CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation
About
Theme detection is a fundamental task in user-centric dialogue systems, aiming to identify the latent topic of each utterance without relying on predefined schemas. Unlike intent induction, which operates within fixed label spaces, theme detection requires cross-dialogue consistency and alignment with personalized user preferences, posing significant challenges. Existing methods often struggle with sparse, short utterances for accurate topic representation and fail to capture user-level thematic preferences across dialogues. To address these challenges, we propose CATCH (Controllable Theme Detection with Contextualized Clustering and Hierarchical Generation), a unified framework that integrates three core components: (1) context-aware topic representation, which enriches utterance-level semantics using surrounding topic segments; (2) preference-guided topic clustering, which jointly models semantic proximity and personalized feedback to align themes across dialogue; and (3) a hierarchical theme generation mechanism designed to suppress noise and produce robust, coherent topic labels. Experiments on a multi-domain customer dialogue benchmark (DSTC-12) demonstrate the effectiveness of CATCH with 8B LLM in both theme clustering and topic generation quality.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Theme Detection | DSTC Travel domain 12 (test) | Semantic Relevance (SR)86.2 | 6 | |
| Conversation-level Topic Discovery and Labeling | DSTC-12 Travel domain (blind test) | Accuracy0.358 | 6 | |
| Theme Distribution | Finance Out-of-domain | Accuracy55.8 | 4 | |
| Theme Distribution | Insurance Out-of-domain | Acc54.5 | 4 | |
| Theme Distribution | BANKING | Acc0.567 | 4 | |
| Theme Label Quality | Finance Out-of-domain | ROUGE-142.4 | 4 | |
| Theme Label Quality | Insurance Out-of-domain | ROUGE-141.8 | 4 | |
| Theme Label Quality | BANKING | ROUGE-135.3 | 4 |