[Paper Review] Deep Learning on Knowledge Graph for Recommender System: A Survey

Yang Gao, Yifan Li+3 more|arXiv (Cornell University)|Mar 25, 2020

Advanced Graph Neural Networks42 references34 citations

TL;DR

A comprehensive survey of GNN-based knowledge-aware deep recommender systems (GNN-KADR), introducing a taxonomy of graph embedding components and summarizing datasets, evaluations, and open-source codes.

ABSTRACT

Recent advances in research have demonstrated the effectiveness of knowledge graphs (KG) in providing valuable external knowledge to improve recommendation systems (RS). A knowledge graph is capable of encoding high-order relations that connect two objects with one or multiple related attributes. With the help of the emerging Graph Neural Networks (GNN), it is possible to extract both object characteristics and relations from KG, which is an essential factor for successful recommendations. In this paper, we provide a comprehensive survey of the GNN-based knowledge-aware deep recommender systems. Specifically, we discuss the state-of-the-art frameworks with a focus on their core component, i.e., the graph embedding module, and how they address practical recommendation issues such as scalability, cold-start and so on. We further summarize the commonly-used benchmark datasets, evaluation metrics as well as open-source codes. Finally, we conclude the survey and propose potential research directions in this rapidly growing field.

Motivation & Objective

• Define the concepts of knowledge graphs and graph neural networks in recommender systems.
• 提供 a unified taxonomy for graph embedding modules in GNN-KADR systems (aggregators and updaters).
• Review state-of-the-art GNN-KADR models and their handling of practical issues like cold-start and scalability.
• Summarize benchmark datasets, evaluation metrics, and open-source codes.
• Suggest future research directions including dynamicity, interpretability, and fairness.

Proposed method

• Propose a new taxonomy: aggregators (relation-unaware, relation-aware subgraph, relation-aware attentive) and updaters (context-only, single-interaction, multi-interaction).
• Categorize and compare representative GNN-KADR models across aggregator/updater choices.
• Discuss practical recommendation issues (scalability, cold-start).
• Summarize datasets, metrics, and available open-source implementations.
• Provide discussion on limitations and future directions.

Experimental results

Research questions

• RQ1 What are the core graph embedding components used in knowledge-aware deep recommender systems?
• RQ2 How do different aggregators and updaters affect performance and scalability in KG-based recommender systems?
• RQ3 What datasets, metrics, and open-source resources dominate this research area?
• RQ4 What are the key challenges and potential directions for future work in GNN-KADR?
• RQ5 How do GNN-based approaches address issues like cold-start and dynamicity in recommendations?

Key findings

• Introduces a taxonomy of GNN-based knowledge-aware deep recommender (GNN-KADR) systems centered on graph embedding modules.
• Reviews representative models and compares their aggregators and updaters.
• Highlights solutions to practical issues such as scalability and cold-start.
• Provides a consolidated summary of benchmark datasets, evaluation metrics, and available open-source codes.
• Discusses limitations and proposes future directions like dynamicity, interpretability, fairness.
• Concludes that relation-aware attentive aggregators offer scalable and explainable modeling for large KG-based RS.