I don't think they trained an LLM. This looks like TFIDF on a bag of words to me, possibly getting informative features from a prediction model. Did something

I don't think they trained an LLM. This looks like TFIDF on a bag of words to me, possibly getting informative features from a prediction model. Did something similar on the same data at school many years back, but I used the category labels to try and predict good and bad trips. The model worked abysmally, but the features made for great word clouds.