A Personalized Review Recommendation Method Based on Graph Bidirectional Aggregation Network Link Prediction Model

Abstract

A personalized review recommendation method based on a graph bidirectional aggregation network link prediction model, in which review features are aggregated to user features in a user-review bipartite graph. Combined with the social network, the user's neighbor features are fused to obtain the user's embedding representation. Then the user embedding representation is aggregated into comments after removing the original user features, and the user embedding representation is adjusted by comparing the difference before and after the aggregation of comments. On this basis, combined with the forwarding network, the score of the edge is calculated by the inner product of the user node features at both ends of the edge, and finally the recommendation is made according to the score. The recommendation system is used to assist users in screening, and the recommendation task of reviews is transformed into a link prediction task between users in a small range. And provide a feature extraction method that can handle heterogeneous data, making the final result more representable.

Description

technical field [0001] The invention relates to the technical field of personalized comment recommendation, in particular to a personalized comment recommendation method based on a graph bidirectional aggregation network link prediction model. Background technique [0002] Malicious comments on the Internet are a phenomenon that has been observed for a long time. The study found that the best strategy for dealing with malicious comments is to avoid responding, and the best way is to screen. Although the comments themselves have been screened by means of reverse chronological order, popularity, manual review, etc., the problem of malicious comments has not been improved. The development of recommender systems helps to solve this problem, but the quality of the recommendation results largely depends on the quality of the input features. Most of the existing methods are for the extraction of independent and the same type of data. Although they have certain expressive ability,...

AI Technical Summary

Problems solved by technology

Most of the existing methods are for extracting independent and same type of data. Although they have a certain expressive ability, due to the lack of related information, the features obtained only through the same type of data can no longer meet the needs of the development of recommendation systems, especially for recommendation systems. There are relatively few feature extraction methods for heterogeneous data that are ubiquitous in the system

Images