Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Industry-Aligned Granular Topic Modeling

Sae Young Moon, Myeongjun Erik Jang, Haoyan Luo, Chunyang Xiao, Antonios Georgiadis, Fran Silavong

TL;DR

TIDE introduces an industry-aligned granular topic modeling framework that leverages large language models to generate fine-grained topics while providing business-friendly utilities such as targeted summarization, topic hierarchy detection, distillation for efficient classification, and an automatic topic evaluation tool. The cluster-then-generate approach reduces LLM usage to $O(N)$ and enables seamless topic assignment, with optional business definitions guiding topic generation. Across public and real-world datasets, TIDE demonstrates robust, granular topic discovery, outperforming several strong baselines and delivering practical benefits for business analytics, including improved topic completeness and readability. Limitations include LLM context-length constraints and potential impacts on distillation performance; future work will address these limits and further integrate business-specific prompts and evaluation.

Abstract

Topic modeling has extensive applications in text mining and data analysis across various industrial sectors. Although the concept of granularity holds significant value for business applications by providing deeper insights, the capability of topic modeling methods to produce granular topics has not been thoroughly explored. In this context, this paper introduces a framework called TIDE, which primarily provides a novel granular topic modeling method based on large language models (LLMs) as a core feature, along with other useful functionalities for business applications, such as summarizing long documents, topic parenting, and distillation. Through extensive experiments on a variety of public and real-world business datasets, we demonstrate that TIDE's topic modeling approach outperforms modern topic modeling methods, and our auxiliary components provide valuable support for dealing with industrial business scenarios. The TIDE framework is currently undergoing the process of being open sourced.

Industry-Aligned Granular Topic Modeling

TL;DR

TIDE introduces an industry-aligned granular topic modeling framework that leverages large language models to generate fine-grained topics while providing business-friendly utilities such as targeted summarization, topic hierarchy detection, distillation for efficient classification, and an automatic topic evaluation tool. The cluster-then-generate approach reduces LLM usage to and enables seamless topic assignment, with optional business definitions guiding topic generation. Across public and real-world datasets, TIDE demonstrates robust, granular topic discovery, outperforming several strong baselines and delivering practical benefits for business analytics, including improved topic completeness and readability. Limitations include LLM context-length constraints and potential impacts on distillation performance; future work will address these limits and further integrate business-specific prompts and evaluation.

Abstract

Topic modeling has extensive applications in text mining and data analysis across various industrial sectors. Although the concept of granularity holds significant value for business applications by providing deeper insights, the capability of topic modeling methods to produce granular topics has not been thoroughly explored. In this context, this paper introduces a framework called TIDE, which primarily provides a novel granular topic modeling method based on large language models (LLMs) as a core feature, along with other useful functionalities for business applications, such as summarizing long documents, topic parenting, and distillation. Through extensive experiments on a variety of public and real-world business datasets, we demonstrate that TIDE's topic modeling approach outperforms modern topic modeling methods, and our auxiliary components provide valuable support for dealing with industrial business scenarios. The TIDE framework is currently undergoing the process of being open sourced.
Paper Structure (55 sections, 5 equations, 2 figures, 7 tables)

This paper contains 55 sections, 5 equations, 2 figures, 7 tables.

Table of Contents

  1. Introduction
  2. TIDE Framework
  3. Summarization
  4. Granular Topic Modeling
  5. Topic Hierarchy Detection
  6. Distillation
  7. Automatic Evaluation Tool
  8. Label accuracy.
  9. Topic accuracy/completeness.
  10. Experiments on Public Datasets
  11. Experiment Settings.
  12. Datasets.
  13. Evaluation metrics.
  14. Baseline models.
  15. Experimental Results
  16. ...and 40 more sections

Figures (2)

  • Figure 1: TIDE architecture diagram
  • Figure 2: TIDE in-depth architecture diagram