with the present invention, fig. 6 a leaf node structure as used with the present invention, and fig. 7 schematically the structure of a search engine with the search system according to the present invention.
The search system according to the invention consists essentially of three parts, namely the data structure, the metrics for approximate matching and ~t~Eh~f3~D ShiE~
, CA 02337079 2001-Ol-06 ~ ~ _ , ! ( z, t r ( r C t' f r r ! r ' ~,' ' ~rc rt r't .rf~.t t r s « ~ t t r E s.
F f C C f J' ( f F E i C
E t E f' f C t C v F f' t 1 E f f f C t rt C C E C : f F f C i the search algorithm. When full text retrieval is the target, as essentially will be the case with the search system according to the present invention, then the entire data set which shall be retrievable, will be stored in a data structure which supports a high query performance.
The basic concepts underlying the present invention shall first be discussed in some detail. Stored information in the forn of text T is divided into words s and word sequences S. Words are substrings of the entire text separated by word boundary terms. The set of word boundary terms is denoted BTWOra. A
common set of word boundary terms could be the set {'; ': ; 'It; 'I~ ; '10 ;
'. ;
;' ; '?') where \t denotes a tab character, \n denotes a linefeed character and \0 denotes an end-.of document indicator. In connection with the following description of the present invention it will be useful with some definitions concerning strings and sequences.
Definition l: String A string is a sequence of symbols taken from an alphabet, such as the ASCII
characters. Then the length of a string is the number of instances of symbols or characters comprising the string, and is denoted ~x~. If x has the length m the string may also be written as xjxZ...x~...x"t, where x; represents the ith symbol in the string. ' A substring of x is a string given by a contiguous group of symbols within x.
Thus, a substring may be obtained from x by deleting one or more characters from the beginning or the end of the string.
Definition 2: Substrin~, suffix and prefix A substring of x is a string x/ = x;xl+ j ...x~ for some 1< i < j < n. The string x; =xln = x~...x,~ is a suffix of string x and the string x~ =xl = x1 x2...xj is a prefix of string x.
Also the notion of a word sequence will be used.
Definition 3: Word sequence -A word sequence is a sequence of separated, consecutive words. A word sequence S' = sl,s~, ...,sn consists of h single words (or strings) s~,s2, up to sn.
Word sequences are delimited by sequence boundary terms. The set sequence boundary terms are denoted BTseq. A common set of sequence boundary terms could be the set {'0\'~, where \0 indicates an end-of document marker.
~a,~~3 S'~~'~
°
~ CA 02337079 2001-Ol-06 i i r F t f r i t c t f f t i f I f C ~ f t d ~ t t i . I t t ~ t i t f ~ d t 4 ' t t l C d F' i a t d s t C E ( f f t f L C t C' C
i t 1 C F C t i i f' f ! i [ f ( i 'f The concept approximate word matching can be described as follows.
Given a string s = s js2... s,~ and a query term q = qlq2... q"~. Then the task is to . find all occurrences of q in s that is at most k errors away from the original query term q, A proximity metric determines how to calculate the errors between q and a potential match s1... s~.
A common metric for approximate word matching is the Levenstein distance or edit distance (V.I. Levenstein, "Binary codes capable of correcting deletions, insertions, and reversals", (Russian) Doklady Akademii nauk SSSR, VoI. 163, No. 4, pp. 845-8 (1965); also Cybernetics and Control Theory, Vol. 10, No. 8, pp. 707-10, (1966)). This metric is defined as the minimum number of edit operations needed to transform one string into another. An edit operation is given by any rewrite rule, for instance:
~ (a~8), deletion ~ (spa), insertion ~ (a~b), change Let p and nz be two words of size i and j,~respectively. Then D(i,j) denotes the edit distance between the ith prefix of p and the jth prefix of m. The edit distance can then recursively be defined as:
D(i,0) = D(0, i) = i D(i-l, j)+1 Dpi, j) = min D(i, j -1~+1 (1) D(i-l, j -1)+~7~i, j) where a~i, j ) = 0 if p~ = n~ j else 1 It is also possible to define an approximate matching on the level of words in a word sequence and this can be described as follows.
Given a text T consisting of the ~ words w~,wa...w" where each of the words is a string of characters. A sequence pattern P consists of the ~r~ words p1, p~, ..., p",. The sequence pattern P is said to have an approximate occurrence in T if the sequence p~, p2, ..., pm differs with at most k errors from a sequence w~, Wl+la ..., YV j for some ij, such that 1 < i <_ j <_ ~z.
Again, a proximity metric determines how to calculate the number of errors between the. two sequences.
~~~tVD~'D SHEET
' CA 02337079 2001-Ol-06 . ' t r i r r c a i , r r r c ~ f < r r c f i ~r r f i ~ f t t r r f r a r c r s r r f r f S i r F i t f F F
S ~ t f F r r f' ~ t F f ( t f r. i r A text that shall be retrieved in a search system must be indexed in a manner which facilitates searching the data. Consequently the data structure is a .
kernel data structure of the search system according to the present invention and is based on so-called suffix trees and particularly a sparse suffix tree.
These two kinds of structures shall be defined in the following. A suffix tree S(T) is a tree representation of all possible suffixes in the text T. All unary nodes in a suffix tree S(T) are concatenated with its child to create a compact vari ant.
Fig. 1 shows the suffix tree for the text T= "structure".
Even more particularly the present invention is based on sparse suffix trees.
These were introduced by J. Karkka.inen ~ E. LTkkonen, in "Sparse Suffix Trees", Proceedings of the Second Annual International Computing and Combinatorics Conference (COCOON '96), Springer Verlag, pp.219-230, which again was based on ideas published by D.R. Morrison, "PATRICIA -Practical Algorithm To Retrieve Information Coded in Alphanumeric", Journal of the AC1VI, 15, pp. 514-534 (1968). A sparse suffix tree is defined as follows.
Definition 4: Sparse suffix tree A sparse suffix tree SST(T) of the text T is a suffix tree, containing only a subset of the suffixes present in the suffix tree ST(T) of the text.
When using the search system according to the present invention searching for entire words, advantageously a non-evenly spaced sparse suffzx tree may be created by storing suffixes starting at word boundaries only. The concept words-spaced sparse suffix tree is defined as follows.
Definition 5: Word-spaced sparse suffix tree A word-spaced sparse suffix tree SSTws(T) of a text T is a sparse suffix tree SST(T) containing only the suffixes starting at a word separator character in the text.
Fig. ~ S110WS two .examples of word-spaced sparse suffix trees. Parts of the suffixes have been omitted to enhance the readability. The word-spaced sparse suffix tree for T = "to be the best" is the left structure, and T = "to malce the only major modification" is the right structure in fig. 2.
~A~ED
' CA 02337079 2001-Ol-06 I , ~ 4 I ~ t t r r f f f f t f t ( t c f r ' t t f f t t f 4 ( t ( 7~1~i4 ! P ~f 4 t E f: ( l C ( 4 f ; t f.
f f t ( t t l' f t C ( 4 7 ( ( f' ( f f t t' ( t t f l ( t f t In the search system of the present invention the text is naturally divided into words which are stored independently in the word-spaced sparse suffix tree.
As the atomic search term for searching is the word itself, advantageously each suffix will be terminated at the end of the word. This reduces the sparse suffix tree to a so-called PATRICIA trie (Morrison, op.cit.). A trie as defined in the literature is a rooted tree with the properties that each node, except the root, contains a symbol of the alphabet and that no two children of the same node contain the same symbol. It should be noted that the word trie derives from the word "retrieval" and hence indicates that the trie is a tree structure suitable for retrieval of data. A PATRICIA trie is defined as a keyword-spaced sparse suffix tree (KWS tree) where the suffixes stored in the leaf nodes are limited by keyword delimiters. An example of a PATRICIA trie for the set of keywords ~"avoid", "abuse", "be", "become", "breathe", "say"~ is shown in fig. 3. The structure used in the search system of the present invention differs from the PATRICIA trie because the search system explicitly stores sequence information of the words. Reducing the suffix length requires that the representation of the leaf node is changed.
