A data augmentation model integrating supervised and unsupervised learning for recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Supervised learning in Graph Neural Networks (GNNs) for recommendations suffers from extremely sparse label data, limiting embedding quality.
• Insufficient labeled data leads to overfitting in recommendation models.
• Existing data augmentation methods often rely on traditional labeled data.

Purpose of the Study:

• To propose a novel recommendation model, DARec, that addresses the challenge of sparse labels in Graph Neural Networks.
• To fuse supervised and unsupervised learning tasks with advanced data augmentation techniques for improved recommendation performance.
• To leverage unlabeled data effectively for enhanced learning efficiency in recommendation systems.

Main Methods:

• Proposed DARec, a recommendation model integrating supervised and unsupervised learning tasks.
• Utilized diffusion models for data augmentation in supervised learning tasks.
• Employed edge dropout on user-item interaction graphs and knowledge graphs (KGs) for unsupervised learning.
• Generated supervisory signals from input data, eliminating reliance on traditional labeled data.

Main Results:

• DARec demonstrated superior performance compared to state-of-the-art recommendation models on three public datasets.
• The model successfully learned feature representations without explicit labels, enhancing learning efficiency.
• Minimizes damage to the original interaction matrix and graph structure during the learning process.

Conclusions:

• DARec offers a robust solution for recommendation systems facing sparse label data by effectively utilizing data augmentation.
• The proposed approach enhances learning efficiency by leveraging unlabeled data through self-generated supervisory signals.
• DARec represents a significant advancement in developing high-quality embedding representations for recommendation models.