Best practices for sentiment analysis of chat transcripts

Hi, slight beginner here. I’m looking for best practice for doing sentiment analysis on chat transcripts. Any and all advice would be appreciated, but specific items I’m unsure about:

  • If it is between two people, one being the customer support and the other being the customer, do I need to classify who is who or can I just feed the whole transcript to a classifier pipeline?

  • How do I know which model to use?

  • What other things can be done besides sentiment analysis? Still exploring what to do with all this raw transcript data :thinking: