Heterogeneous graph neural network traditional Chinese medicine target prediction method based on attention mechanism

The invention discloses an attention mechanism-based heterogeneous graph neural network traditional Chinese medicine target prediction method, which sequentially comprises the following steps of: 1, constructing a traditional Chinese medicine target heterogeneous graph according to input traditional Chinese medicine target related data, and initializing a feature vector of each node; 2, extracting all node pairs in the herbal medicine target heterogeneous graph, obtaining a message vector of each node pair through a message passing mechanism, aggregating the attention vector and the message vector to each corresponding node by using an aggregation mechanism, mapping the vectors of the nodes back to specific distribution of the types of the nodes, and generating feature representation of a target node; and 3, predicting the interaction relationship between the traditional Chinese medicine target pairs by using a bilinear layer and two full-connection layers. By extracting the topological structure and semantic information of the traditional Chinese medicine target heterogeneous network, the vector representation capable of fully expressing rich features of the traditional Chinese medicine and the target is generated, and the traditional Chinese medicine target prediction problem can be more effectively solved.