In JKLC bot, for phase 1 some utterances are trained in multilingual model (bot languages : English, Hindi and Gujarati). For phase 2, when we are training the bot with some set of utterances and hit the train button, many utterances are failing.
We would like to understand for new utterances, why bot is not able to understand.
I’ve added 20 new utterances in English language and 10 are failing from that.
Generated a report and highlighted in blue for reference. Can someone please help us understand as we are facing issues and this is a blocker.
Sharing the utterances report for reference:
We also noticed when we train the bot to sentence encoder and train it back in multilingual the confidence gets increased for those utterances by 90% but causing some issues on keywords.
CC ; @B_Madhuri_Vishwanath @Harshitha_MH @Kaustubh_Nawade