Temporal Collaborative Filtering with Graph Convolutional Neural Networks
Host Publication: International Conference on Pattern Recognition (ICPR)
Authors: E. Rodrigo, D. Nguyen and N. Deligiannis
Publication Year: 2020
Temporal collaborative filtering (TCF) methods aimat modelling non-static aspects behind recommender systems,such as the dynamics in users preferences and social trendsaround items. State-of-the-art TCF methods employ recurrentneural networks (RNNs) to model such aspects. These methodsdeploy matrix-factorization-based approaches to learn the userand item representations. Recently, graph-neural-network-based(GNN-based) approaches have shown improved performance inproviding accurate recommendations over traditional MF-basedapproaches in non-temporal CF settings. Motivated by this, wepropose a novel TCF method that leverages GNNs to learn userand item representations and RNNs to model their temporaldynamics. A challenge with this method lies in the increaseddata sparsity, which makes it more complicated to obtain qualityrepresentations with GNNs. To overcome this challenge, wetrain a GNN model at each time step using a set of observedinteractions accumulated time-wise. Comprehensive experimentson real-world data show the improved performance obtainedby our method over several state-of-the-art temporal and nontemporalCF models.