Open thaynos opened 3 weeks ago
I suggested this in https://github.com/mlomb/chat-analytics/issues/25, but it's sadly not possible because it would require storing the raw messages in the database which would be far too large. Also, the model isn't trained specifically for chat-analytics (chat-analytics uses multilang-sentiment) so chat-analytics isn't really able to "refine the model" (might be theoretically possible, but it would be futile and out-of-scope).
it would be very helpful if we could see which messages were tagged as negative or positive as that could give us a deeper insight into the accuracy. perhaps maybe add a accurate or not button to further help refine the model?