Graph Embedding Model over Review-Based Edges

• 1、Artificial Intelligence School, Beijing University of Posts and Telecommunications, Beijing, 100876

Abstract：In order to solve the problem of insufficient information utilization in user and item graph embedding training of recommendation system, The model RHG is proposed to train user and item graph embeddings by using reviews and interactive information in the recommendation system. Based on the edge based graph neural network, the model first constructs the edge in the graph by using the review text to aggregate the graph node information. Next, the model designs different feature spaces to improve the efficiency of the node information aggregation, and then a weighted loss function is proposed according to the importance of different nodes to optimize the training. Finally, on Amazon\'s three real datasets, the graph embedding effect is evaluated by the scoring task in the recommendation system. The mean square error of RHG is reduced by 2.75% compared with the current state-of-the-art model CGAT, which proves the effectiveness of the model.

Keywords： Pattern recognition and intelligent system graph embedding review text feature spaces