Pointers to the original text are replaced by the suffix string itself. A
suffix length reduction of this kind is shown in .fig. 4 for the same two strings as shown in fig. 2. In other words fig. 4 shows word-spaced sparse suffix trees with suffixes cut off at word boundaries. The word-spaced sparse suffix tree for T= "to be the best" is shown at left and the word-spaced sparse suffix tree for T = "to make the only maj or modification" is shown at right in the figure. A leaf node will contain a list of all positions where the word represented by the leaf node occurs.
Instead of using the implicit sequence of information found in the original text, the present invention explicitly stores sequence information in the word-spaced sparse suffix tree. This is done, by using pointers between the leaf nodes that represent consecutive words in the original text. As at least all the occurrences of the word represented by a particular leaf node are available, a pointer must be added to the next consecutive leaf.
A leaf node contains only the suffix of the word it represents, so when traversing the sequence pointers in the occurrence list only the suffixes of each of the consecutive words are revealed. This is handled by storing the entire word in the leaf node instead of just the suffix and thus also data structure of the invention differs from the PATRICIA trie in this respect. The a~f~.G'~ u~~~~
CA 02337079 2001-Ol-06 r t a r i r r t L t T f c r V i t r ( i i r , , , r r r!2, r , r , r ~ r ' t r r r i a a r i s s r c r a i c r r . r s. t i r c s r r r r a c ~ t r r r r data structure for explicitly stored word sequence information with an occurrence list with pointers to the next consecutive word and to its occurrence is shown in fig. 5.
The search system according to the present invention uses a PATRICIA trie for organizing the occurrence list (Morrison, op.cit.). The PATRICIA trie enables the search system to access the list of all consecutive words matching the string p2 in a time O(~p2~), where ~p2~ of course is the length of p2. By using a PATRICIA trie to organize the list of occurrences, a completely defined tree structure is obtained for storing words from a text and maintaining the sequence information. A typical leaf node, with both a PATRICIA trie for the organized occurrence list and the extra unsorted list of occurrences, is shown in fig. 6. As an example the memory requirement for an occurrence list as used in the search system of the present invention, a database with about 742358 documents has a total of 333 856 744 words and a lexicon of 53 8 244 distinct words. The total size of the database is 2054.52 MB. The average word length is thus 6.45 bytes. A sparse suffix tree will use 8 bytes for each internal node, using 32 bit pointers. It is assumed that an average of 3 internal nodes is used for each word. The leaf node would then require 6.45 bytes for storing the entire word plus 32 bits for a pointer to an occurrence list. A total of 34.45 bytes/word gives a total size of 18.108 MB.
In addition the occurrence list has the size of 4 bytes per entry and 12 bytes if the full version is to be used. Hence the total memory requirement of the occurrence list varies from 1273 MB to 3820 MB. The data structure using a sparse suffix tree will have a size between 60% to 200% of the original text.
This is comparable with the requirements of an inverted file, but the sparse suffix tree as used in the search system according to the invention provides much faster searching, enables approximate matching and makes sequence matching easy to perform.
In approximate searching, a metric is used to give an error measure of a possible match. The search system according to the present invention employs several metrics, and particularly a unique combination of metrics.
These metrics along with the combined metric shall be discussed in the following.
An edit distance metric as defined above allows the operations deletion, insertion and change which intuitively apply to words as well as characters.
Afi~i~D SHE~~
' ,,~~,, CA 02337079 2001-Ol-06 , . , .. f f : f V f t : t t I t ( 1 f -a t~f. t, f t t f t' , t d y t . d :ljt t t t d f t t ~ Y Y S f t f ( t a t t ~ f f t S f ( f F " 9 t t f f f f t f. t ~ f . f C t f f : f t t Common errors in matching phrases are missing, extra or changed words.
Hence the edit distance metric as previously defined shall be adapted and extended in order to apply to the approximate word sequence snatching problem. Edit operations for sequences are defined below.
Definition 6: Edit operations for sequences For transforming one sequence S of words into another sequence P of words, the edit operations allowed on the word in the sequences may be written according to the following rewrite rules: ' ~ (ass), deletion of word a from the sequence ~ (spa), insertion of word a into the sequence ~ (a~b), change of word a into word b ~ (ab-~ba), transposition of adjacent words a and b.
Instead of characters as atoms, the search system according to the invention applies the edit operations to words which then should be regarded as the operational atoms. .
A cost function cealr(x~Y) is a constant which is defined as 1 delete _ 1 insert edit ~'~' ~ .~~ 2 1 transpose a(x, y) change where a(x,y) is defined as ~y~= 0 x=y ( a x 1 else By using the edit operations as defined above the edit distance for sequences can now be defined.
Definition 7: Edit distance for sequences The edit distance metric for sequences defines the distance DSe9(S,P) between the sequence S = sl,s,~, ...,sn and the sequence P = pl,p~, ...,p,n as the minimum sum of cost c(x-~y) for the sequence of edit operations transforming the sequence S into the sequence P.
The search system according to the present invention enhances the edit distances metric for sequences to weight the cost of the edit operations by the ~~a~tD 5~1~
CA 02337079 2001-Ol-06 -t t , t t ; t. t t r t t d -n , ~ ~ ~ ~ a ~ a t Y t f ~ t i f ! f ~
F t t f . t C~~~'t i t C C - t t P ( t t t f t t d t t f d t F f i t ~ t t i i d F t F 1 t i t f i t t f' t E d 2 t d size of the words operated. upon.
Definition 8: Word size-dependent edit distance for sequences _ The word size dependent edit distance for sequences is defined as the minimum sum of costs for the editing operations needed to transform one sequence into the other. The cost functions axe dependent on the word size of its operands.
In the search system according to the invention a definition of cost functions is given by the equations Chtsert (E ~ a) _ l al .
Z
Cdelete (a ~ ~) _ l al l (4) ~trpnspose (ab -~ ba) =1 max(Ila~-Ibll,1) Cchnnge (a ~ b~ = l where Z denotes the average length of a word in the two sequences being compared. The cost of each edit operation is weighted by a size proportional to the change in the total length of the sequence or by the ratio of the current word length and the average word length in the sequences considered.
Now the distance metric reflects the assumption of some relation between the word length and how important the word is to the semantic context of the word sequence. Furthermore the search system according to the invention employs proximity at the character level when the change edit operation (a~b) is used. Replacing a word a by another word b should be related to the similarity between these two words. The new cost function fox the change edit operation hence is given as:
max a-b ,l Cchange (a ~ b~ - C'3aPProx (a~ b~ ~~ I I ~~ (5) l where a~PPrax (a~ bJ - 1.7(a, b) Where D(a, b) is the normalized edit distance measuring function for words, 0 means full similarity, 1 means no similarity.
~s~~
CA 02337079 2001-Ol-06 r r ( i , t : c W r ' . ~ a r t r ( C ( 'Y , a t r r t ( ( t r ( L'S ( a ( t ( ( ( t ( f t 1 t ( < F 't t ( ' ( t 1 t t t F . t ( i ( F ( i ( t i f' i i i t E f E t ( E
The search system according to the invention combines the edit distance metric for sequences with the cost functions as given by formulas (4), (5) and (6), with an edit distance metric for words as given by formula (1). This means that sequence edit operations are only used when the words being matched are more than k errors away from each other.
The algorithms used in the search system according to the invention perform efficient searching of the described structures. Matches are found according to the metrics as given above.
Approximate word matching in a word-spaced sparse suffix tree is done by combining the calculation of the edit distance matrix and a traversal of the suffix tree. An algorithm for this is written in pseudo-code and given in table I.
This algorithm is adapted from a trie-matching algorithm as proposed by H.
Shang 8~ T.H. Merrettal, "Tries for Approximate String Matching", IEEE
Transactions on Knowledge and Data Engineering, Vol.S, No. 4, pp. 540-547 (1996). The expected worst case running time of the algorithm is O(k~E~k) according to Shang ~ Merrettal (op.cit.).
