Context Enhanced Recurrent Neural Network for Session-Aware Recommendation

Abstract

Recommender systems, which suggest items that users might find most interesting based on their previous web-clicks or purchased items, have a wide range of applications. While recent methods employing recurrent neural networks, such as session-based, session-aware, and context-aware recommendations, have achieved impressive results, there remains potential for further enhancements. In this study, we introduce a two-layered Gated Recurrent Unit (GRU) architecture augmented with context through the integration of an Attention Mechanism, termed CAII (Context-Aware II). CAII employs Context-Aware Modeling and Attentive Session Modeling to predict the items users are most likely to be interested in. The performance of CAII is assessed using the Amazon EC dataset, which includes context information like prices and images. Evaluated based on Recall and Mean Reciprocal Rank (MRR), CAII not only surpasses traditional methods such as Most-Popular and Item-k-Nearest Neighbor (Item-kNN), but also outperforms established models like GRU4Rec and Multi-View Recurrent Neural Network (MVRNN). Compared to the advanced recommendation method Weighted II-RNN, CAII shows an improvement of 0.6% for Recall@20 and 1.19% for MRR@20. Interestingly, the highest Recall was observed when CAII was enhanced solely with price context. However, when both price and image contexts were incorporated, CAII achieved the best MRR. These results demonstrate the efficacy of CAII in enhancing recommender systems.