Design and implementation of a graph neural network-based book recommendation system

Abstract

In response to the common problems of limited accuracy and insufficient personalization of traditional book recommendation systems, this study has designed and implemented a highly accurate personalized book recommendation system based on graph neural networks. Through the analysis of the research on recommendation systems and graph neural networks, a user-book bipartite graph model was constructed using graph neural networks. Four recommendation models, namely Two-Tower, Matrix Factorization, Neural Graph Collaborative Filtering, and LightGCN, were compared on the Amazon Reviews 2023 book dataset. The experimental results show that LightGCN outperforms the other models in all indicators, with a Recall@20 of 0.10089, verifying the effectiveness of the graph neural network-based method in the book recommendation task.

The backend of the system uses the FastAPI framework and RESTful API interfaces to implement the recommendation service; the PostgreSQL database is used to store user, book, and review data; the front end is implemented using Vue.js for the interactive interface. The system supports personalized book recommendations, book search, book detail display, comment management, and history record query functions, and realizes the complete process from the recommendation model to the actual application platform.