Sentiment Analysis is the processing of data to quantify something objective. In tyGraph for Yammer we use basic sentiment analysis to quantify the tone of messages on a scale of 0 to 1.
Sentiment Value
This is the value from zero to one we give to messages. We use the the Microsoft Azure Text Analytics API from Azure Cognitive Services to scan each message and return a sentiment value. We then average these values to calculate a conversation, group, or users sentiment score.
Some companies opt to only return Sentiment Values for a recent window (say the last 3 months of messages) when they first enable tyGraph to reduce cost. You may notice sentiment values stop at a certain point in your historic data for this reason.
Where to Find It
The Sentiment Value column can be found in the messages table, though it is often hidden in report view to prevent confusion when editing visualizations.
How is a Messages Sentiment Determined
For details on how these values are calculated please see this article from Microsoft:
“The Text Analytics API uses a machine learning classification algorithm to generate a sentiment score between 0 and 1. Scores closer to 1 indicate positive sentiment, while scores closer to 0 indicate negative sentiment. Sentiment analysis is performed on the entire document, instead of individual entities in the text. This means sentiment scores are returned at a document or sentence level.
The model used is pre-trained with an extensive corpus of text and sentiment associations. It utilizes a combination of techniques for analysis, including text processing, part-of-speech analysis, word placement, and word associations. For more information about the algorithm, see Introducing Text Analytics. Currently, it isn’t possible to provide your own training data.“
There’s a tendency for scoring accuracy to improve when documents contain fewer sentences rather than a large block of text. During an objectivity assessment phase, the model determines whether a document as a whole is objective or contains sentiment. A document that’s mostly objective doesn’t progress to the sentiment detection phase, which results in a 0.50 score, with no further processing. For documents that continue in the pipeline, the next phase generates a score above or below 0.50. The score depends on the degree of sentiment detected in the document.”