Heterogeneous Graph Neural Network with Attribute Sentiment for Product Recommendation
Establishing the relationship between user interests and product attributes term is helpful to improve the accuracy of product recommendation system.We propose a graph neural network product recommendation model based on a multi-head attention mechanism by integrating attribute term sentiment,which realizes the updating of edges and nodes in the graph.We utilize BERT-SAN/ChatGLM-Turbo to automatically obtain the attribute term sentiment information of the review data,and calculate the user's preference for the attribute term and the attribute's contribution to the product reputation.On this basis,a bipartite graph of the relationship between users and products is constructed with the attribute term.Finally,a stable user and product association bipartite graph neural network is trained in terms of MSE loss.Experiments on Yelp Restaurant and Digital Music datasets show that the proposed model is significantly better than state-of-the-art methods.
attribute term sentiment analysisgraph neural networkuser preferenceattribute term contributioninterpretability
NETL
NSTL
万方数据
