Research On Traditional Chinese Medicine Prescription Recommendation Algorithm Based On Graph Neural Network Multi-layer Information Fusion And Knowledge-driven

Traditional Chinese Medicine(TCM)prescription is a precious wealth accumulated in the long-term development of TCM,and it is also one of the inheritance methods of TCM.It is of great significance to deeply excavate the potential rules of TCM prescriptions for the actual clinical application of TCM and the discovery of new prescriptions.With the rise of artificial intelligence technology,machine learning and deep learning provide core technologies for intelligent assisted diagnosis and treatment based on TCM and related knowledge mining.The using of artificial intelligence technology to build a TCM prescription recommendation model to assist TCM physicians in diagnosing patients and discovering new prescriptions will help promote the application and development of artificial intelligence in the field of TCM.In the past,traditional bag-of-words-based statistical models were often used to build TCM prescription recommendation model,but statistical models were not suitable for short text corpora such as TCM prescriptions.This makes it difficult for statistical models to gain insight into the complex correlations between symptoms and herbs in TCM prescriptions.At present,graph neural network is often used to build TCM prescription recommendation model,but the influence of attributes of herb has been ignored in previous work.Therefore,in view of the shortcomings of previous work,this thesis proposes a multi-layer information fusion based on graph neural network for knowledgedriven TCM prescription recommendation model(KDHR).The main work and contributions are as follows:(1)Aiming at the fact that TCM prescription data only contains symptom name and herb name without additional auxiliary information,a knowledge graph of herb attributes is constructed,and relevant attribute information of herb is introduced to alleviate the weak correlation between the data.(2)A multi-layer information fusion based on graph neural network for knowledgedriven TCM prescription recommendation model is designed.KDHR uses the attributes of herb as additional auxiliary information and uses the multi-layer perceptron to fuse the different levels of information trained by each layer of graph neural network to obtain the node feature representation with rich information and less noise.(3)The effectiveness of each component design in the KDHR algorithm is verified by four ablation experiments.(4)Through seven comparative experiments on the TCM prescription benchmark dataset,including the most advanced TCM prescription recommendation model,the effectiveness and advancement of the KDHR algorithm in the TCM prescription recommendation task are verified.