Approximate word sequence matching requires the calculation of the word sequence edit distance for all possible matches. However, the number of possible matches can be limited by starting the calculation of the edit distance only on the possible words. The cost of deleting a word from the sequences determines the number of possible start words. If the accumulated cost of deleting the i first words in a query sequence PQ rises above a given error threshold, the candidate sequence starting with the ith word of the query cannot possibly be a match. Therefore for a query sequence PQ of i words, at most i possible start words will be tried. Since there are no baclcpointers in the sequence structure of the tree, it will not be ensured that all possible matches are obtained. Adding backpointers would solve this problem. The algorithm for approximate word-sequence matching as used in the search system according to the present invention, is given in pseudo-code in table II below. This algorithm tries to match the first keyword with pl,p2...
sequentially, testing all possible start positions.
Aid .~wEf~
' CA 02337079 2001-Ol-06 , v ~ t f f ~ t a f 1 t t F t ~ i f t S ~ f t ,. .. _ ~ .. . i ~ t i t t L6ttt t f t Y f F t , t S t i t S ( t E t ' t t t t S t t t s Y 3'ttF .r . if t: Y Lt St In the ApproxSequenceMatch algorithm in table II the App~oxMateIzRest function is defined by the algorithm in table III below. This function matches the remaining sequence, using an initial error value.
> ' CA 02337079 2001-Ol-06 t t t f i t t t t r t r t ' ~ ' ~ ' ' ~ c f . i F ' F f a t ( f t J 1 ~ ~~ ( t ~ i a i S 7 i f C t f i I t y . r t C i F f i f F f S F
° F F A i ( S F t i f t i t t t t Table I
FindApproximate (root,p,k) node E- root;
i E- 1;
nodes ~- Children (node) ; // A stack of nodes for all v E nodes do if IsZeaf (v) then for j E- i to length (Suffix (node) ) do w~ F- Suffix (node) ~_i;
if w~ - '$' then // '$' is a stopchar output wl.,_~ ;
return;
if EditDist (i) - co then break;
else //Inte rnal node .7. ~ 1-~1;
wi ~- label ( v) if EditDist (i) - oo then , break;
nodes E- Children ( v) U nodes;
// end for EditDistance(j) // Calculates jth row for i~ 1 to length ( P) do if pi = w~ then 8 ~- 0 else 8 ~ 1:
Cl = D~i-lr j~ ,+ Cins (mj) ;
C2. = D~i,j-1~ + Cdel (pi) i Cg = D~1-l, j-1~ -f- Cchange (pir mj) i Cq = D~'1-~, j-2~ -1- Ctranspose(pirmj-1) i D~1, j~ ~ Cfraction (~ ~-Z ) ' I~1.1.11 ( C1, C~, C3, Cq ) ;
a.f D[i, j~ > k return oo; // No distance below k return D(i, j~
CA 02337079 2001-Ol-06 t t t a t f t: f , . s E r f t 1 f t f [ f < ~ f f . ' E f ( t~ ~.' t t t f~ C f f t f E C . S t t F i t E C t f E C t f t ' ( t f 5 f t t 6 . v l l i E Y F C E f v Table II
- ApproxSequenceMatch ED (root, P( pl,,pa.~--.Pmt ~k) m~.~p) matches startError ~ 0 startTndex E- 1 while startError _< k ~R startlndex S m do StartNOde E- .FlndExact (,pStartlndex~ ~
list ~ UnorderedOccurrence.List (startNodeJ ;
for all v~E lists do if ApproxMatchRest2 (v,P,k,startError) then matches ~ U v;
startError ~- startError + Cdel (,jJstartlndex~
startlndex E- startlndex + 1;
t~.~ y ~ ~_;~s a =3 ~
~ ~ CA 02337079 2001-Ol-06 t F i t t F r F r r c c <
F, t, t t ~ i t ~ f a c F L f r t il~ s~ f < . t F r c f t t t f t a r t ' f f ~ F t t f ( t ( F f T
t c t ! t ( f i i f F 3 f 7 ! ! t Table III
ApproxMa tch.Rest ( a o P, K, s tartError) error E- startError;
lastError ~ startError;
column ~ O ;
node ~ u;
for a ~ p2 to p~p~ do node ~ NextOccurrence(node);
word ~ Keyword(node);
IastError ~- error;
error ~- startError + EditDistance (column);
if error > k AND IastError > k then return false;
return true;
Edit.Distance(j) // Calculates jth row for iE- 1 to length ( P) do if pi = w~ then a ~ 0 else r7 E- 1;
Cl = D[1-l, ]' + Cins (mj) i C~ = D[1, j-1~ + Cdel (pi) i Cg =~ D[2-l, j-1~ + C~hange(~pirmj) r Cq = D[1-~, j-2~ + Ctranspose(pirmj-1) i D[.1, j~ ~ Cfraction (,~l-Z ) ' Illlll ( C1, C~, C3, Cq ) ;
if D[i, j] >- k return oo; // No distance below k return D[i, j~
~;~~~r~Q s~~
CA 02337079 2001-Ol-06 a ~ . _ t f r a t c ; s r a t i t , ' i~i t f t f ' a t c t ~ t z t : f t ' F S f f a ~ t f F t t' 1 f ! S
S t ' d I f ~ t f t t : < Z F 1 t f t C f t f ~ f' O t a S r F f t t The algorithms in tables II and III are written in the same pseudo-code as the algorithm in table I.
The Fi~cclExact function used to find the leaf node matching the first word in the sequence performs a simple traversal of the tree and its running time is O~pj~ where pl denotes the first word in a query sequence PQ. Calculating the edit distance can be done in ~P~2 time using straightforward dynamic programming or in O(k) time (where k denotes the error threshold) using improved versions of the calculation algorithm, see E. Ukkonen, "Finding Approximate Patterns in Strings", Journal of Algorithms, vol. 6, pp. 132-137 IO (1985).
If ~~a~~(p;) denotes the total sum of the number of occurrences of each word pj in the word sequence, then the worst case running time is O(k~~co~C(pz)).
Finally the implementation of a search engine based on the search system according to the invention shall briefly be discussed. Particularly a search engine based on the search system according to the invention is implemented as an approximate search engine (ASE) and is intended as a search engine for indexing large document collections and providing algorithms for exact and approximate searching of these document collections. ASE shall provide a data structure for storing large texts or collection of documents. It is to be understood that the data structure may be generated from documents which contain additional information, such as images, video, sound, and the text may be formatted and/or annotated. The data structure is identical to the word-spaced sparse suffix tree as discussed above and it is, of course, to be understood that the words is the keywords of the search system, hence the word-spaced sparse suffix tree may instead be termed a keyword-spaced sparse suffix tree (KWS tree). The ASE shall contain algorithms for indexing documents in the KWS tree. These algorithms, of course, do not form a part of the search system according to the present invention, but they are well-known to persons skilled in the art and described in the literature, see for instance J. Karkkainen & E. ZJkkonen (op.cit.) and D.R. Morrison (op. cit) .
The search system according to the invention and as used in the ASE
employs algorithms both for exact and approximate matching of a pattern in a KWS tree: The algorithms given above in table I and table II are used for approximate word and word sequence snatching with the non-uniform edit ~~~D 5:~~
CA 02337079 2001-Ol-06 , . ! I . .
a i S n . . a t . , t .. . ' ~i . < , i ~~lCti t [
t i . i F [ [ t f t k I d ~ . ( [ d < ! f t ; .
t f i F t f S ( t F ; ~ d' t i S t distance as a metric. Finding an exact match of keyword p with length ~a in a .
KWS tree is known in the art and easily implemented as a simple traversal of the tree structure. An appropriate algorithm for exact keyword matching written in pseudo-code is given in table IV. The search system according to the invention also shall be able to support algorithms for exact keyword sequence matching. Algorithms for exact keyword sequence matching are known in the art and easily implemented as e.g. shown in pseudo-code in table V below. The algorithm given here will find the exact match of the first keyword, if any. Then it will for all occurrences of the first keyword check if the second keyword matches the second keyword of the query. If so, the MatchRest procedure in table V is used to determine if the occurrence of the two first keywords are matching in the entire sequence. For approximate keyword matching in a KWS tree the search system implements the algorithm in table I above. For approximate lceyword sequence snatching the search system implements the algorithm in table II above, matching a first keyword sequentially with p,,p~... and testing all possible start positions, applying the ApproxMatchRest function as given in table III to match a sequence starting at a particular position and handle the initial error value.
