[Paper Review] Graph Neural Networks for Social Recommendation

Wenqi Fan, Yao Ma+5 more|arXiv (Cornell University)|Feb 19, 2019

Advanced Graph Neural Networks42 references60 citations

TL;DR

GraphRec is a Graph Neural Network framework for social recommendation that jointly models a user-item interaction graph and a user-user social graph with opinion-aware interactions and attention mechanisms, achieving superior performance over baselines on Ciao and Epinions.

ABSTRACT

In recent years, Graph Neural Networks (GNNs), which can naturally integrate node information and topological structure, have been demonstrated to be powerful in learning on graph data. These advantages of GNNs provide great potential to advance social recommendation since data in social recommender systems can be represented as user-user social graph and user-item graph; and learning latent factors of users and items is the key. However, building social recommender systems based on GNNs faces challenges. For example, the user-item graph encodes both interactions and their associated opinions; social relations have heterogeneous strengths; users involve in two graphs (e.g., the user-user social graph and the user-item graph). To address the three aforementioned challenges simultaneously, in this paper, we present a novel graph neural network framework (GraphRec) for social recommendations. In particular, we provide a principled approach to jointly capture interactions and opinions in the user-item graph and propose the framework GraphRec, which coherently models two graphs and heterogeneous strengths. Extensive experiments on two real-world datasets demonstrate the effectiveness of the proposed framework GraphRec. Our code is available at \url{https://github.com/wenqifan03/GraphRec-WWW19}

Motivation & Objective

• Motivate leveraging social relations to enhance recommender systems.
• Develop a unified GNN framework that integrates user-item interactions and social connections.
• Capture both interactions and opinions in the user-item graph and account for heterogeneous social tie strengths.
• Demonstrate the effectiveness of GraphRec on real-world datasets and compare with strong baselines.

Proposed method

• Proposes GraphRec with three components: user modeling, item modeling, and rating prediction.
• Uses two types of aggregation: item aggregation (from user-item graph) and social aggregation (from social graph) to learn user latent factors.
• Introduces opinion-aware interaction representations by fusing item embeddings with opinion embeddings via an MLP.
• Implements attention mechanisms to weight interactions (alpha) and social ties (beta) for personalized aggregation.
• Learns item latent factors via user aggregation of opinion-aware interactions with attention (mu) and predicts ratings through an MLP using concatenated user and item factors.
• Optimizes with RMSprop on a squared loss, with embeddings for users, items, and opinions learned jointly.

Experimental results

Research questions

• RQ1 How can we coherently model two heterogeneous graphs (user-item and user-user) in a single GNN framework for social recommendation?
• RQ2 Can opinion information on user-item interactions be incorporated to improve user/item representations and rating predictions?
• RQ3 Do attention mechanisms over interactions and social ties improve the quality of learned representations?
• RQ4 What is the empirical performance of GraphRec compared with traditional and neural-network based social recommender systems on real-world datasets?

Key findings

• GraphRec consistently outperforms baseline methods across Ciao and Epinions in RMSE and MAE.
• Incorporating social network information improves recommendation performance over models that rely only on rating data.
• Including opinion information on user-item interactions provides additional predictive gains.
• Attention mechanisms for both item-level interactions and social ties contribute to performance improvements.
• Ablation variants removing either social information or opinions show degraded performance, highlighting their importance.