search

shobrook / TypeSense

Chrome extension that analyzes a Messenger conversation's sentiment in real-time
MIT License
7 stars 2 forks source link

Contextual Sentiment Analysis Strategy? #2

Closed alichtman closed 6 years ago

alichtman commented 6 years ago

Do we want to analyze the sentiment of each message individually (out of context) or in sets of 3-4 messages (in context)?

ABABA? ABA? Sets of 3/4/5 messages, regardless of who sent them? What if all of the last messages are all sent by one person?

Azure forces sentiment analysis of only one string at a time, so we can't do contextualized sentiment scores unless we join them into one string.

Opinions on what we should do?

alichtman commented 6 years ago

http://www.aclweb.org/anthology/W15-1829

This paper seems to have done them all individually so I might do that

shobrook commented 6 years ago

We could concatenate previous messages into a single string, call it ABA, and get its sentiment score. And then concatenate the most recent message to the string, so it's now ABAB, and get the updated sentiment score. If the updated sentiment score is greater than the previous, then the new message's sentiment is positive, and negative for when the sentiment score is less than the previous.