An Unsupervised Sentence Embedding Method by Mutual Information Maximization
This work addresses the problem of generating effective sentence embeddings without labeled data for researchers and practitioners in NLP, offering a domain-agnostic solution, though it is incremental as it builds on existing BERT and SBERT frameworks.
The paper tackles the inefficiency of BERT for sentence-pair tasks and the reliance of Sentence BERT on labeled data by proposing an unsupervised method using mutual information maximization to derive sentence embeddings, achieving significant performance gains over unsupervised baselines and competitive results with supervised methods in label-scarce settings.
BERT is inefficient for sentence-pair tasks such as clustering or semantic search as it needs to evaluate combinatorially many sentence pairs which is very time-consuming. Sentence BERT (SBERT) attempted to solve this challenge by learning semantically meaningful representations of single sentences, such that similarity comparison can be easily accessed. However, SBERT is trained on corpus with high-quality labeled sentence pairs, which limits its application to tasks where labeled data is extremely scarce. In this paper, we propose a lightweight extension on top of BERT and a novel self-supervised learning objective based on mutual information maximization strategies to derive meaningful sentence embeddings in an unsupervised manner. Unlike SBERT, our method is not restricted by the availability of labeled data, such that it can be applied on different domain-specific corpus. Experimental results show that the proposed method significantly outperforms other unsupervised sentence embedding baselines on common semantic textual similarity (STS) tasks and downstream supervised tasks. It also outperforms SBERT in a setting where in-domain labeled data is not available, and achieves performance competitive with supervised methods on various tasks.