BERTopic generates document embedding with pre-trained transformer-based language models, clusters these embeddings, and finally, generates topic representations with the class-based TF-IDF procedure.
https://ritvik19.medium.com/papers-explained-193-bertopic-f9aec10cd5a6