BPGNN-SBR: Behavior Progressive Graph Neural Networks for Session-Based Recommendation

Zekun Xu,Wenlong Wu,Zhanzuo Yin,Xinzhe Zhao,Junnan Zhuo,Bohan Li

DOI: https://doi.org/10.1007/978-981-97-7244-5_43

2024-01-01

Abstract:Session-based recommendation leverages anonymous user interaction sequences to predict their next interaction. Most of the previous methods exploit a single user behavior for prediction, and some recent methods utilize multiple behaviors of users. However, they only consider the impact of behaviors on user preferences from a single perspective and do not explore the relationships between behaviors, resulting in inaccurate modeling. To tackle the issues, we propose Behavior Progressive Graph Neural Networks (BP-GNN) for Session-Based Recommendation, in which we model behavioral relationships from both macro and micro perspectives. In the macro view, we use item representations learned from clicking behavior as inputs to reinforce the learning process of buying behavior. In the micro view, input sequences are transformed into behavior-agnostic and operation sequences, with item representations learned for each. Then BP-GNN effectively captures information within the progressive relationships of behaviors. Experiments on three public datasets demonstrate that BP-GNN outperforms state-of-the-art models.