# Exploiting Cross-Session Information for Session-based Recommendation with Graph Neural Networks

**Authors:** Ruihong Qiu, Zi Huang, Jingjing Li, Hongzhi Yin  
**arXiv:** [2107.00852](https://arxiv.org/abs/2107.00852)  
**Date:** 2021-07-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2107.00852v2/

## Key points

- Sessions defined as sequences of interactions capturing user's recent interests
- Limitations of traditional recommender systems in capturing complex item dependency relationships
- Proposal of using graph neural networks (GNNs) for session-based recommendation
- Introduction of Full Graph Neural Network (FGNN) to learn intricate item dependencies within sessions
- Creation of Broadly Connected Session (BCS) graph to connect different sessions and enhance session embeddings
- Incorporation of cross-session information to improve recommendation accuracy
- Evaluation through experiments on Yoochoose and Diginetica datasets showing outperformance of proposed method against state-of-the-art models

## Summary

In their paper titled "Exploiting Cross-Session Information for Session-based Recommendation with Graph Neural Networks," authors Ruihong Qiu, Zi Huang, Jingjing Li, and Hongzhi Yin introduce the concept of sessions to address the limitations of traditional recommender systems. Sessions are defined as sequences of interactions between users and multiple items over a period, aimed at capturing the user's recent interests. The existing session-based recommender systems typically focus on mining sequential patterns within individual sessions but may not adequately capture complex dependency relationships among items. To overcome this challenge, the authors propose a novel approach leveraging graph neural networks (GNNs). They represent each session as a graph rather than a linear sequence structure and introduce the Full Graph Neural Network (FGNN) to learn intricate item dependencies within sessions. Additionally, they construct a Broadly Connected Session (BCS) graph to connect different sessions and develop a Mask-Readout function to enhance session embeddings based on the BCS graph. By incorporating cross-session information into individual session representation learning, their model aims to improve recommendation accuracy. The effectiveness of the proposed method is evaluated through extensive experiments conducted on two e-commerce benchmark datasets: Yoochoose and Diginetica. The experimental results demonstrate that their approach outperforms state-of-the-art session-based recommender models in terms of recommendation performance. Overall, this study contributes valuable insights into enhancing session-based recommendation systems using graph neural networks and leveraging cross-session information for improved user recommendations in various online platforms.