Dynamic Topic Segmentation in Dialogues: Enhancing Boundaries with Topic-Aware Propagation

Dialogue Topic Segmentation (DTS) forms the foundation for various dialogue modeling tasks.However, existing methods do not fully leverage crucial dialogue information, such as positional cues and inter-topic associations.Additionally, these segmentation strategies lack flexibility, as they typically rely on a one-time global semantic comparison to determine segmentation boundaries without accounting for the dynamic influence of topic shifts on semantics.To address these limitations, we propose an innovative dynamic Dialogue Topic Segmentation (DyDTS) framework that employs a dynamic boundary optimization strategy based on topic-aware propagation matching, which enhances boundary refinement.Experimental results on two public datasets show the proposed method outperforms mainstream baselines.The source code and datasets used in this study are available at:https://github.com/Mark131434/DyDTS.