A Graph Neural Network Recommendation Method Incorporating Review Information

Abstract

The invention discloses a graph neural network recommendation method incorporating comment information, using user comment text as a source of node information, and using a BERT model to perform feature extraction of text data to obtain a content expression vector of each node. Then build a user-item bipartite graph based on the interaction between the user and the item, and learn the structural representation of the user and the item by building a graph neural network on the bipartite graph to aggregate the first-order and third-order neighbor information of the nodes. Representations are fused with content representations as the final embedded representations of users and items. Finally, the user's interaction probability for each item is predicted by the multi-layer perceptron MLP. For the obtained prediction results, use Top-N sorting to generate a list of recommended items. The invention can more accurately capture user preference, find user's point of interest, and perform more accurate and effective recommendation on it.

Description

technical field [0001] The invention relates to the field of computer information recommendation methods, in particular to a graph neural network recommendation method incorporating comment information. Background technique [0002] The recommendation system can solve the problem of information overload and develop rapidly. For example, the software on the mobile phone, shopping, music, video, etc. are all related to recommendation. Most of the core is to analyze the user's behavior to study the user's interests and hobbies, and recommend the user's interest from a large amount of data. [0003] Traditional recommendation algorithms include content-based recommendation, collaborative filtering-based recommendation and hybrid recommendation. Content-based recommendation uses information comparison and filtering technology. The content of the item does not need to be scored, but it needs easy-to-extract content features and strong structure. The recommendation accuracy is not...

AI Technical Summary

Problems solved by technology

[0005] (1) Sensitive to data sparsity and cold start

[0006] (2) Single use of interaction information between users and items cannot make good use of other useful information

[0007] (3) For high-order neighbors, the user's concerns cannot be well identified

Images