A Method for Chinese Word Similarity Based on Data Augmentation SBERT

With the remarkable performance of transformer-based deep learning models in various natural language processing (NLP) tasks in recent years, it has become possible to utilize these models to address Chinese word similarity problems. Among them, SBERT has achieved excellent results in sentence similarity, and we can use the method of processing sentences to process Chinese word pairs. This paper proposes a Chinese word similarity calculation method based on SBERT. We introduce a data augmentation technique leveraging SBERT, aimed at addressing the challenge of generating additional labeled data from a limited annotated dataset. This approach enhances the accuracy of similarity calculations. Our novel method demonstrates a substantial performance improvement compared to the original approach.