Yammer logo on a window
Editorial

When Artificial Intelligence Sentiment Analysis Meets Yammer

5 minute read
Laurence Lock Lee avatar
SAVED
Bringing some of Microsoft's artificial intelligence capabilities to bear on collaboration tools opens up a new realm of productivity.

Microsoft is undoubtedly the leader in office productivity software, with generations having now grown up using its Office products suite. Less heralded in the productivity space, however, are the artificial intelligence (AI) capabilities Microsoft has developed over a similar time frame. 

Much of the AI found in the Office suite is hidden or embedded in functions like the spelling and grammar checker in MS Word. Other AI applications are more explicit, such as Cortana, which can, among other things, remind us of commitments we may have made but overlooked. Microsoft is making much of this AI capability available to development partners with API connections through its Cognitive Services unit. While it is still early days, we have observed some active piloted integrations with selected Microsoft products like Dynamics 365.

Putting Microsoft's AI Capabilities to Work

The Microsoft Cognitive Services offerings are broken into the following categories: Vision, Speech, Language, Knowledge and Search. How could a business make use of the Language, Knowledge and Search offerings in particular in the context of conversational platforms like Yammer and Teams? 

Using text analytics to identify emotion and sentiment across large repositories of conversational text could provide strong insights into culture and behaviors across whole organizations. Knowledge acquisition bots could respond to questions posed online by prompting responses from those who had previously demonstrated the capability to answer effectively, or potentially source the answer from previous similar questions. 

In this way, the problem solving capabilities of whole organizations and beyond could be dramatically improved in an efficient and timely manner.

Exploring Sentiment Analysis with Yammer

To further explore the possibilities in combining AI and productivity, we developed an integration between the Microsoft text analysis and Yammer to assess the real potential. We previously assessed the viability of the commercially available sentiment analysis products and found sentiment analysis can fall short when dealing with deeper cognitive themes, such as the use of irony or sarcasm. 

Learning Opportunities

Mark Parkinson, collaboration development senior manager at Mars Inc., spoke at Microsoft's Ignite conference about his company's use of AI and Yammer. He noted humans are also inconsistent when assessing sentiment in text, suggesting only a 70 percent agreement limit. Parkinson recommended augmenting sentiment classifications with human intelligence. By using sentiment to surface highly emotive and potentially negative posts on Yammer, community managers were able to apply their own overriding assessments on just a handful of cases, rather than having to trawl through the bulk of online conversations.

A Window Into Your Organization’s Culture?

While most AI deployments focus on helping with tactical challenges, we wanted to explore how something like AI and Yammer might provide a window into an overall organizational culture. Yammer may not be the most used application in the Office 365 suite, but our prior research had shown it hosts the most enterprise-wide conversations and therefore is a good place from which to explore enterprise culture.

After assessing a number of individual organizations, we anonymously combined the Yammer activities of four organizations operating across different industry sectors, encompassing some 20,000 employees, to provide a more generic picture of organizational culture. We undertook sentiment analysis of over 15,000 pages of text messages exchanged on Yammer over a six-month period. We then applied the results to enrich the behavioral profiles which reflect the different personas employees adopt when collaborating online. The large sample set made us confident that the cultural picture could be a rich one — and we weren’t disappointed. Our ‘cultural’ summary is provided below:

Online Behavioral Personas 

Cultural Signatures 

 
online collaboration personas: Catalyst

Catalysts spark reactions online. Sentiment analysis told us they:

  • Showed the most positive and most negative sentiments 
  • Made the longest posts on average 

Whatever the sentiment, Catalysts contribute "energy" to the organization and contribute to a positive culture.

 
online collaboration personas: Engager

Engagers connect networks by balancing their contributions with the reactions they receive. Sentiment told us they are:

  • Middle of the road for emotion — positive and negative sentiment. 
  • Their messages are on average second in length to the Catalyst 

Sentiment reinforces the Engager as those special individuals who can balance competing demands and their emotions at the same time. They can be the "engine room" for an organization in getting things done.

 
online collaboration personas: responders

Responders sustain networks by ensuring that participants are welcomed and supported. Sentiment showed us they:

  • Showed the most emotion, largely positive
  • Were the most succinct in their responses
  • Shared their positivity more broadly. 

Responders are good for an organization’s culture, by spreading positivity.

 
online collaboration personas: broadcaster

Broadcasters are concerned with sending a message, without much concern for engaging in conversation. Sentiment showed us that they are:

  • Largely emotionless in their delivery 
  • Were second to Responders on their breadth of audience 

Culturally, Broadcasters add little positivity to an organizational culture. Too many can even be destructive.

online collaboration personas: observer
 

Observers are those who infrequently join conversations. They may either be simply reading conversations or avoiding them alltogether. The Observer pattern was the most common though. Sentiment told as that they are:

  • Mostly negative in their sentiment 
  • The least engaging with others 
  • Had the narrowest reach 

Observers are often the "silent majority" hidden in an organization. Observers need to be transformed to more positive behavioral personas, if a positive cultural change is desired.

An organization’s culture is exhibited through the behaviors of its staff. Establishing the relative proportions of behavioral personas with the help of sentiment analysis, can we believe provide a rich picture of an organization’s evolving culture over time.

Bring AI Into the Productivity Conversation

When we can combine the power of AI and the conversational platforms contained within Office 365, the potential for breakthrough productivity movements could be immense. We have already seen evidence of how chatbots can help us in our day-to-day tasks. What we've demonstrated here is the potential to have an even deeper impact at the very heart of all organizations: its culture.

fa-solid fa-hand-paper Learn how you can join our contributor community.

About the Author

Laurence Lock Lee

Laurence Lock Lee is the co-founder and chief scientist at Swoop Analytics, a firm specializing in online social networking analytics. He previously held senior positions in research, management and technology consulting at BHP Billiton, Computer Sciences Corporation and Optimice. Connect with Laurence Lock Lee:

Main image: Cole Camplese