Finally, the ASE shall need a simple front end which gives the user control of indexing and querying the document collection. The front end should also be able to furnish statistics of the document collection and provide both a networlc interface for remote access, e.g. via WWW, and a local server user interface.
The ASE with the search system according to the invention should be general in a manner that allows for the adding new indexing and searching algorithms easily. Also, storing extra information about each document or keyword shall be possible to implement in an easy manner. Particularly the front end should be independent of the data structure and the search algorithms, such that internal changes in these has no effect on the design of the former.
The use of the search system according to the invention the ASE should be designed to have as low memory overhead as possible in the data structure.
Also,,searching should be designed to be as fast as possible. However, there will usually be a trade-off between these two factors.
~~~~3 ~i~~~
i CA 02337079 2001-Ol-06 f 1 t 3 1 .( i v F r ! s t ., . . ' , . t . ' ~ ; t' ;.~
t ~ , 9 i r Jy Y t , R t t t G t 5 c i t t t a t t , C r t t E f t t ' f f f t r F ! F t f' f a C r t t f .
Table IV
FindExact (root,p) _ i~ 1;
node ~- Find Child (root,pi) ;
while node AND i _< length (p) do if IsLeaf (node) AND Suffix (node) - Pi...Pm then return node;
i ~ i+1 node ~- FindChild (node, pi);
return NIL;
Table V
MatchSequenceExact (P., root) ma tches ~ Q~;
v ~ FindExact (pl,root) ;
if ~ P ~ > 1 then if v ~ NIL then list ~ UnorderedOccurrenceList (v);
for all a E List do if NextKeyword(u) - p2 then if MatchRest (p3...pm, u) then matches ~ matches U Occurrence (u) ;
return matches;
MatchRest (P, u) node ~
u;
for v ~ p1 to p~P~ do ~
node ~ NextOccurrence(node);
word ~ Keyword(node);
if v ~ word then return false;
a~~~~~~.~ yAt~~~, °
, CA 02337079 2001-Ol-06 i r c r t ; r f r t r r r r s' t f r a r i a r r r t t t t - r t23 , trt i~ t i ( i: f C° r f f t t C a f f f C ( ( t f v f ( f ( f t i f f t f f f t f C . t f C C t' r f To sum up, an ASE with a search system according to the invention shall comprise four major modules.
1. Document indexing module DIM for indexing documents in the KWS tree structure. This module should also contain all extensions to support several document types.
2. Data storage module DSM based on a keyword-spaced sparse suffix tree (KWS tree).
3. Search algorithm module SAM for searching the KWS tree, comprising algorithms for exact and/or approximate matching of respectively words and word sequences.
4. User interface front-end module FEM comprising both a local server user interface and a networlc interface for remote queries.
The four modules of the ASE works together to offer a complete search engine functionality. The data flow between the different modules is shown in fig. 7. Indexing a collection of documents in done in the document indexing module DIM comprising indexing algorithms. This module is, of course, not a part of the search system according to the invention, but indexing algorithms that can be used are well-lcnown in the art. The text found in the documents is passed on to the data storage DSM module for storage. The data storage module is, of course, a part of the search system according to the invention and is as stated based on the KWS tree structure.
The search algorithm module SAM contains algorithms for searching the data located in the data storage module. This module implements the search system according to the present invention and allows for a search process querying the data structure for tree and node information, while maintaining state variables. The front-end module may for instance be implemented on a work station or a personal computer and the like, providing the functionality as stated above.
As already stated in the introduction, it is to be understood that the search system according to the invention can be implemented as software written in a suitable high-level language on commercially available computer systems, including workstations. It may also as stated be implemented in the form of a dedicated processor device which advantageously may comprise a large number of parallel processors being able to process large word sequences in ~l,~t~a SF
CA 02337079 2001-Ol-06 " ~ .. ~ '. x i f s't tse,t t'r ti ,x a I f t t I .~ t t c f_ t . t ~7L~. t t t f i ~ a t t t f ~ x t f ( I t t t t !
t a x t ! I I = t . t t ~ C f x f ! t C f t f r f f ! Y
parallel for approximate matching with a large number of query word sequences. The fixed operational parameters of the processor may then be entered in a low-level. code, while keyword sequences input from the KWS
tree structure allows for an extremely fast processing of queries on a huge S amount of data, and the search system according to present invention shall hence in high degree be suited for perfonning searches on e.g. the World Wide Web, even in a KWS tree structure large enough to index all documents .
presently offered on the World Wide Web and moreover capable of handling the expected data volume growth on the World Wide Web in the future.
A~~~(~ 5~~~~
The four modules of the ASE works together to offer a complete search engine functionality. The data flow between the different modules is shown in fig. 7. Indexing a collection of documents in done in the document indexing module DIM comprising indexing algorithms. This module is, of course, not a part of the search system according to the invention, but indexing algorithms that can be used are well-lcnown in the art. The text found in the documents is passed on to the data storage DSM module for storage. The data storage module is, of course, a part of the search system according to the invention and is as stated based on the KWS tree structure.
The search algorithm module SAM contains algorithms for searching the data located in the data storage module. This module implements the search system according to the present invention and allows for a search process querying the data structure for tree and node information, while maintaining state variables. The front-end module may for instance be implemented on a work station or a personal computer and the like, providing the functionality as stated above.
As already stated in the introduction, it is to be understood that the search system according to the invention can be implemented as software written in a suitable high-level language on commercially available computer systems, including workstations. It may also as stated be implemented in the form of a dedicated processor device which advantageously may comprise a large number of parallel processors being able to process large word sequences in ~l,~t~a SF
CA 02337079 2001-Ol-06 " ~ .. ~ '. x i f s't tse,t t'r ti ,x a I f t t I .~ t t c f_ t . t ~7L~. t t t f i ~ a t t t f ~ x t f ( I t t t t !
t a x t ! I I = t . t t ~ C f x f ! t C f t f r f f ! Y
parallel for approximate matching with a large number of query word sequences. The fixed operational parameters of the processor may then be entered in a low-level. code, while keyword sequences input from the KWS
tree structure allows for an extremely fast processing of queries on a huge S amount of data, and the search system according to present invention shall hence in high degree be suited for perfonning searches on e.g. the World Wide Web, even in a KWS tree structure large enough to index all documents .
presently offered on the World Wide Web and moreover capable of handling the expected data volume growth on the World Wide Web in the future.
A~~~(~ 5~~~~
Claims (9)
1. A search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words and/or symbols and sequences thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric M which measures the degree of matching between the query Q and retrieved information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords kw, characterized in that the data structure comprises a tree structure in the form of a non-evenly spaced sparse suffix tree ST(T) for storing suffixes of words and/or symbols s and sequences S thereof in the text T, that the metric M comprises a combination of an edit distance metric for an approximate degree of matching between words and/or symbols in the text T and a query Q and an edit distance metric for an approximate degree of matching between sequences S of words and/or symbols in the text T and a query sequence P of words and/or symbols in the query Q, the latter edit distance metric including weighting cost functions for edit operations which transform a sequence S of words and/or symbols in the text T into the sequence P of words and/or symbols in the query Q, the weighting taking place with a value proportional to a change in the length of sequence upon a transformation or dependent on the size of the words and/or symbols in sequences to be matched, that the implemented search algorithms comprise a first algorithm for determining the degree of matching between words and/or symbols in the suffix tree representation of respectively the text T and a query Q, and a second algorithm for determining the degree of matching between sequences of words and/or symbols in the suffix tree representation of respectively the text T and the query Q, said first and/or second algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words or sequences of symbols or combinations thereof, such that information R is retrieved on the basis of query Q with a specified degree of matching between the former and the latter, and that the search algorithms optionally also comprise a third algorithm for determining exact matching between words and/or symbols in the suffix tree representation of respectively the text T and the query Q and/or a fourth algorithm for determining exact matching between sequences of words and/or symbols in the suffix tree representation of respectively the text T and the query Q, said third and/or fourth algorithms searching the data structure with queries Q in the form of either words, symbols, sequences of words, or sequences of symbols or combinations thereof, such that information R is retrieved on the basis of the query Q with an exact matching between the former and the latter.
