MLLMIE: Multiple Large Language Model Information Extraction in Ancient Medical Texts

Ancient medical texts, which reflect traditional cultural practices, can play an important role in the promotion of medical traditions. Nevertheless, the complexity of extracting information from classical Chinese medical texts and the potential inconsistencies and hallucinations in large language models during information extraction present significant challenges. In domains such as medicine, where the precision of information extraction is of paramount importance, the paper augmented existing methodologies for information extraction from large language models and proposed a novel approach, designated as Multiple Large Language Model Information Extraction (MLLMIE). The method initially defined a pattern layer to explicitly delineate the concepts, terms, and their interrelationships within the text. Subsequently, it employed prompt engineering to extract information from multiple models, thereby obtaining candidate medical ancient text knowledge results. Finally, the model validation layer was employed to filter the candidate results, thereby obtaining the final medical ancient text knowledge. The method attains F1 scores of 92.68% for named entity recognition and 94.91% for relation extraction in ancient medical texts, thereby demonstrating high-quality medical ancient text information extraction. This method effectively enhances the accuracy of ancient medical text information extraction and facilitates the discovery and application of ancient medical text knowledge.