Preference and Concurrence Aware Bayesian Graph Neural Networks for Recommender Systems

TL;DR

The paper addresses noise and incompleteness in observed user-item graphs by enhancing Bayesian Graph Neural Networks with a flexible graph generative model. It introduces PECO, a bipartite graph generation approach that jointly accounts for user preference, item concurrence, and node degree to produce informative synthetic graphs for BGNN training. Through extensive experiments on four public datasets, PECO consistently improves over Node-Copy and strong baselines, with dataset-specific tuning of the concurrence weight. The method offers a practical, adaptable framework to robustify recommender systems against graph noise and missing signals, with potential for further refinements in node classification and hyperparameter learning.

Abstract

Graph-based collaborative filtering methods have prevailing performance for recommender systems since they can capture high-order information between users and items, in which the graphs are constructed from the observed user-item interactions that might miss links or contain spurious positive interactions in industrial scenarios. The Bayesian Graph Neural Network framework approaches this issue with generative models for the interaction graphs. The critical problem is to devise a proper family of graph generative models tailored to recommender systems. We propose an efficient generative model that jointly considers the preferences of users, the concurrence of items and some important graph structure information. Experiments on four popular benchmark datasets demonstrate the effectiveness of our proposed graph generative methods for recommender systems.

Figures (1)

• Figure 1: The item degree distribution. Item indices are sorted according to the item degree in the original graph.