Research On Herb Recommendation Method Using Seq2Seq And Graph Network

With the development of artificial intelligence,concepts such as Internet medical care have been proposed.Combining Traditional Chinese Medicine(TCM)with artificial intelligence,making full use of the advantages of deep learning,and expanding the application scenarios of TCM(such as auxiliary diagnosis and treatment,disease risk assessment,remote consultation,etc.)have become current research hotspots.This paper makes an in-depth research on how to apply deep learning algorithm to TCM auxiliary diagnosis and treatment tasks.The task of this paper is to recommend herbs.This is the main task of TCM in the context of artificial intelligence.Through the analysis of patients’symptoms and combined with related theories such as syndrome differentiation and treatment,precise prescriptions are given to assist TCM doctors in clinical diagnosis and treatment,which has important application significance.Based on this,this paper mainly does the following three aspects of work:(1)Aiming at TCM entity representation learning,this paper proposes a TCM entity representation method based on hypernetwork model.Firstly,symptom-herb knowledge graph was constructed based on TCM data.Secondly,the convolution filters of the head node feature and the relation feature are generated by using the hypernetwork model.The former is convolved with relation,and the latter is convolved with entity.Finally,the connection between the head and tail entities is constructed by mapping the relationship to the entity semantic space.Experiments show that this method can effectively improve the accuracy of link prediction.(2)For herb recommendation task,this paper proposes a herb recommendation method based on graph convolutional neural network integrating multi-graph features.This method introduces a simplified graph convolutional neural network,which simplifies graph convolutional neural network by removing nonlinear operations that have little effect on graph convolution operations and reducing redundant weight matrices.The multi-layer neural network is used and the multi-head attention mechanism is introduced to realize the fusion of Chinese medicine and symptom node features.The experimental results show that the method proposed in this paper can effectively improve the accuracy of traditional Chinese medicine recommendation.(3)For herb recommendation task,this paper proposes a Transformer-based herb recommendation method.This method regards the clinical diagnosis and treatment process of TCM practitioners as the process of sequence generation,and applies Transformer to the herb recommendation task.In addition,part of the method(1)is used as a pre-training method to deal with the data sparse problem.The experimental results verify the effectiveness of the method on the recommendation task of traditional Chinese medicine.