[Paper Review] A Survey of Graph Neural Networks for Social Recommender Systems

Kartik Sharma, Yeon-Chang Lee+5 more|arXiv (Cornell University)|Dec 8, 2022

Recommender Systems and Techniques131 references18 citations

TL;DR

This survey provides a comprehensive analysis of Graph Neural Networks (GNNs) in social recommender systems, categorizing GNN architectures—such as attention-based, fusion, hierarchical, and temporal models—and user sociology concepts including heterogeneous link strengths, temporal dynamics, and global preferences. It identifies attention mechanisms as the most prevalent architectural component and highlights underexplored areas like weak ties and bidirectional vs. unidirectional social networks as key future research directions.

ABSTRACT

Social recommender systems (SocialRS) simultaneously leverage the user-to-item interactions as well as the user-to-user social relations for the task of generating item recommendations to users. Additionally exploiting social relations is clearly effective in understanding users' tastes due to the effects of homophily and social influence. For this reason, SocialRS has increasingly attracted attention. In particular, with the advance of graph neural networks (GNN), many GNN-based SocialRS methods have been developed recently. Therefore, we conduct a comprehensive and systematic review of the literature on GNN-based SocialRS. In this survey, we first identify 84 papers on GNN-based SocialRS after annotating 2151 papers by following the PRISMA framework (preferred reporting items for systematic reviews and meta-analyses). Then, we comprehensively review them in terms of their inputs and architectures to propose a novel taxonomy: (1) input taxonomy includes 5 groups of input type notations and 7 groups of input representation notations; (2) architecture taxonomy includes 8 groups of GNN encoder notations, 2 groups of decoder notations, and 12 groups of loss function notations. We classify the GNN-based SocialRS methods into several categories as per the taxonomy and describe their details. Furthermore, we summarize benchmark datasets and metrics widely used to evaluate the GNN-based SocialRS methods. Finally, we conclude this survey by presenting some future research directions. GitHub repository with the curated list of papers are available at https://github.com/claws-lab/awesome-GNN-social-recsys.

Motivation & Objective

• To systematically survey Graph Neural Network (GNN)-based approaches in social recommender systems, focusing on architectural diversity and sociological modeling.
• To identify and categorize the most prominent GNN architectures used in social recommendation, including attention, fusion, hierarchical, and temporal models.
• To analyze the integration of user sociology concepts such as heterogeneous link strengths, temporal evolution, and global preferences in GNN-based recommendation systems.
• To highlight gaps in current research, particularly regarding weak ties and platform-specific dynamics (e.g., Facebook vs. Twitter), and propose future research directions.
• To establish a taxonomy of GNN architectures and user sociology factors, serving as a foundational reference for future research in GNN-based social recommendation.

Proposed method

• Conducted a systematic literature review using the PRISMA framework on the Scopus database, filtering for papers published after 2009 with 'social', 'graph', and 'recommendation' in title or abstract.
• Applied predefined inclusion criteria to manually label 1958 initial results into 'Yes', 'No', or 'Maybe' relevance categories, culminating in a curated set of 41 relevant papers.
• Classified GNN architectures into categories: attention-based, fusion-based, items-as-nodes, hierarchical, temporal, and mutualistic, based on model design and component integration.
• Mapped user sociology concepts into dimensions such as heterogeneous link strength, temporal dynamics, and global preference modeling, analyzing their prevalence and impact.
• Synthesized findings into two taxonomies: one for GNN architectures and one for user sociology factors, with detailed analysis of representative models like GraphRec, HIDM, ASR, and GAT-NSR.
• Identified cross-cutting trends, such as the frequent combination of attention with other components (e.g., hierarchical or temporal), and highlighted models with novel structural innovations.

Experimental results

Research questions

• RQ1 What are the dominant architectural designs used in GNN-based social recommender systems, and how do they differ in structure and component integration?
• RQ2 How are user sociology concepts—such as heterogeneous link strength, temporal dynamics, and global preferences—modeled within GNN frameworks for social recommendation?
• RQ3 What are the most common architectural components (e.g., attention, fusion, temporal modeling) and how are they combined across state-of-the-art models?
• RQ4 Which user sociology factors are underexplored in current literature, and what implications do they have for future recommendation personalization?
• RQ5 How do platform-specific characteristics (e.g., bidirectional vs. unidirectional social networks) influence the design and effectiveness of GNN-based social recommendation systems?

Key findings

• Attention mechanisms are the most widely adopted architectural component, enabling models to dynamically weigh the influence of different neighbors in the social graph.
• The majority of models combine multiple architectural concepts—such as attention with hierarchical or temporal structures—indicating a trend toward hybrid, multi-component designs.
• Heterogeneous link strength modeling is a key factor in capturing varying degrees of social influence, with models like HeteroGraphRec and ASR demonstrating improved performance through this approach.
• Temporal modeling is increasingly integrated into GNNs, with models like TGRec and EGFRec showing enhanced performance in capturing evolving user preferences and social dynamics.
• Despite the prevalence of attention and fusion models, only a small subset of works explore weak ties or platform-specific social dynamics, indicating a significant research gap.
• The survey identifies 41 relevant papers and constructs two comprehensive taxonomies—one for GNN architectures and one for user sociology factors—providing a foundational reference for future research.