Boosting Knowledge Graph with Diverse-Aware Intent Inference for recommendations.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Knowledge graphs (KGs) enhance recommendation systems through semantic structure.
• Traditional methods like collaborative filtering have limitations.
• Graph Neural Networks (GNNs) model KG relationships but face challenges with information redundancy and limited diversity.

Purpose of the Study:

• To propose a novel framework, Knowledge Graph with Diverse-Aware Intent Inference (KGDII), to address limitations in GNN-based recommendation systems.
• To enhance both the quality and diversity of recommendations.
• To improve user intent inference by selecting diverse relationships and utilizing attention mechanisms.

Main Methods:

• Developed the Knowledge Graph with Diverse-Aware Intent Inference (KGDII) framework.
• Implemented a mechanism for selecting diverse subsets of relationships within the KG to generate user intents.
• Utilized an attention mechanism to prioritize significant relationships, reducing redundancy and improving intent representation.

Main Results:

• KGDII demonstrated superior performance over state-of-the-art methods in recommendation accuracy and diversity on real-world datasets.
• Ablation studies confirmed the effectiveness of the proposed components.
• Case analyses highlighted the strong interpretability of the KGDII framework.

Conclusions:

• KGDII effectively enhances recommendation quality and diversity by inferring diverse user intents.
• The framework's attention mechanism and diverse relationship selection contribute to improved performance and reduced redundancy.
• KGDII presents a promising approach for advancing recommendation system performance and interpretability.