Graph Intention Embedding Neural Network for tag-aware recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Tag-aware recommender systems enhance user profiling and item representation using tag data.
• Deep learning models have advanced tag-aware recommendations but face limitations in exploiting tagging history and user intentions.
• Existing methods often lack interpretability due to ignoring the intent behind user-item interactions.

Purpose of the Study:

• To propose a novel model, Graph Intention Embedding Neural Network (GIENN), to address limitations in current tag-aware recommender systems.
• To fully exploit rich content and features from users' tagging history.
• To incorporate user interaction intentions for enhanced interpretability and recommendation accuracy.

Main Methods:

• Constructing a tag-aware interaction graph (TAIG) from user tagging history.
• Employing a two-layer attention mechanism to weigh node neighbors and types within the graph.
• Leveraging tag semantics to infer user interaction intentions and propagate this information through the graph embeddings.

Main Results:

• The proposed GIENN model demonstrates superior performance compared to state-of-the-art baselines.
• Experiments on three public datasets validate the effectiveness of GIENN in tag-aware top-N recommendation tasks.
• GIENN successfully integrates tag semantics to reveal user intentions, improving recommendation quality.

Conclusions:

• GIENN effectively addresses the limitations of traditional tag-aware recommender systems.
• The model enhances recommendation accuracy and interpretability by considering user intentions and tag significance.
• GIENN represents a significant advancement in leveraging graph structures and attention mechanisms for personalized recommendations.