Neighborhood exploration method based on heterogeneous graph neural network

Abstract

The invention discloses a neighborhood exploration method based on a heterogeneous graph neural network. The neighborhood exploration method comprises the following steps: step 1, defining a parameter alpha for guiding an exploration direction; 2, in the given heterogeneous graph and meta path phi, for each direct neighbor of the node, no carrying out operation on the probability of alpha, and skipping to the next direct neighbor node; carrying out biased migration sampling when the probability is 1-alpha, and replacing original direct neighbors with meta-path neighbors obtained by migration sampling; 3, adding the direct neighbors into a neighbor set according to the probability of alpha; and adding the meta-path neighbors into the neighbor set according to the probability of 1-alpha. According to the method, two exploration strategies of depth-first exploration and breadth-first exploration are smoothly spliced to adapt to different heterogeneous network structures, and specific semantic neighbors are captured, so that the interpretability and accuracy of a recommendation system are improved.

Description

technical field [0001] The invention relates to the field of computer network analysis, in particular to a neighborhood exploration method based on a heterogeneous graph neural network. Background technique [0002] In recent years, recommender systems have played an increasingly important role in various Internet products, as it can help users discover items of interest, such as movies, commodities, etc., in a huge database. The recommendation system is used to mine the user's historical behavior, and establish respective feature matrices according to the characteristics of users and products. Traditional recommendation methods, such as collaborative filtering, mainly use neighbor users or neighbor items with high similarity to predict the ratings of target user candidate items. A common practice is to first construct a user-product rating matrix, then calculate the similarity to determine the neighbor set, and finally predict the rating to generate a recommendation list. ...

AI Technical Summary

Problems solved by technology

Most recommendation algorithms based on heterogeneous information networks use meta-path-based similarity to improve accuracy, and cannot extract latent features of users or items well

[0004] Patent No. 202010094718.7, proposes a movie recommendation method based on network embedding of meta-paths, recommends different meta-paths with differen

Images