Research map / Label-efficient and robust learning / Recognition and long-tail
CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation
Rui Ke, Jiahui Xu, Shenghao Yang, Kuang Wang, Feng Jiang, Haizhou Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add preference-enhanced clustering (PeC) (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 48.8 -> 55.8 | Improved |
| add PeC (ablation) | CATCH · DSTC-12 Finance | LLM score and ROUGE-2: LLM 98.4 -> 97.3; Rouge-2 26.7 -> 24.5 | Mixed |
| add topic segmentation (TopSeg) context representation (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 41.6 -> 55.8 | Improved |
| add hierarchical theme generation (HieGen) instead of flat (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 36.8 -> 55.8 | Improved |
| add hierarchical generation instead of flat (ablation) | CATCH · DSTC-12 Finance | ROUGE-L: 19.6 -> 42.4 | Improved |
| add utterance-level adaptation of the topic segmenter (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 50.6 -> 55.8 | Improved |
| add conversation-level adaptation of segmenter (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 44.3 -> 50.6 | Improved |
| replace spectral clustering with ClusterLLM (ablation) | CATCH · DSTC-12 Finance | clustering accuracy: 55.8 -> 70.1 | Improved |
| replace LLaMA3-8B with GPT-4o for label generation (ablation) | CATCH · DSTC-12 Finance | CosSim: 59.3 -> 61.1 | Improved |
| replace LLaMA3-8B with Qwen-3-8B for labels (ablation) | CATCH · DSTC-12 Finance | CosSim: 59.3 -> 59.3 | No clear effect |
| oracle clustering (PeC=1) vs real clustering (ablation) | CATCH · DSTC-12 Finance | ROUGE-1: 42.4 -> 51.6 | Improved |
| CATCH vs GRM preference-reward baseline | DSTC-12 Banking · DSTC-12 Banking | clustering accuracy: 46.9 -> 56.7 | Improved |
Closest papers by meaning
- Topic-Aware Multi-turn Dialogue Modeling (AAAI 2021 · similarity 0.84)
- Personalized Visual Content Generation in Conversational Systems (NeurIPS 2025 · similarity 0.83)
- Sentiment Classification in Customer Service Dialogue with Topic-Aware Multi-Task Learning (AAAI 2020 · similarity 0.83)
- A Controllable Model of Grounded Response Generation (AAAI 2021 · similarity 0.83)
- Topic-Oriented Spoken Dialogue Summarization for Customer Service with Saliency-Aware Topic Modeling (AAAI 2021 · similarity 0.83)
- Aligning Spoken Dialogue Models from User Interactions (ICML 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
