FIRE: knowledge-enhanced recommendation with feature interaction and intent-aware attention networks.

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Retrieval

Background:

• Recommender systems are crucial for managing information overload and improving user experience in web applications.
• Knowledge Graphs (KGs) enhance recommender systems by providing real-world facts and entities.
• Graph Neural Networks (GNNs) are increasingly used for knowledge-aware recommendation (KGR), but existing models neglect feature interaction and user intent modeling.

Purpose of the Study:

• To address the limitations of current GNNs-based KGR models, specifically in high-order feature interaction and user intent modeling.
• To propose a novel Knowledge-enhanced Recommendation with Feature Interaction and Intent-aware Attention Networks (FIRE) model.
• To enhance user and item representation learning for improved KGR performance.

Main Methods:

• Developed the FIRE model, integrating GNNs with Convolutional Neural Networks (CNNs) for multi-granular feature interactions.
• Employed vertical (bit-level) and horizontal (vector-level) convolutions within CNNs to model high-order feature interactions.
• Utilized a two-level attention mechanism (node- and intent-level) to capture users' latent intent factors.

Main Results:

• The FIRE model demonstrated superior performance compared to state-of-the-art baselines on three public KG datasets.
• Ablation studies confirmed the effectiveness of the proposed feature interaction and intent-aware attention mechanisms.
• Model studies provided insights into the working mechanisms and plausibility of the FIRE approach.

Conclusions:

• The proposed FIRE model effectively addresses key limitations in existing KGR methods by enhancing feature interaction and user intent modeling.
• FIRE significantly improves user and item representation learning, leading to state-of-the-art performance in knowledge graph recommendation.
• The novel integration of CNNs and attention mechanisms offers a promising direction for future research in knowledge-aware recommender systems.