Dual intent view contrastive learning for knowledge aware recommender systems.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Knowledge-aware recommendation systems struggle with sparse data and redundant entity relations.
• Existing methods often fail to fully leverage knowledge graphs for improved recommendations.

Purpose of the Study:

• To propose a novel recommendation model, DIVCL, to enhance knowledge-aware recommendation systems.
• To address the limitations of sparse supervision signals and redundant entity relations in knowledge graph-based recommendations.

Main Methods:

• Developed the Dual-Intent-View Contrastive Learning network (DIVCL) utilizing Graph Neural Networks (GNNs).
• Employed a dual-view representation learning approach (local and global views).
• Integrated intents as nodes to refine user-item interactions and filter knowledge graph relations.

Main Results:

• DIVCL demonstrated superior performance compared to state-of-the-art models on three benchmark datasets.
• The model effectively handles sparse signals and redundant relations in knowledge graphs.
• Experimental results validate the efficacy of the proposed dual-view and intent-integration strategies.

Conclusions:

• DIVCL offers a robust solution for knowledge-aware recommendation challenges.
• The integration of intents significantly enhances the model's ability to learn from sparse and noisy knowledge graph data.
• The proposed approach provides a promising direction for future research in recommender systems.