2. The search system according to claim 1, wherein the non-evenly spaced sparse suffix tree ST(T) is a word-spaced sparse suffix tree SST ws (T) comprising only a subset of the suffixes in the text T.
3. The search system according to claim 2, wherein the word-spaced sparse suffix tree SST ws (T) is a keyword-spaced sparse suffix tree SST kws (T).
4. The use of a search system according to claim 1 in an approximate search engine.
5. A method in a search system for information retrieval, particularly information stored in form of text, wherein a text T comprises words and/or symbols and sequences S
thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric M which measures the degree of matching between the query Q and retrieved information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords kw, wherein the information in the text T is divided into words and word sequences, the words being substrings of the entire text separated by word boundary terms and forming a sequence of symbols, and wherein each word is structured as a sequence of symbols, characterized by generating the data structure as a word-spaced sparse suffix tree SSTws (T) of a text T for representing all the suffixes starting at a word separator symbol in the text T, storing sequence information of the words s in the text T
in the word-spaced sparse suffix tree SSTws (T), generating a combined edit distance metric M comprising an edit distance metric D(s,q) for words in the text T and a query word q in a query Q and a word-size dependent edit distance metric Dws (S,P) for sequences S of words s in the text T and a sequence P of words q in the query Q, the edit distance metric Dws (S,P) being the minimum sum of costs for edit operations transforming a sequence S into the sequence P, the minimum sum of costs being the minimum sum of cost functions for each edit operation weighted by a value proportional to the change in the total length of the sequence S or by the ratio of the current word length and average word length in the sequences S;P, and determining the degree of matching between words s,q by calculating the edit distance D(s,q) between the words s of the retrieved information R and the word q of a query Q, or in case the words s,q are more than k errors from each other, determining the degree of matching between the word sequences SR ; PQ of retrieved information R and a query Q respectively by calculating the edit distance Dws (S R,P Q) for all matches.
thereof, wherein the information retrieval takes place with a given or varying degree of matching between a query Q, wherein the query Q comprises words and/or symbols and sequences thereof, and retrieved information R comprising words and/or symbols and sequences thereof from the text T, wherein the search system comprises a data structure for storing at least a part of the text T, and a metric M which measures the degree of matching between the query Q and retrieved information R, and wherein the search system implements search algorithms for executing a search, particularly a full text search on the basis of keywords kw, wherein the information in the text T is divided into words and word sequences, the words being substrings of the entire text separated by word boundary terms and forming a sequence of symbols, and wherein each word is structured as a sequence of symbols, characterized by generating the data structure as a word-spaced sparse suffix tree SSTws (T) of a text T for representing all the suffixes starting at a word separator symbol in the text T, storing sequence information of the words s in the text T
in the word-spaced sparse suffix tree SSTws (T), generating a combined edit distance metric M comprising an edit distance metric D(s,q) for words in the text T and a query word q in a query Q and a word-size dependent edit distance metric Dws (S,P) for sequences S of words s in the text T and a sequence P of words q in the query Q, the edit distance metric Dws (S,P) being the minimum sum of costs for edit operations transforming a sequence S into the sequence P, the minimum sum of costs being the minimum sum of cost functions for each edit operation weighted by a value proportional to the change in the total length of the sequence S or by the ratio of the current word length and average word length in the sequences S;P, and determining the degree of matching between words s,q by calculating the edit distance D(s,q) between the words s of the retrieved information R and the word q of a query Q, or in case the words s,q are more than k errors from each other, determining the degree of matching between the word sequences SR ; PQ of retrieved information R and a query Q respectively by calculating the edit distance Dws (S R,P Q) for all matches.
6. The method according to claim 5, further comprising additionally weighting an edit operation which changes a word s into word q with a parameter for the proximity between the characters of the words s;q, thus taking the similarity of the words s;q in regard when determining the cost of the edit operation in question.
7. The method according to claim 5, further comprising limiting the number of matches by calculating the edit distance Dws (S R,P Q) for restricted number of words in the query word sequence P Q.
8. The method according to claim 5, further comprising defining the edit distance D(s,q) between words s and a word q recursively and calculating the edit distance D(s,q) by means of a dynamic programming procedure.
9. The method according to claim 5, further comprising defining the edit distance Dws (S,P) between sequences S and a sequence P recursively and calculating the edit distance Dws (S,P) by means of a dynamic programming procedure.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CA002524207A CA2524207A1 (en) | 1998-07-10 | 1999-07-09 | A search system and method for retrieval of data, and the use thereof in a search engine |
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
NO983175A NO983175L (en) | 1998-07-10 | 1998-07-10 | Search system for data retrieval |
NO19983175 | 1998-07-10 | ||
PCT/NO1999/000233 WO2000003315A2 (en) | 1998-07-10 | 1999-07-09 | A search system and method for retrieval of data, and the use thereof in a search engine |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002524207A Division CA2524207A1 (en) | 1998-07-10 | 1999-07-09 | A search system and method for retrieval of data, and the use thereof in a search engine |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2337079A1 CA2337079A1 (en) | 2000-01-20 |
CA2337079C true CA2337079C (en) | 2006-07-04 |
Family
ID=19902243
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002337079A Expired - Lifetime CA2337079C (en) | 1998-07-10 | 1999-07-09 | A search system and method for retrieval of data, and the use thereof in a search engine |
Country Status (20)
Country | Link |
---|---|
US (1) | US6377945B1 (en) |
EP (1) | EP1095326B1 (en) |
JP (1) | JP3581652B2 (en) |
KR (1) | KR100414236B1 (en) |
CN (1) | CN1317114A (en) |
AT (1) | ATE212736T1 (en) |
AU (2) | AU772525B2 (en) |
BR (1) | BR9912015B1 (en) |
CA (1) | CA2337079C (en) |
DE (1) | DE69900854T2 (en) |
DK (1) | DK1095326T3 (en) |
ES (1) | ES2173752T3 (en) |
HK (1) | HK1040784A1 (en) |
HU (1) | HUP0201630A2 (en) |
IL (1) | IL140606A0 (en) |
IS (1) | IS5796A (en) |
NO (1) | NO983175L (en) |
PL (1) | PL345714A1 (en) |
PT (1) | PT1095326E (en) |
WO (1) | WO2000003315A2 (en) |
Families Citing this family (122)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6564206B1 (en) * | 1998-10-05 | 2003-05-13 | Canon Kabushiki Kaisha | Information search apparatus and method, and storage medium |
