Comprehensive Evaluation of Large Language Models for Topic Modeling

TL;DR

This paper quantitatively evaluates large language models for topic modeling, assessing their topic quality, hallucination tendencies, and controllability, revealing strengths in coherence but limitations in focus and control.

Contribution

It provides a comprehensive quantitative analysis of LLMs in topic modeling, addressing gaps in prior qualitative evaluations.

Findings

LLMs produce coherent and diverse topics with few hallucinations.

They tend to take shortcuts by focusing on parts of documents.

Controllability of topics via prompts is limited.

Abstract

Recent work utilizes Large Language Models (LLMs) for topic modeling, generating comprehensible topic labels for given documents. However, their performance has mainly been evaluated qualitatively, and there remains room for quantitative investigation of their capabilities. In this paper, we quantitatively evaluate LLMs from multiple perspectives: the quality of topics, the impact of LLM-specific concerns, such as hallucination and shortcuts for limited documents, and LLMs' controllability of topic categories via prompts. Our findings show that LLMs can identify coherent and diverse topics with few hallucinations but may take shortcuts by focusing only on parts of documents. We also found that their controllability is limited.