Metapath-fused heterogeneous network representation recommendation algorithm

Abstract

The invention relates to the technical field of recommendation models of heterogeneous networks, and discloses a heterogeneous network representation recommendation algorithm fusing meta-paths, which comprises the following steps of: 1) extracting information on a heterogeneous information network; 2) carrying out vector fusion based on meta-path weight; 3) predicting a recommendation score in combination with matrix decomposition, firstly obtaining a node sequence based on different meta-paths by using a random walk strategy, learning implicit vector representation of a user and an item in a unified dimension space by using a metapath2vec++ algorithm for different paths, then learning preference weights of nodes for different meta-paths by using a multi-layer perceptron (MLP) based on an attention mechanism, and finally, in combination with a recommendation model of matrix factorization, utilizing a path structure Hetesim similarity to constrain an implicit factor vector decomposed by a scoring matrix, and predicting a project score.

Description

technical field [0001] The invention relates to the technical field of recommendation models for heterogeneous networks, in particular to a heterogeneous network representation recommendation algorithm that integrates meta-paths. Background technique [0002] Most recommendation models based on heterogeneous networks first extract the feature vector representation of nodes through the preset meta-path, and then fit the rating matrix to achieve rating prediction. Although the existing heterogeneous network methods have achieved certain results in improving accuracy, most methods still have the following problems: [0003] (1) Only a single meta-path is used to extract node information, and an accurate representation of nodes cannot be obtained when faced with sparse data; [0004] (2) The user's preference for different path semantics is ignored when merging multiple meta-path information, resulting in inaccurate representation of nodes in complex networks. [0005] In view...

Images