US6584465B1 (en) * | 2000-02-25 | 2003-06-24 | Eastman Kodak Company | Method and system for search and retrieval of similar patterns |
US6745181B1 (en) * | 2000-05-02 | 2004-06-01 | Iphrase.Com, Inc. | Information access method |
US6704728B1 (en) * | 2000-05-02 | 2004-03-09 | Iphase.Com, Inc. | Accessing information from a collection of data |
US7127450B1 (en) * | 2000-05-02 | 2006-10-24 | International Business Machines Corporation | Intelligent discard in information access system |
US6711561B1 (en) * | 2000-05-02 | 2004-03-23 | Iphrase.Com, Inc. | Prose feedback in information access system |
US8478732B1 (en) | 2000-05-02 | 2013-07-02 | International Business Machines Corporation | Database aliasing in information access system |
US6718325B1 (en) * | 2000-06-14 | 2004-04-06 | Sun Microsystems, Inc. | Approximate string matcher for delimited strings |
US6408277B1 (en) | 2000-06-21 | 2002-06-18 | Banter Limited | System and method for automatic task prioritization |
US8290768B1 (en) | 2000-06-21 | 2012-10-16 | International Business Machines Corporation | System and method for determining a set of attributes based on content of communications |
US9699129B1 (en) | 2000-06-21 | 2017-07-04 | International Business Machines Corporation | System and method for increasing email productivity |
US6915294B1 (en) * | 2000-08-18 | 2005-07-05 | Firstrain, Inc. | Method and apparatus for searching network resources |
US7080073B1 (en) | 2000-08-18 | 2006-07-18 | Firstrain, Inc. | Method and apparatus for focused crawling |
US7103838B1 (en) * | 2000-08-18 | 2006-09-05 | Firstrain, Inc. | Method and apparatus for extracting relevant data |
US20020152202A1 (en) * | 2000-08-30 | 2002-10-17 | Perro David J. | Method and system for retrieving information using natural language queries |
GB2368414B (en) * | 2000-10-27 | 2002-09-11 | One Stop To Ltd | Searching procedures |
US7644057B2 (en) * | 2001-01-03 | 2010-01-05 | International Business Machines Corporation | System and method for electronic communication management |
US6681219B2 (en) * | 2001-03-29 | 2004-01-20 | Matsushita Electric Industrial Co., Ltd. | Method for keyword proximity searching in a document database |
US6976016B2 (en) * | 2001-04-02 | 2005-12-13 | Vima Technologies, Inc. | Maximizing expected generalization for learning complex query concepts |
US7136846B2 (en) | 2001-04-06 | 2006-11-14 | 2005 Keel Company, Inc. | Wireless information retrieval |
US7054855B2 (en) * | 2001-07-03 | 2006-05-30 | International Business Machines Corporation | Method and system for performing a pattern match search for text strings |
US7209913B2 (en) * | 2001-12-28 | 2007-04-24 | International Business Machines Corporation | Method and system for searching and retrieving documents |
US7343372B2 (en) * | 2002-02-22 | 2008-03-11 | International Business Machines Corporation | Direct navigation for information retrieval |
US7287026B2 (en) * | 2002-04-05 | 2007-10-23 | Oommen John B | Method of comparing the closeness of a target tree to other trees using noisy sub-sequence tree processing |
US6928445B2 (en) * | 2002-06-25 | 2005-08-09 | International Business Machines Corporation | Cost conversant classification of objects |
FI121583B (en) * | 2002-07-05 | 2011-01-14 | Syslore Oy | Finding a Symbol String |
US6983280B2 (en) * | 2002-09-13 | 2006-01-03 | Overture Services Inc. | Automated processing of appropriateness determination of content for search listings in wide area network searches |
US7970621B2 (en) * | 2002-10-18 | 2011-06-28 | Cerner Innovation, Inc. | Automated order entry system and method |
US20050187913A1 (en) | 2003-05-06 | 2005-08-25 | Yoram Nelken | Web-based customer service interface |
US8495002B2 (en) * | 2003-05-06 | 2013-07-23 | International Business Machines Corporation | Software tool for training and testing a knowledge base |
CA2429909A1 (en) * | 2003-05-27 | 2004-11-27 | Cognos Incorporated | Transformation of tabular and cross-tabulated queries based upon e/r schema into multi-dimensional expression queries |
CN1567303A (en) * | 2003-07-03 | 2005-01-19 | 富士通株式会社 | Method and apparatus for automatic division of structure document information block |
US7165119B2 (en) | 2003-10-14 | 2007-01-16 | America Online, Inc. | Search enhancement system and method having rankings, explicitly specified by the user, based upon applicability and validity of search parameters in regard to a subject matter |
US7640232B2 (en) | 2003-10-14 | 2009-12-29 | Aol Llc | Search enhancement system with information from a selected source |
US7657423B1 (en) * | 2003-10-31 | 2010-02-02 | Google Inc. | Automatic completion of fragments of text |
US7418455B2 (en) * | 2003-11-26 | 2008-08-26 | International Business Machines Corporation | System and method for indexing weighted-sequences in large databases |
US7343378B2 (en) * | 2004-03-29 | 2008-03-11 | Microsoft Corporation | Generation of meaningful names in flattened hierarchical structures |
IL161874A (en) * | 2004-05-07 | 2012-09-24 | Yves Villaret | System and method for searching strings of records |
EP1825395A4 (en) * | 2004-10-25 | 2010-07-07 | Yuanhua Tang | Full text query and search systems and methods of use |
US20080077570A1 (en) * | 2004-10-25 | 2008-03-27 | Infovell, Inc. | Full Text Query and Search Systems and Method of Use |
US20070266406A1 (en) * | 2004-11-09 | 2007-11-15 | Murali Aravamudan | Method and system for performing actions using a non-intrusive television with reduced text input |
US7895218B2 (en) | 2004-11-09 | 2011-02-22 | Veveo, Inc. | Method and system for performing searches for television content using reduced text input |
US20060101504A1 (en) * | 2004-11-09 | 2006-05-11 | Veveo.Tv, Inc. | Method and system for performing searches for television content and channels using a non-intrusive television interface and with reduced text input |
CN1614607B (en) * | 2004-11-25 | 2011-08-31 | 中国科学院计算技术研究所 | Filtering method and system for e-mail refuse |
CA2509496A1 (en) * | 2005-06-06 | 2006-12-06 | 3618633 Canada Inc. | Search-enhanced trie-based syntactic pattern recognition of sequences |
US9041744B2 (en) | 2005-07-14 | 2015-05-26 | Telecommunication Systems, Inc. | Tiled map display on a wireless device |
US7779011B2 (en) * | 2005-08-26 | 2010-08-17 | Veveo, Inc. | Method and system for dynamically processing ambiguous, reduced text search queries and highlighting results thereof |
US7788266B2 (en) | 2005-08-26 | 2010-08-31 | Veveo, Inc. | Method and system for processing ambiguous, multi-term search queries |
US7644054B2 (en) * | 2005-11-23 | 2010-01-05 | Veveo, Inc. | System and method for finding desired results by incremental search using an ambiguous keypad with the input containing orthographic and typographic errors |
US7882119B2 (en) * | 2005-12-22 | 2011-02-01 | Xerox Corporation | Document alignment systems for legacy document conversions |
US7664343B2 (en) * | 2006-01-23 | 2010-02-16 | Lockheed Martin Corporation | Modified Levenshtein distance algorithm for coding |
US7529741B2 (en) | 2006-03-06 | 2009-05-05 | Veveo, Inc. | Methods and systems for segmenting relative user preferences into fine-grain and coarse-grain collections |
US8073860B2 (en) | 2006-03-30 | 2011-12-06 | Veveo, Inc. | Method and system for incrementally selecting and providing relevant search engines in response to a user query |
WO2007124429A2 (en) | 2006-04-20 | 2007-11-01 | Veveo, Inc. | User interface methods and systems for selecting and presenting content based on user navigation and selection actions associated with the content |
