KDSTM: Neural Semi-supervised Topic Modeling with Knowledge Distillation

TL;DR

KDSTM is a semi-supervised topic modeling approach that leverages knowledge distillation, requiring minimal labeled data and no pretraining, and it outperforms existing methods in accuracy, robustness, and efficiency.

Contribution

Introduces KDSTM, a novel semi-supervised topic modeling method that does not need pretrained embeddings and is effective for resource-constrained text classification.

Findings

Outperforms existing supervised topic models in accuracy and robustness

Achieves similar performance to state-of-the-art weakly supervised methods

Requires fewer labeled documents and no pretraining

Abstract

In text classification tasks, fine tuning pretrained language models like BERT and GPT-3 yields competitive accuracy; however, both methods require pretraining on large text datasets. In contrast, general topic modeling methods possess the advantage of analyzing documents to extract meaningful patterns of words without the need of pretraining. To leverage topic modeling's unsupervised insights extraction on text classification tasks, we develop the Knowledge Distillation Semi-supervised Topic Modeling (KDSTM). KDSTM requires no pretrained embeddings, few labeled documents and is efficient to train, making it ideal under resource constrained settings. Across a variety of datasets, our method outperforms existing supervised topic modeling methods in classification accuracy, robustness and efficiency and achieves similar performance compare to state of the art weakly supervised text classification methods.