Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DASH: Dialogue-Aware Similarity and Handshake Recognition for Topic Segmentation in Public-Channel Conversations

Sijin Sun, Liangbin Zhao, Ming Deng, Xiuju Fu

TL;DR

DASH-DTS tackles the challenging problem of topic segmentation in public-channel dialogues by integrating handshake recognition, similarity-guided in-context learning with carefully selected exemplars, and context-aware topic labeling. It adds interpretability through segment-level explanations and confidence scores and contributes a first public maritime DTS dataset, VHF-Dial. Empirical results show state-of-the-art or competitive performance across standard benchmarks and particularly strong gains on VHF-Dial, with ablations validating the contribution of each component. The framework enables more reliable monitoring and decision support in safety-critical communications and sets the stage for extending to other domains like air traffic control and emergency dispatch.

Abstract

Dialogue Topic Segmentation (DTS) is crucial for understanding task-oriented public-channel communications, such as maritime VHF dialogues, which feature informal speech and implicit transitions. To address the limitations of traditional methods, we propose DASH-DTS, a novel LLM-based framework. Its core contributions are: (1) topic shift detection via dialogue handshake recognition; (2) contextual enhancement through similarity-guided example selection; and (3) the generation of selective positive and negative samples to improve model discrimination and robustness. Additionally, we release VHF-Dial, the first public dataset of real-world maritime VHF communications, to advance research in this domain. DASH-DTS provides interpretable reasoning and confidence scores for each segment. Experimental results demonstrate that our framework achieves several sota segmentation trusted accuracy on both VHF-Dial and standard benchmarks, establishing a strong foundation for stable monitoring and decision support in operational dialogues.

DASH: Dialogue-Aware Similarity and Handshake Recognition for Topic Segmentation in Public-Channel Conversations

TL;DR

DASH-DTS tackles the challenging problem of topic segmentation in public-channel dialogues by integrating handshake recognition, similarity-guided in-context learning with carefully selected exemplars, and context-aware topic labeling. It adds interpretability through segment-level explanations and confidence scores and contributes a first public maritime DTS dataset, VHF-Dial. Empirical results show state-of-the-art or competitive performance across standard benchmarks and particularly strong gains on VHF-Dial, with ablations validating the contribution of each component. The framework enables more reliable monitoring and decision support in safety-critical communications and sets the stage for extending to other domains like air traffic control and emergency dispatch.

Abstract

Dialogue Topic Segmentation (DTS) is crucial for understanding task-oriented public-channel communications, such as maritime VHF dialogues, which feature informal speech and implicit transitions. To address the limitations of traditional methods, we propose DASH-DTS, a novel LLM-based framework. Its core contributions are: (1) topic shift detection via dialogue handshake recognition; (2) contextual enhancement through similarity-guided example selection; and (3) the generation of selective positive and negative samples to improve model discrimination and robustness. Additionally, we release VHF-Dial, the first public dataset of real-world maritime VHF communications, to advance research in this domain. DASH-DTS provides interpretable reasoning and confidence scores for each segment. Experimental results demonstrate that our framework achieves several sota segmentation trusted accuracy on both VHF-Dial and standard benchmarks, establishing a strong foundation for stable monitoring and decision support in operational dialogues.

Paper Structure

This paper contains 28 sections, 10 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Early Methods
  4. Sequence Modeling Methods
  5. Pre-trained Models
  6. Large Language Models
  7. Methodology
  8. Dialogue-level Similarity
  9. Semantic Similarity Calculation
  10. Exemplar Selection
  11. In-Context Learning with Selected Exemplars
  12. Handshake Statement Tag
  13. Content-Aware Topic Generation
  14. Contextual Input
  15. Posi/Nega-tive Generation
  16. ...and 13 more sections

Figures (2)

  • Figure 1: Dynamically select the most semantically relevant exemplars for each input conversation, enhancing the accuracy of topic segmentation in contextual learning for large models.
  • Figure 2: Workflow of the Handshake Statement Tagging Component Based on Few-shot LLM Learning.