Cross-border e-commerce recommendation method based on heterogeneous graph expression learning

Abstract

The invention discloses a cross-border e-commerce recommendation method based on heterogeneous graph expression learning. Comprising the following steps: carrying out quantitative analysis on a real cross-border e-commerce data set; obtaining topic probability distribution of cross-border e-commerce products through a latent semantic topic model LDA; selecting the theme corresponding to the maximum probability distribution value as the theme of the final product; constructing a 'user-product-theme' tripartite graph; for users and items having a high-order side relationship in a user-product-theme tripartite graph, proposing HNGR to carry out embedded propagation learning, including information propagation and information aggregation, so that high-quality user and product expression vectors are obtained; and modeling user-product interaction through a multi-layer perceptron (MLP) to generate a recommendation result. According to the invention, the user purchase record data of the cross-border e-commerce platform is used as a drive, and the graph neural network of heterogeneous graph expression learning is used as a model, so that the interest preference of the user can be identified, and meanwhile, the sparsity problem of a user-product purchase matrix can be relieved.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, in particular to a cross-border e-commerce recommendation method (Heterogeneous Neural Graph Recommendation, HNGR for short) based on heterogeneous graph expression learning. Background technique [0002] For traditional shopping scenarios, some mature recommendation algorithms have been widely used. Among them, the most classic ones are recommendation based on collaborative filtering, recommendation based on matrix factorization and recommendation based on content. However, due to cross-border e-commerce products The variety and complexity of information types, the extremely sparse "user-item" matrix and the prominent cold start problem make these three models difficult to work in the recommendation process. In addition, improved recommendation models based on collaborative filtering or matrix decomposition only consider the "explicit" and "implicit" feedback information of users o...

AI Technical Summary

Problems solved by technology

[0002] For traditional shopping scenarios, some mature recommendation algorithms have been widely used. Among them, the most classic ones are recommendation based on collaborative filtering, recommendation based on matrix factorization and recommendation based on content. However, due to cross-border e-commerce products Various types of information, complex types, extremely sparse "user-item" matrix and prominent cold start problem make these three models difficult to work in the recommendation process

In addition, improved recommendation models based on collaborative filtering or matrix decomposition only consider the "explicit" and "implicit" feedback information of users on products, while ignoring the relationship between graph structure information composed of users and items and products. The latent semantic topic association, the recommendation performance is difficult to meet the requirements of the platform and users

Images