BERTopic: Neural topic modeling with a class-based TF-IDF procedure

TL;DR

BERTopic introduces a neural topic modeling approach that combines transformer-based embeddings, clustering, and a class-based TF-IDF method to produce coherent and competitive topics from document collections.

Contribution

It extends clustering-based topic modeling with a novel class-based TF-IDF, improving topic coherence and performance.

Findings

Produces coherent topics effectively

Remains competitive across multiple benchmarks

Combines transformer embeddings with clustering and TF-IDF

Abstract

Topic models can be useful tools to discover latent topics in collections of documents. Recent studies have shown the feasibility of approach topic modeling as a clustering task. We present BERTopic, a topic model that extends this process by extracting coherent topic representation through the development of a class-based variation of TF-IDF. More specifically, BERTopic generates document embedding with pre-trained transformer-based language models, clusters these embeddings, and finally, generates topic representations with the class-based TF-IDF procedure. BERTopic generates coherent topics and remains competitive across a variety of benchmarks involving classical models and those that follow the more recent clustering approach of topic modeling.