Sentiment analysis on the conversation you were having?

Among other things, sure. More simply, keyword analysis.

Remember semantically aware models are still fairly new and even they lack the context for a particular field of text.

All of these “models” are useless garbage but it doesn’t stop them from trying to absolutely cram them everywhere they can.

Unnecessary

None of what they do is “necessary”. They could just ask you what your relevant interests are and you could tell them, but they do it anyway. They go to great lengths for any seemingly insignificant amount of data they can get their hands on.