Exploiting Dual-Attention Networks for Explainable Recommendation in Heterogeneous Information Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Retrieval

Background:

• Traditional recommendation systems struggle with sparse data and limited interpretability.
• Heterogeneous Information Networks (HINs) offer rich contextual data but are challenging to integrate.
• Explainable recommendation aims to provide both item suggestions and the reasoning behind them.

Purpose of the Study:

• To propose a novel framework, Dual-Attention Networks for Explainable Recommendation (DANER), for HINs.
• To improve recommendation performance and provide interpretable explanations.
• To effectively integrate complex contextual information from HINs into recommendation models.

Main Methods:

• Utilized multiple meta-paths in HINs to capture high-order semantic relations and generate similarity matrices.
• Employed matrix decomposition for low-dimensional sparse user and item representations.
• Introduced dual-level attention networks (local and global) to integrate representations from different meta-paths.
• Used a multi-layer perceptron for user-item interaction modeling and rating prediction.

Main Results:

• DANER demonstrated superior recommendation performance compared to state-of-the-art methods on real-world datasets.
• The dual-attention mechanism effectively identified critical meta-paths for generating relevant explanations.
• A case study confirmed the interpretability of the proposed DANER framework.

Conclusions:

• DANER effectively leverages HINs for enhanced explainable recommendation.
• The dual-attention mechanism is key to improving both accuracy and interpretability.
• The framework offers a promising approach for building more transparent and effective recommendation systems.