Douban books score prediction based on graph neural network and BERT

Abstract

This article aims to predict comment ratings on Douban Reading Platform by combining graph neural network (GCN), TF-IDF (word frequency-inverse document frequency) statistical methods, BERT model, and Heterogeneous information network (HGCN). First of all, we built a diagram of relationships including user books, users and user reviews. Secondly, we use the Bert model to encode the book review to extract text characteristics. Finally, we fixed the GCN or HGCN graph with the text of BERT and built a scoring prediction model. The research in this project aims to provide a new idea for recommendation systems based on graph neural networks and pre-trained language models.