Session-based Recommendation with Context-Aware Attention Network

Abstract

Session-based recommendation aims to generate recommendation results based on user's anonymous session. Previous studies model the session as a sequence and use Recursive Neural Network (RNN) to represent user behavior for recommendations. Although achieved promising result, previous studies ignore the relationship between session's items and external context of session, which fails in revealing the intrinsic relation between them. To tackle the problem mentioned above, we propose a novel method, i.e., Session-based Recommendation with Context-Aware Attention Network, SR-CAAN, which enhances the ability of modeling the user preference by combining sequence prediction with session external context aware method. In the proposed method, we incorporate external knowledge with Knowledge Graph (KG) to obtain the external context of session by using attention mechanism. Each session is presented as a composition of the external context of session and user's long-short term interest is obtained by Recurrent Neural Networks (RNNs). Experiments conducted on real word datasets demonstrate that SR-CAAN outperform the state-of-the-art significantly.