The information retrieval system helps users find relevant information from large knowledge bases according to their search requests. With the massive growth of medical and health information, the problem of medical and health information retrieval has attracted the attention of the industry and academia. In order to improve the performance of medical and health information retrieval, a medical and health information retrieval method based on natural language processing is proposed. Based on the Chinese word segmentation model framework of dictionary enhanced graph convolution network, the medical and health information retrieval language word segmentation processing is implemented. The TTSimhash algorithm in natural language processing technology is used to calculate the word similarity and find the extended words of conditional keywords. Design a relevance retrieval method of query documents based on BERT multi-level information structure to achieve medical and health information retrieval. The test results show that the design method MAP@k is higher.