Coherence-Aware Neural Topic Modeling

TL;DR

This paper introduces a neural topic modeling approach that integrates topic coherence directly into the training process, resulting in models with comparable perplexity but significantly improved topic coherence.

Contribution

It proposes a novel method to incorporate topic coherence into neural variational inference, enhancing the semantic quality of topics during training.

Findings

Models achieve similar perplexity to baselines

Significantly higher topic coherence scores

Coherence-aware training improves semantic quality of topics

Abstract

Topic models are evaluated based on their ability to describe documents well (i.e. low perplexity) and to produce topics that carry coherent semantic meaning. In topic modeling so far, perplexity is a direct optimization target. However, topic coherence, owing to its challenging computation, is not optimized for and is only evaluated after training. In this work, under a neural variational inference framework, we propose methods to incorporate a topic coherence objective into the training process. We demonstrate that such a coherence-aware topic model exhibits a similar level of perplexity as baseline models but achieves substantially higher topic coherence.