491 / 2024-04-25 16:37:06

A Patient Symptom Description Matching Model Combining Medical Knowledge Graph and Semantic Enhancement

Semantic enhancement,Patient Symptom Description,Online medical service platform,The Medical Knowledge Graph

With the improvement of residents 'awareness of health management, online medical service platforms are widely concerned and have been developed rapidly. However, in the process of searching for a suitable doctor according to their symptoms on the platform, patients' descriptions of symptoms tend to be colloquial, and patients are not professional enough in medicine and do not understand the symptoms of different diseases, so they often cannot find the right doctor, which reduces the satisfaction of patients using the platform. Therefore, the patient's colloquial description of symptoms and the professional description of symptoms need to correspond and match.

In this paper, the patient's description text data is taken as the corpus set, Chinese word segmentation is performed by Jieba tool, and the Word2Vec model is called to vectorize the symptom words of the entire corpus set. Then, the symptom knowledge in the knowledge graph is introduced, and the cosine similarity and the TF-IDF weight based on probability are calculated to realize the correspondence between patient description and professional symptoms. In particular, considering the problems of inaccurate description and insufficient semantic information contained in patients' colloquized symptom descriptions, the texts below the set number of symptom words in the symptom texts are taken as semantic enhancement objects, the importance and correlation of symptom words are used to realize the semantic enhancement of patient description texts, and the semantic enhancement matching model of patient symptoms based on knowledge graph is established. Through data crawler, 26625 patient question and answer records are collected from well-known online medical service platforms in China. Experimental results show that our model outperforms the baseline model and the current main matching algorithms in terms of accuracy and recall rate, which provides scientific guidance for patients 'symptom description.