Graph Learning based Recommender Systems: A Review (2105.06339v1)

Abstract: Recent years have witnessed the fast development of the emerging topic of Graph Learning based Recommender Systems (GLRS). GLRS employ advanced graph learning approaches to model users' preferences and intentions as well as items' characteristics for recommendations. Differently from other RS approaches, including content-based filtering and collaborative filtering, GLRS are built on graphs where the important objects, e.g., users, items, and attributes, are either explicitly or implicitly connected. With the rapid development of graph learning techniques, exploring and exploiting homogeneous or heterogeneous relations in graphs are a promising direction for building more effective RS. In this paper, we provide a systematic review of GLRS, by discussing how they extract important knowledge from graph-based representations to improve the accuracy, reliability and explainability of the recommendations. First, we characterize and formalize GLRS, and then summarize and categorize the key challenges and main progress in this novel research area. Finally, we share some new research directions in this vibrant area.

Review of "Graph Learning based Recommender Systems: A Review"

In the paper "Graph Learning based Recommender Systems: A Review," the authors present a comprehensive examination of the recent developments in Graph Learning based Recommender Systems (GLRS), a swiftly advancing area within the field of Artificial Intelligence. By employing graph learning techniques, GLRS aim to model user preferences and item characteristics in recommender systems using graph structures. The paper reviews various approaches within this paradigm, categorizes prevailing challenges, and offers a perspective on ongoing and future research directions.

Formalization and Categorization

GLRS leverage graphs to represent the relationships among users, items, and associated attributes, forming connections either explicitly or implicitly. The core of the paper is focused on defining GLRS, understanding how they utilize graph-based methods, and categorizing them into classes based on the nature of the graphs and the recommendation tasks they aim to solve. They are mainly categorized into models dealing with general interaction data, sequential interaction data, incorporation of side information, and external knowledge such as item ontology or common domain knowledge.

Technical Progress in Graph Learning Approaches

The paper systematically categorizes GLRS approaches into several main types:

1. Random Walk Approach: This method models the propagation over a graph to capture relations between nodes (e.g., users and items) by simulating random paths. Though effective in exploring indirect and high-order node relations, the random walk approach often suffers from computational inefficiency on large-scale graphs.
2. Graph Embedding Approach: Graph embedding translates graph data into lower-dimensional vector representations. Methods like Graph Factorization, Graph Distributed Representation, and Graph Neural Embedding focus on capturing complex inter-node relations that aid in recommendations.
3. Graph Neural Networks (GNN): Subtypes such as Graph ATtention networks (GAT), Gated Graph Neural Networks (GGNN), and Graph Convolutional Networks (GCN) are highlighted. These techniques utilize neural networks to learn node representations with an emphasis on attention mechanisms, recurrent networks, and convolutional operations to capture nuanced graph features.

Implications and Future Directions

The paper outlines several implications of GLRS, emphasizing how these systems can address challenges related to sparse data and user-item interaction modeling. Practical implications include enhanced recommendation accuracy and improved capability to address the cold-start problem. The theoretical implications point toward a richer understanding of user behavior through graph-based data analysis.

The authors suggest promising future developments:

• Self-evolutionary RS: Addressing the dynamic nature of graph topology as users and their interactions evolve over time.
• Explainable RS: Integration of causal graph learning to provide deeper insights into user decisions and item characteristics.
• Cross-domain RS: Use of multiplex graph learning to handle inter-domain interactions and enhance cross-domain recommendations.
• Efficient Online RS: Developing algorithms capable of handling large-scale graphs efficiently for real-time recommendations.

The paper serves as a comprehensive resource for understanding the promising capabilities of GLRS and emphasizes the need for ongoing research to explore their full potential. It encourages further exploration to overcome existing limitations and leverage graph learning beyond traditional paradigms.