CN100401300C (en) * | 2006-04-29 | 2008-07-09 | 上海世纪互联信息系统有限公司 | Searching engine with automating sorting function |
US20070260595A1 (en) * | 2006-05-02 | 2007-11-08 | Microsoft Corporation | Fuzzy string matching using tree data structure |
US7849078B2 (en) * | 2006-06-07 | 2010-12-07 | Sap Ag | Generating searchable keywords |
US8577328B2 (en) | 2006-08-21 | 2013-11-05 | Telecommunication Systems, Inc. | Associating metro street address guide (MSAG) validated addresses with geographic map data |
EP2062171A4 (en) * | 2006-09-14 | 2010-10-06 | Veveo Inc | Methods and systems for dynamically rearranging search results into hierarchically organized concept clusters |
JP4274221B2 (en) * | 2006-10-02 | 2009-06-03 | ソニー株式会社 | Information processing apparatus and method, program, and recording medium |
WO2008045690A2 (en) | 2006-10-06 | 2008-04-17 | Veveo, Inc. | Linear character selection display interface for ambiguous text input |
US8078884B2 (en) | 2006-11-13 | 2011-12-13 | Veveo, Inc. | Method of and system for selecting and presenting content based on user identification |
US8131722B2 (en) * | 2006-11-20 | 2012-03-06 | Ebay Inc. | Search clustering |
US20080172380A1 (en) * | 2007-01-17 | 2008-07-17 | Wojciech Czyz | Information retrieval based on information location in the information space. |
WO2008104621A1 (en) | 2007-02-28 | 2008-09-04 | Classe Qsl, S.L. | System for the recovery of information units |
US8549424B2 (en) * | 2007-05-25 | 2013-10-01 | Veveo, Inc. | System and method for text disambiguation and context designation in incremental search |
US20090055436A1 (en) * | 2007-08-20 | 2009-02-26 | Olakunle Olaniyi Ayeni | System and Method for Integrating on Demand/Pull and Push Flow of Goods-and-Services Meta-Data, Including Coupon and Advertising, with Mobile and Wireless Applications |
US20090055242A1 (en) * | 2007-08-24 | 2009-02-26 | Gaurav Rewari | Content identification and classification apparatus, systems, and methods |
US20090098889A1 (en) * | 2007-09-11 | 2009-04-16 | Bob Barcklay | Wireless device location alerts on battery notification events |
US7716228B2 (en) * | 2007-09-25 | 2010-05-11 | Firstrain, Inc. | Content quality apparatus, systems, and methods |
US7904433B2 (en) * | 2007-10-09 | 2011-03-08 | O2Micro International Limited | Apparatus and methods for performing a rule matching |
US20090144247A1 (en) * | 2007-11-09 | 2009-06-04 | Eric Wistrand | Point-of-interest panning on a displayed map with a persistent search on a wireless phone using persistent point-of-interest criterion |
CN101246501B (en) * | 2008-03-27 | 2010-06-23 | 腾讯科技(深圳)有限公司 | Method and system for polymerizing the same subject network document files |
WO2009126231A1 (en) * | 2008-04-07 | 2009-10-15 | Telecommunication Systems, Inc | Proximity search for point-of-interest names combining inexact string match with an expanding radius search |
US8027990B1 (en) * | 2008-07-09 | 2011-09-27 | Google Inc. | Dynamic query suggestion |
US8027973B2 (en) * | 2008-08-04 | 2011-09-27 | Microsoft Corporation | Searching questions based on topic and focus |
US9424339B2 (en) | 2008-08-15 | 2016-08-23 | Athena A. Smyros | Systems and methods utilizing a search engine |
US7882143B2 (en) * | 2008-08-15 | 2011-02-01 | Athena Ann Smyros | Systems and methods for indexing information for a search engine |
US9092517B2 (en) * | 2008-09-23 | 2015-07-28 | Microsoft Technology Licensing, Llc | Generating synonyms based on query log data |
WO2010042172A1 (en) * | 2008-10-06 | 2010-04-15 | Telecommunication Systems, Inc. | Probabilistic reverse geocoding |
US8594627B2 (en) | 2008-10-06 | 2013-11-26 | Telecommunications Systems, Inc. | Remotely provisioned wirelessly proxy |
DE102008062830B3 (en) * | 2008-12-23 | 2010-07-22 | Cerebrix Gmbh | Information storing, searching and displaying device i.e. document management system, has assigning unit for assigning information to two data structures such that information is searchable and displayable |
US20100287177A1 (en) * | 2009-05-06 | 2010-11-11 | Foundationip, Llc | Method, System, and Apparatus for Searching an Electronic Document Collection |
US20100287148A1 (en) * | 2009-05-08 | 2010-11-11 | Cpa Global Patent Research Limited | Method, System, and Apparatus for Targeted Searching of Multi-Sectional Documents within an Electronic Document Collection |
US20100293179A1 (en) * | 2009-05-14 | 2010-11-18 | Microsoft Corporation | Identifying synonyms of entities using web search |
US8533203B2 (en) * | 2009-06-04 | 2013-09-10 | Microsoft Corporation | Identifying synonyms of entities using a document collection |
US9166714B2 (en) | 2009-09-11 | 2015-10-20 | Veveo, Inc. | Method of and system for presenting enriched video viewing analytics |
US8364679B2 (en) * | 2009-09-17 | 2013-01-29 | Cpa Global Patent Research Limited | Method, system, and apparatus for delivering query results from an electronic document collection |
US20110082839A1 (en) * | 2009-10-02 | 2011-04-07 | Foundationip, Llc | Generating intellectual property intelligence using a patent search engine |
WO2011056086A2 (en) * | 2009-11-05 | 2011-05-12 | Google Inc. | Statistical stemming |
US20110119250A1 (en) * | 2009-11-16 | 2011-05-19 | Cpa Global Patent Research Limited | Forward Progress Search Platform |
US8244754B2 (en) | 2010-02-01 | 2012-08-14 | International Business Machines Corporation | System and method for object searching in virtual worlds |
US20110191332A1 (en) | 2010-02-04 | 2011-08-04 | Veveo, Inc. | Method of and System for Updating Locally Cached Content Descriptor Information |
US8339094B2 (en) * | 2010-03-11 | 2012-12-25 | GM Global Technology Operations LLC | Methods, systems and apparatus for overmodulation of a five-phase machine |
US10643227B1 (en) | 2010-03-23 | 2020-05-05 | Aurea Software, Inc. | Business lines |
US8463789B1 (en) | 2010-03-23 | 2013-06-11 | Firstrain, Inc. | Event detection |
US10546311B1 (en) | 2010-03-23 | 2020-01-28 | Aurea Software, Inc. | Identifying competitors of companies |
US9760634B1 (en) | 2010-03-23 | 2017-09-12 | Firstrain, Inc. | Models for classifying documents |
US9600566B2 (en) | 2010-05-14 | 2017-03-21 | Microsoft Technology Licensing, Llc | Identifying entity synonyms |
EP2423830A1 (en) | 2010-08-25 | 2012-02-29 | Omikron Data Quality GmbH | Method for searching through a number of databases and search engine |
US8745061B2 (en) * | 2010-11-09 | 2014-06-03 | Tibco Software Inc. | Suffix array candidate selection and index data structure |
WO2012082859A1 (en) * | 2010-12-14 | 2012-06-21 | The Regents Of The University Of California | High efficiency prefix search algorithm supporting interactive, fuzzy search on geographical structured data |
US9069767B1 (en) | 2010-12-28 | 2015-06-30 | Amazon Technologies, Inc. | Aligning content items to identify differences |
US8798366B1 (en) | 2010-12-28 | 2014-08-05 | Amazon Technologies, Inc. | Electronic book pagination |
US9846688B1 (en) | 2010-12-28 | 2017-12-19 | Amazon Technologies, Inc. | Book version mapping |
US9881009B1 (en) | 2011-03-15 | 2018-01-30 | Amazon Technologies, Inc. | Identifying book title sets |
US8782042B1 (en) | 2011-10-14 | 2014-07-15 | Firstrain, Inc. | Method and system for identifying entities |
US8745019B2 (en) | 2012-03-05 | 2014-06-03 | Microsoft Corporation | Robust discovery of entity synonyms using query logs |
US8977613B1 (en) | 2012-06-12 | 2015-03-10 | Firstrain, Inc. | Generation of recurring searches |
US10032131B2 (en) | 2012-06-20 | 2018-07-24 | Microsoft Technology Licensing, Llc | Data services for enterprises leveraging search system data assets |
