SiReN: Sign-Aware Recommendation Using Graph Neural Networks.

Area of Science:

• Artificial Intelligence
• Computer Science
• Machine Learning

Background:

• Recommender systems commonly use network embedding (NE) with graph neural networks (GNNs) to improve accuracy.
• Existing methods often overlook low rating scores, which can provide valuable user preference information.
• There's a need to effectively utilize negative feedback in NE-based recommender systems.

Purpose of the Study:

• To develop a novel sign-aware recommender system (SiReN) using GNNs.
• To effectively incorporate low rating scores into user preference representation.
• To enhance recommendation accuracy by utilizing both positive and negative user-item interactions.

Main Methods:

• Constructing a signed bipartite graph with positive and negative edges.
• Generating separate embeddings for positive and negative interactions using GNNs and MLPs, combined via an attention model.
• Employing a sign-aware Bayesian personalized ranking (BPR) loss function for optimization.

Main Results:

• SiReN demonstrates consistent performance improvements over state-of-the-art NE-aided recommendation methods.
• The proposed method effectively utilizes information from low rating scores.
• Signed graph construction and sign-aware optimization enhance preference representation.

Conclusions:

• SiReN offers a significant advancement in NE-based recommender systems by effectively utilizing signed user-item interactions.
• The system provides a robust framework for incorporating negative feedback, leading to superior recommendation accuracy.
• This approach highlights the importance of sign-aware learning in modern recommender systems.