ExpGCN: Review-aware Graph Convolution Network for explainable recommendation.

Area of Science:

• Artificial Intelligence
• Computer Science

Background:

• Recommender systems often use reviews as explanations, formulating generation as a ranking task.
• Graph Neural Networks (GNNs) are used to learn representations on heterogeneous user-item-explanation graphs.
• Existing GNNs face limitations in message passing and efficiency for complex graphs, impacting recommendation performance.

Purpose of the Study:

• To propose a novel Explanation-aware Graph Convolution Network (ExpGCN) to address limitations in current explainable recommendation methods.
• To improve both explanation generation and item recommendation performance by effectively modeling heterogeneous interactions.

Main Methods:

• The proposed ExpGCN divides the heterogeneous interaction graph into subgraphs based on edge types.
• It employs task-oriented graph convolution to aggregate information from distinct subgraphs.
• This approach generates separate node representations for explanation ranking and item recommendation tasks.

Main Results:

• ExpGCN significantly outperforms state-of-the-art baselines across four public datasets.
• The model demonstrates high computational efficiency.
• Task-oriented convolution effectively reduces aggregation complexity and mitigates performance conflicts between tasks.

Conclusions:

• ExpGCN offers a more effective and efficient approach to explainable recommendations.
• The method successfully addresses the performance degeneration caused by conflicting learning objectives in GNNs.
• ExpGCN showcases significant potential for enhancing user experience in recommender systems through better explanations.