Improving Contrastive Learning of Sentence Embeddings from AI Feedback
This work addresses a bottleneck in natural language processing for researchers and practitioners by improving sentence embedding quality through a hybrid approach, though it is incremental as it builds on existing contrastive learning methods.
The paper tackles the problem of generating high-quality sample pairs for contrastive learning of sentence embeddings by proposing CLAIF, which uses AI feedback from large language models to construct pairs with fine-grained similarity scores, achieving state-of-the-art performance on semantic textual similarity and transfer learning tasks.
Contrastive learning has become a popular approach in natural language processing, particularly for the learning of sentence embeddings. However, the discrete nature of natural language makes it difficult to ensure the quality of positive and negative sample pairs generated through data augmentation methods. Although supervised contrastive learning can produce more accurate sample pairs with human feedback labels, it still lacks fine-grained training signals. In this paper, we propose to improve \textbf{C}ontrastive \textbf{L}earning of sentence embeddings from \textbf{AI} \textbf{F}eedback \textbf{(CLAIF)}. Our method utilizes AI feedback from large pre-trained language models (LLMs) to construct sample pairs with fine-grained sample similarity scores to improve contrastive learning. Besides, we combine human feedback and AI feedback to provide better supervision signals for supervised contrastive learning of sentence embeddings. Experimental results show that our method achieves state-of-the-art performance on several semantic textual similarity (STS) and transfer learning tasks compared to other unsupervised and supervised contrastive learning methods.