# CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation

Source: [CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation](<https://www.sota2.com/research/paper/69520ad653fdd5c36173d2dd>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-12-30T05:00:04.959Z

Authors: Rui Ke, Jiahui Xu, Shenghao Yang, Kuang Wang, Feng Jiang, Haizhou Li

Published: 2025-12-25T15:33:25.000Z

## Abstract

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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Theme Distribution on Finance Out-of-domain](<https://www.sota2.com/research/sota/theme-distribution-on-finance-out-of-domain.md>): Theme Distribution; Finance Out-of-domain; Accuracy: 55.8.

- [Theme Distribution on Insurance Out-of-domain](<https://www.sota2.com/research/sota/theme-distribution-on-insurance-out-of-domain.md>): Theme Distribution; Insurance Out-of-domain; Acc: 54.5.

- [Theme Distribution on Banking](<https://www.sota2.com/research/sota/theme-distribution-on-banking.md>): Theme Distribution; BANKING; Acc: 0.567.

- [Theme Label Quality on Finance Out-of-domain](<https://www.sota2.com/research/sota/theme-label-quality-on-finance-out-of-domain.md>): Theme Label Quality; Finance Out-of-domain; ROUGE-1: 42.4.

- [Theme Label Quality on Insurance Out-of-domain](<https://www.sota2.com/research/sota/theme-label-quality-on-insurance-out-of-domain.md>): Theme Label Quality; Insurance Out-of-domain; ROUGE-1: 41.8.

- [Theme Label Quality on Banking](<https://www.sota2.com/research/sota/theme-label-quality-on-banking.md>): Theme Label Quality; BANKING; ROUGE-1: 35.3.

- [Theme Detection on DSTC Travel domain 12 (test)](<https://www.sota2.com/research/sota/theme-detection-on-dstc-travel-domain-12-test.md>): Theme Detection; DSTC Travel domain 12 (test); Semantic Relevance (SR): 86.2.

- [Conversation-level Topic Discovery and Labeling on DSTC-12 Travel domain (blind test)](<https://www.sota2.com/research/sota/conversation-level-topic-discovery-and-labeling-on-dstc-12-travel-domain-blind-test.md>): Conversation-level Topic Discovery and Labeling; DSTC-12 Travel domain (blind test); Accuracy: 0.358.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2512.21715>)

[PDF](<https://arxiv.org/pdf/2512.21715.pdf>)