A Social Recommendation Model Based on Basic Spatial Mapping and Bilateral Generative Adversarial Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Social recommender systems aim to improve recommendations using social connections, particularly when user-item interaction data is scarce.
• Effectively fusing heterogeneous information from both interaction and social spaces is a key challenge in social recommendation.

Purpose of the Study:

• To propose a novel social recommendation model, MBSGAN, that addresses the challenge of fusing interaction and social information.
• To enhance recommendation performance by effectively mining and exploiting heterogeneous information in interaction and social spaces.

Main Methods:

• Proposing a base space mapping to integrate interaction and social spaces, overcoming information heterogeneity.
• Constructing bilateral generative adversarial networks (GANs) within both interaction and social spaces.
• Utilizing generators for candidate sample selection and discriminators for distinguishing positive/negative examples to learn complex information.

Main Results:

• The proposed MBSGAN model demonstrated superior effectiveness compared to eight other social recommendation models.
• MBSGAN also outperformed six generative adversarial network-based models across four public datasets (Douban, FilmTrust, Ciao, Epinions).

Conclusions:

• The MBSGAN model effectively fuses heterogeneous interaction and social information for improved social recommendation.
• The use of bilateral GANs and base space mapping is a promising approach for enhancing recommendation systems with social data.