From Documents to Segments: A Contextual Reformulation for Topic Assignment

In three linesNovel topic modeling approach (SBTA) assigns topics to text segments rather than entire documents, reducing topic contamination. Authors construct SemEval-STM, a dataset annotated via LLM + human refinement, and validate improved clustering quality and interpretability across multiple models.

Summary generated by Claude — human-verified