This idea is part of the A Dollar Worth of Ideas series, with potential open source, research or data science projects or contributions for people to pursue. I would be interested in mentoring some of them. Just contact me for details.
This idea piggy-backs on sentiment analysis and music generation. It can be done using rule-based music generation systems (such as MURAL) or with more deep learning approaches.
The system provides a feedback to the streamer of the sentiment in the chat by generating music reacting to the sentiment in that sentiment. It also addresses the issue of copyrighted music on streaming platforms.