[Paper Review] Graph Neural Networks in Recommender Systems: A Survey

Shiwen Wu, Fei Sun+3 more|arXiv (Cornell University)|Nov 4, 2020

Advanced Graph Neural Networks109 citations

TL;DR

A comprehensive survey of GNN-based recommender systems, proposing a taxonomy by information type and task, and discussing challenges, models, and future directions.

ABSTRACT

With the explosive growth of online information, recommender systems play a key role to alleviate such information overload. Due to the important application value of recommender systems, there have always been emerging works in this field. In recommender systems, the main challenge is to learn the effective user/item representations from their interactions and side information (if any). Recently, graph neural network (GNN) techniques have been widely utilized in recommender systems since most of the information in recommender systems essentially has graph structure and GNN has superiority in graph representation learning. This article aims to provide a comprehensive review of recent research efforts on GNN-based recommender systems. Specifically, we provide a taxonomy of GNN-based recommendation models according to the types of information used and recommendation tasks. Moreover, we systematically analyze the challenges of applying GNN on different types of data and discuss how existing works in this field address these challenges. Furthermore, we state new perspectives pertaining to the development of this field. We collect the representative papers along with their open-source implementations in https://github.com/wusw14/GNN-in-RS.

Motivation & Objective

• Motivate the use of graph neural networks to model graph-structured data in recommender systems.
• Provide a systematic taxonomy of GNN-based recommendation models based on information types and tasks.
• Identify core challenges in applying GNNs to various data modalities (e.g., social, knowledge graphs) and discuss solutions.
• Summarize representative models, datasets, and evaluation practices to guide future research and practice.
• Outline nine potential future directions to advance GNN-based recommendation.

Proposed method

• Present a new taxonomy classifying models by information type and recommendation task (user-item collaborative filtering, sequential, social, knowledge-graph-based, and other tasks).
• Review and summarize five categories of GNN techniques (GCN, GraphSAGE, GAT, GGNN, HGNN) and their relevance to recommendation.
• Discuss graph construction strategies, neighbor aggregation methods, and information update mechanisms in GNN-based recommender models.
• Analyze open challenges and how existing works address issues like scalability, heterogeneity, and high-order connectivity.
• Provide a survey of benchmark datasets and evaluation metrics used in GNN-based recommender research.
• Offer guidance on future research directions based on observed gaps and trends.

Experimental results

Research questions

• RQ1 What are the dominant information types used to build GNN-based recommender systems and how do they map to recommendation tasks?
• RQ2 How do different GNN architectures (GCN, GraphSAGE, GAT, GGNN, HGNN) perform in various recommendation settings and what challenges arise (scalability, heterogeneity, high-order relations)?
• RQ3 What strategies exist for graph construction and neighborhood aggregation to improve recommendation performance?
• RQ4 What are the standard datasets and evaluation practices for GNN-based recommender systems, and what real-world deployments illustrate their utility?

Key findings

• The authors propose a new taxonomy organizing GNN-based recommendation models by information type and task (five categories).
• The survey provides a comprehensive review of representative models, highlighting how they address core issues in each category.
• They identify nine potential future directions to guide subsequent research in GNN-based recommender systems.
• The paper discusses practical deployment considerations, including graph construction, neighbor sampling, and aggregation methods for scalability.
• They synthesize the landscape by linking graph learning advances to recommender system needs and data characteristics.