Identification of intent-sentiment co-reference from conversational texts

Abstract

Our goal in this thesis work is to develop models based on judicial dialogues that can recognise the feelings of the utterances, the intent or intents of the speakers, the tokens that determine them, and any relationships between them. One issue with a chatbot designed primarily for judicial help is that user intentions can become muddled and overlap. Therefore, judging intents solely from the words said by the user is difficult. The wording and context completely alter the intentions. In order to correctly serve people, it’s crucial for a chatbot or agent to keep track of their feelings during a conversation. When a dialogue veers off into an intense emotion, it loses all effectiveness. Therefore, it is a significant problem for the chatbot to understand which topics can cause those divisive feelings.