Fusion feature recommendation method based on attention mechanism

The invention relates to the technical field of recommendation systems, in particular to a fusion feature recommendation method based on an attention mechanism, and the method mainly comprises the steps: obtaining a data set; during data preprocessing, the social network of the user is fused into the scoring matrix of the user; decomposing and fusing the obtained user score matrix and the user view matrix to obtain a matrix with user characteristics and a matrix with item characteristics; inputting the fused matrix into a multi-head attention layer to obtain a corresponding feature weight; comparing the test set with the evaluation index update weight; and obtaining a final recommendation method through reception training. Compared with a traditional recommendation method, the method has the advantages that various user item features are fused, so that feedback from low-order and high-order combined parts can be received simultaneously during learning of an embedded layer, better feature representation is lea