LITA: An Efficient LLM-assisted Iterative Topic Augmentation Framework

TL;DR

The paper addresses domain-specific topic modeling where conventional methods struggle with coherence and specificity. It introduces LITA, an LLM-assisted iterative topic augmentation framework that seeds topics with user-provided words, embeds documents, clusters them, and selectively reassigns ambiguous documents via an LLM to dynamically augment topics. Key contributions include cost-efficient LLM integration by focusing on ambiguous instances, iterative refinement with agglomerative clustering to identify new topics, and empirical validation showing improvements over $K$-topic baselines on two datasets across multiple topic-quality and clustering metrics. The approach offers a scalable, adaptable workflow for high-quality topic clustering in domain-specific corpora, with practical potential for real-world analytics and deployment.

Abstract

Topic modeling is widely used for uncovering thematic structures within text corpora, yet traditional models often struggle with specificity and coherence in domain-focused applications. Guided approaches, such as SeededLDA and CorEx, incorporate user-provided seed words to improve relevance but remain labor-intensive and static. Large language models (LLMs) offer potential for dynamic topic refinement and discovery, yet their application often incurs high API costs. To address these challenges, we propose the LLM-assisted Iterative Topic Augmentation framework (LITA), an LLM-assisted approach that integrates user-provided seeds with embedding-based clustering and iterative refinement. LITA identifies a small number of ambiguous documents and employs an LLM to reassign them to existing or new topics, minimizing API costs while enhancing topic quality. Experiments on two datasets across topic quality and clustering performance metrics demonstrate that LITA outperforms five baseline models, including LDA, SeededLDA, CorEx, BERTopic, and PromptTopic. Our work offers an efficient and adaptable framework for advancing topic modeling and text clustering.

Figures (3)

• Figure 1: The Illustration of LITA
• Figure 2: Number of Ambiguous Instances in each Iteration
• Figure 3: Sensitivity Analysis: we report the impact on the ambiguous distance threshold ($\epsilon$) and agglomerative distance threshold ($\gamma$) across both datasets.