Keeping up with internal conversations and understanding what's truly being said on your platform can feel like an impossible task. When it's tough to track sentiment and identify those important discussions, you might overlook signs of employee concern or engagement.

That's where our Sentiment Analysis steps in to help! By giving you insight into the emotions behind the content, you can easily understand employee sentiments, address negative issues, and make thoughtful decisions that truly reflect your workforce's needs.

How does it work?

For our analysis, we consider every timeline post, blog article, wiki article, and multichannel studio post on your platform to figure out its overall sentiment. Here's how we do it:

What we look at

For each content item, we consider:

• Comments: The comments of each content item. The sentiment of comments can range from positive to neutral to negative.
• Reactions: The reactions of each content item. The sentiment of reactions can either be positive or negative. Reactions to comments on the content item are not counted.
• The Timeline post itself: The content in timeline posts. The sentiment of timeline post content can range from positive to neutral to negative.

Content from blog and wiki articles and studio posts are excluded due to their generally neutral sentiment. Consequently, the sentiment for blog and wiki articles and studio posts without comments or reactions is labeled as N/A.

How we calculate the sentiment score

Once we've determined the sentiment of each comment, reaction, and post content, we combine them to create an overall sentiment score. We consider it in this order:

1. Timeline post content matters most. The feeling expressed in the main post carries the most weight.
2. Comments are next in line. What people say in their comments is very important.
3. Reactions are considered last, but even a simple "like" or "dislike" adds to the picture.

There's an important exception: If a timeline post is clearly negative in its main content, we don't factor in any comments or reactions that might seem positive. Why? Because sometimes, a negative post gets "positive" reactions (like "likes") that actually support the original negative idea, rather than disagreeing with it.

AI algorithms can struggle to detect the subtleties of human communication, such as sarcasm, humor, and irony. Therefore, it's essential not to expect 100% accuracy.

• 

  Calculations:

  • W post = Weight of a post. Default = 4
  • W comments = Weight of comments. Default = 2
  • W reactions = Weight of reactions. Default = 1
  • S post = Sentiment of post content. Range: (-1;0;1). Wiki and blog default = 0
  • S comments = Sentiment of a comment. Range: (-1;0;1)
  • S reactions = Sentiment of a reaction. Range: (-1;1):
    • -1 for negative reactions (Dislike; Hmm...)
    • 1 for positive reactions (Like, Insightful, Funny, Great, Applause, Heart)
  • N comments = Number of comments
  • S reactions = Number of reactions

  Scores per label:

  • "Positive": >= 0.4
  • "Neutral": -0.39 - 0.39
  • "Negative": <= -0.4

The final sentiment labels

After all the calculations, we assign a sentiment label to the post:

• Positive
• Neutral
• Negative

As stated above, the sentiment for blog and wiki articles without comments or reactions is labeled as N/A (Not applicable).

How is it presented?

The sentiment for each piece of content is displayed on its Content Details page within the Analytics. You'll see a dedicated Sentiment metric for the content item, and a donut chart providing a quick visual summary of comments and reactions in the selected time period. You can use filters to refine the sentiment results by date range or company, department, or location. 

Additionally, the Sentiment is directly available on the Content Performance table on the main Content dashboard, making it easy to find content that contributes to a specific sentiment.

Learn more in our article on the Content dashboard: Content Analytics.