Both, because the way it's right and wrong are different.

Sentiment analysis might misclassify some of the data, but it doesn't risk making things up wholescale like an LLM would.