MC-TCMNER: A Multi-modal Fusion Model Combining Contrast Learning Method for Traditional Chinese Medicine NER

Traditional Chinese Medicine (TCM) texts contain a wealth of knowledge accumulated over thousands of years, making the extraction of knowledge from these texts a pivotal concern. Named Entity Recognition (NER) can serve as an effective tool for extracting knowledge information from TCM texts. However, TCM texts contain a large number of rare characters and homophones, and the attributions of entities are also more complex, making TCMNER more challenging. In order to address this issue, this paper introduces MC-TCMNER, a novel method that leverages the multi-modal features of Chinese characters and incorporates a training strategy based on contrastive learning. Experiments have shown that our proposed method achieves an F1 score of 94.05% on the TCMNER dataset and 52.84% on the C-CLUE benchmark, demonstrating the effectiveness of MC-TCMNER. Furthermore, owing to the limited availability of a comprehensive dataset for TCMNER, we have taken the initiative to publicly release a TCMNER dataset that we meticulously collected and annotated.