A Multi-granularity Text Matching Method Based on Chinese Linguistic Features

Text semantic matching is a fundamental task in natural language processing, which holds significant meaning and presents challenges for a series of subsequent tasks. Due to the multi-dimensional complexity of semantic expression in Chinese, existing text semantic matching methods tend to focus more on learning features from single granularities, such as characters and words, while neglecting the unique characteristics of Chinese characters, such as pinyin, radicals, and glyphs. These features contain rich semantic information. Therefore, this paper proposes a multi-granularity text matching method based on Chinese linguistic features, investigating the impact of integrating five granularities—character, word, pinyin, radicals, and glyphs—on semantic matching. Extensive experiments on the public datasets BQ and LCQMC demonstrate that the proposed model can effectively improve the performance of semantic matching compared to popular non-BERT models. Specifically, on the BQ dataset, the accuracy and F1 score improved by 1.1% and 1.06%, respectively.