Is Automated Topic Model Evaluation Broken?: The Incoherence of Coherence
This addresses a critical validation problem for researchers and practitioners in NLP and unsupervised learning, highlighting that current automated metrics are broken for neural topic models.
The paper tackles the problem of evaluating topic models by showing that automated coherence metrics, which favor neural models, do not align with human judgments from rating and intrusion tasks, revealing a validation gap. The result is that automated evaluations incorrectly declare winners when human evaluations do not, questioning their validity.
Topic model evaluation, like evaluation of other unsupervised methods, can be contentious. However, the field has coalesced around automated estimates of topic coherence, which rely on the frequency of word co-occurrences in a reference corpus. Contemporary neural topic models surpass classical ones according to these metrics. At the same time, topic model evaluation suffers from a validation gap: automated coherence, developed for classical models, has not been validated using human experimentation for neural models. In addition, a meta-analysis of topic modeling literature reveals a substantial standardization gap in automated topic modeling benchmarks. To address the validation gap, we compare automated coherence with the two most widely accepted human judgment tasks: topic rating and word intrusion. To address the standardization gap, we systematically evaluate a dominant classical model and two state-of-the-art neural models on two commonly used datasets. Automated evaluations declare a winning model when corresponding human evaluations do not, calling into question the validity of fully automatic evaluations independent of human judgments.