Topic Segmentation Using Generative Language Models

TL;DR

This paper explores using generative large language models for topic segmentation, proposing a recursive prompting strategy and boundary similarity metrics, showing they can outperform existing methods but still face challenges.

Contribution

It introduces a novel overlapping and recursive prompting approach for topic segmentation with LLMs and evaluates their effectiveness using boundary similarity metrics.

Findings

LLMs can outperform traditional segmentation methods

The proposed prompting strategy improves segmentation quality

Challenges remain in reliability and consistency of LLM-based segmentation

Abstract

Topic segmentation using generative Large Language Models (LLMs) remains relatively unexplored. Previous methods use semantic similarity between sentences, but such models lack the long range dependencies and vast knowledge found in LLMs. In this work, we propose an overlapping and recursive prompting strategy using sentence enumeration. We also support the adoption of the boundary similarity evaluation metric. Results show that LLMs can be more effective segmenters than existing methods, but issues remain to be solved before they can be relied upon for topic segmentation.