US9594831B2 (en) | 2012-06-22 | 2017-03-14 | Microsoft Technology Licensing, Llc | Targeted disambiguation of named entities |
US9229924B2 (en) | 2012-08-24 | 2016-01-05 | Microsoft Technology Licensing, Llc | Word detection and domain dictionary recommendation |
KR101355254B1 (en) * | 2012-10-22 | 2014-01-28 | 인하대학교 산학협력단 | Parallel suffix partitioning method for parallel suffix tree construction using trie |
US9256593B2 (en) | 2012-11-28 | 2016-02-09 | Wal-Mart Stores, Inc. | Identifying product references in user-generated content |
US10592480B1 (en) | 2012-12-30 | 2020-03-17 | Aurea Software, Inc. | Affinity scoring |
US9244952B2 (en) | 2013-03-17 | 2016-01-26 | Alation, Inc. | Editable and searchable markup pages automatically populated through user query monitoring |
US9589074B2 (en) | 2014-08-20 | 2017-03-07 | Oracle International Corporation | Multidimensional spatial searching for identifying duplicate crash dumps |
CN105446957B (en) | 2015-12-03 | 2018-07-20 | 小米科技有限责任公司 | Similitude determines method, apparatus and terminal |
CN108595584B (en) * | 2018-04-18 | 2022-06-07 | 卓望数码技术(深圳)有限公司 | Chinese character output method and system based on digital marks |
US10761965B2 (en) * | 2018-09-28 | 2020-09-01 | Atlassian Pty Ltd. | Detecting method calls based on stack trace data |
CN110517050A (en) * | 2019-08-12 | 2019-11-29 | 太平洋医疗健康管理有限公司 | A kind of medical insurance, which instead cheats to exchange, encodes digging system and method |
CN111415708B (en) * | 2020-03-24 | 2023-05-05 | 山东大学 | Method and system for realizing large-scale database clustering by double buffer model |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU620994B2 (en) * | 1989-07-12 | 1992-02-27 | Digital Equipment Corporation | Compressed prefix matching database searching |
AU1640992A (en) | 1991-03-08 | 1992-10-06 | Red Brick System | Compaction of a querying in relational databases |
US5511159A (en) | 1992-03-18 | 1996-04-23 | At&T Corp. | Method of identifying parameterized matches in a string |
US5953006A (en) * | 1992-03-18 | 1999-09-14 | Lucent Technologies Inc. | Methods and apparatus for detecting and displaying similarities in large data sets |
US5412807A (en) * | 1992-08-20 | 1995-05-02 | Microsoft Corporation | System and method for text searching using an n-ary search tree |
JPH07210569A (en) * | 1994-01-19 | 1995-08-11 | Oki Electric Ind Co Ltd | Method and device for retrieving information |
DE69422935T2 (en) | 1994-06-30 | 2000-08-17 | Ibm | METHOD AND DEVICE FOR COMPARING VARIABLE LENGTH DATA SEQUENCES |
JP2829259B2 (en) * | 1994-09-14 | 1998-11-25 | 株式会社東芝 | Data processing device, data search device, data processing method, and data search method |
US5724571A (en) * | 1995-07-07 | 1998-03-03 | Sun Microsystems, Inc. | Method and apparatus for generating query responses in a computer-based document retrieval system |
KR980004154A (en) * | 1996-06-29 | 1998-03-30 | 김광호 | Electronic dictionary using prefix and compression and retrieval method connected with it |
KR100284777B1 (en) * | 1997-11-19 | 2001-03-15 | 정선종 | Tri-Dictionaries for Map Terminology and How to Register and Search |
-
1998
- 1998-07-10 NO NO983175A patent/NO983175L/en not_active Application Discontinuation
-
1999
- 1999-07-09 CA CA002337079A patent/CA2337079C/en not_active Expired - Lifetime
- 1999-07-09 WO PCT/NO1999/000233 patent/WO2000003315A2/en not_active Application Discontinuation
- 1999-07-09 DK DK99933296T patent/DK1095326T3/en active
- 1999-07-09 ES ES99933296T patent/ES2173752T3/en not_active Expired - Lifetime
- 1999-07-09 DE DE69900854T patent/DE69900854T2/en not_active Expired - Lifetime
- 1999-07-09 HU HU0201630A patent/HUP0201630A2/en unknown
- 1999-07-09 PL PL99345714A patent/PL345714A1/en unknown
- 1999-07-09 US US09/486,726 patent/US6377945B1/en not_active Expired - Lifetime
- 1999-07-09 JP JP2000559494A patent/JP3581652B2/en not_active Expired - Lifetime
- 1999-07-09 AT AT99933296T patent/ATE212736T1/en not_active IP Right Cessation
- 1999-07-09 IL IL14060699A patent/IL140606A0/en unknown
- 1999-07-09 BR BRPI9912015-1A patent/BR9912015B1/en not_active IP Right Cessation
- 1999-07-09 AU AU49370/99A patent/AU772525B2/en not_active Expired
- 1999-07-09 PT PT99933296T patent/PT1095326E/en unknown
- 1999-07-09 EP EP99933296A patent/EP1095326B1/en not_active Expired - Lifetime
- 1999-07-09 KR KR10-2001-7000436A patent/KR100414236B1/en not_active IP Right Cessation
- 1999-07-09 CN CN99810507A patent/CN1317114A/en active Pending
-
2000
- 2000-12-29 IS IS5796A patent/IS5796A/en unknown
-
2002
- 2002-03-19 HK HK02102107.4A patent/HK1040784A1/en unknown
-
2004
- 2004-07-29 AU AU2004203480A patent/AU2004203480A1/en not_active Abandoned
Also Published As
Publication number | Publication date |
---|---|
AU2004203480A1 (en) | 2004-08-26 |
JP2002520712A (en) | 2002-07-09 |
CA2337079A1 (en) | 2000-01-20 |
JP3581652B2 (en) | 2004-10-27 |
KR100414236B1 (en) | 2004-01-07 |
DE69900854D1 (en) | 2002-03-14 |
EP1095326B1 (en) | 2002-01-30 |
EP1095326A1 (en) | 2001-05-02 |
US6377945B1 (en) | 2002-04-23 |
HUP0201630A2 (en) | 2002-08-28 |
PT1095326E (en) | 2002-07-31 |
IL140606A0 (en) | 2002-02-10 |
AU4937099A (en) | 2000-02-01 |
DE69900854T2 (en) | 2002-08-22 |
ATE212736T1 (en) | 2002-02-15 |
CN1317114A (en) | 2001-10-10 |
DK1095326T3 (en) | 2002-05-13 |
WO2000003315A3 (en) | 2000-02-24 |
KR20010071841A (en) | 2001-07-31 |
ES2173752T3 (en) | 2002-10-16 |
HK1040784A1 (en) | 2002-06-21 |
WO2000003315A2 (en) | 2000-01-20 |
AU772525B2 (en) | 2004-04-29 |
BR9912015A (en) | 2001-04-10 |
NO983175L (en) | 2000-01-11 |
BR9912015B1 (en) | 2011-12-13 |
PL345714A1 (en) | 2002-01-02 |
IS5796A (en) | 2000-12-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2337079C (en) | A search system and method for retrieval of data, and the use thereof in a search engine | |
US10671676B2 (en) | Multiple index based information retrieval system | |
US9817886B2 (en) | Information retrieval system for archiving multiple document versions | |
Faloutsos | Access methods for text | |
US6947920B2 (en) | Method and system for response time optimization of data query rankings and retrieval | |
CN107153647B (en) | Method, apparatus, system and computer program product for data compression | |
Jagadish et al. | On effective multi-dimensional indexing for strings | |
Malki | Comprehensive study and comparison of information retrieval indexing techniques | |
Kanlayanawat et al. | Automatic indexing for Thai text with unknown words using trie structure | |
CA2524207A1 (en) | A search system and method for retrieval of data, and the use thereof in a search engine | |
CN116028698B (en) | High-efficiency information searching system | |
Maarek | Introduction to Information Retrieval for Software Reuse. | |
Pandey et al. | A Novel Approach for Extraction of Relevant Web Pages from WWW Using Data Mining | |
Balaraman et al. | Using CBR Inexact Search for Intelligent Retrieval of Data | |
Kovács et al. | Document Clustering Using Concept Set Representation | |
André-Jösson et al. | Indexing time series using signatures | |
Thanuja et al. | Genetic Algorithm in Data Capturing and Mining | |
Shau-Yin et al. | Finding the maximum grid convex polygon for a convex region on the plane | |
CZ200164A3 (en) | Search system and method for retrieval of data, and the use thereof in a search engine | |
NO311657B1 (en) | Search system and method for retrieving data, and its use in a search engine |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKEX | Expiry |
Effective date: 20190709 |