Comment Sentiment

  • 5 January 2023
  • 2 replies

I don’t understand how sentiment in the comment report is calculated. Is the sentiment based on the entire text of the comment for each open-ended question or each individual keyword or phrase found within a commenter’s response? For example, for a single question, there are only 3 commenters who wrote a single sentence each, but the % positive sentiment is 87% and the % negative sentiment is 13%. This calculation makes it seem like one sentence from a single commenter may have a stronger influence on the sentiment values than other commenters when in reality the commenter may only be using two redundant positive words to explain.  Is there anyway to categorize sentiment based on the full text of the comment?

2 replies

Userlevel 6

hi @mraney 

No one comment/sentence has more weight than another. Each sentence is reduced down to (if you will) key words and key phrases. Our algorithm cannot distinguish key takeaways from full sentences.

Hope this helps.

Happy New Year!

Thanks Judy. So if I am understanding you correctly, it is each keyword and phrase in a response that is counted toward the numerator and denominator totals when calculating sentiment percentages? Meaning that if commenter #1 said something like: “The manager is excellent, a strong leader. I appreciate the flexibility and good communication provided.” and commenter #2 said something like: “The manager is excellent. However, sometimes she is a poor communicator.” the positive sentiment percentage would be [1(excellent-C1) + 1(strong-C1) + 1(flexibility-C1) + 1(good-C1) + 1(excellent-C2)]/[1(excellent-C1) + 1(strong-C1)+ 1(flexibility-C1) + 1(good-C1) + 1(excellent-C2) + 1(poor-C2)] = 83%. Doesn’t this mean that because commenter 1 stated more “positive” keywords that their response is weighted more heavily in the sentiment percentage calculation than commenter 2 and ultimately a handful of survey respondents may actually be driving the sentiment percentages? In the example I provided, excellent and strong are in the same sentence and are driving up the numerator even thought the commenter is simply defining the first positive sentiment rather than expressing a separate positive sentiment. Might this prevent a representative analysis of employees who provided comments?  I appreciate your time!