Contrastive Learning-Based Personalized Tag Recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Personalized tag recommendation systems aim to provide users with relevant tags based on their preferences.
• Traditional methods struggle with data sparsity, hindering accurate user, item, and tag embedding learning.
• This limitation impacts the effectiveness of personalized tag recommendation models.

Purpose of the Study:

• To propose a novel contrastive learning-based personalized tag recommendation algorithm (CLPTR).
• To overcome the data sparsity problem in traditional tag recommendation systems.
• To enhance the accuracy of learning user, item, and tag embeddings.

Main Methods:

• Developed CLPTR, a contrastive learning algorithm for personalized tag recommendation.
• Generated augmented user-tag and item-tag interaction graphs by injecting noise into implicit features.
• Integrated contrastive learning with a graph neural network (GNN) model for self-supervised signal extraction.

Main Results:

• CLPTR effectively preserves the semantics and structural information of interaction graphs.
• The proposed method demonstrates state-of-the-art performance compared to traditional models.
• Extensive experiments on real-world datasets validate the effectiveness of CLPTR.

Conclusions:

• CLPTR offers a significant improvement over traditional personalized tag recommendation approaches.
• The noise-injection strategy in graph augmentation is crucial for preserving data integrity.
• This research advances the field of personalized recommendation systems through effective contrastive learning integration.