MCRec: A Co-Attention Based Meta-Path Contextualized Recommender System for Heterogeneous Information Networks

Recommendation systems play a crucial role in personalized content delivery across various domains, yet the challenge of modeling complex interactions in heterogeneous information networks (HINs) remains underexplored. In this paper, we propose MCRec, a novel recommendation framework designed to address this challenge by leveraging a co-attention mechanism to integrate multi-faceted context from users, items, and meta-paths. MCRec refines the representation of these entities and their relationships, improving recommendation accuracy and providing interpretability. Through a priority-based sampling strategy, we ensure that the most informative meta-paths are utilized for training. Extensive experiments on real-world datasets, including MovieLens and Yelp, show that MCRec outperforms state-of-the-art recommendation algorithms in terms of both precision and interpretability. We also highlight the framework’s strengths in cold-start scenarios and its ability to provide meaningful insights into user-item interactions. Finally, we discuss the limitations of MCRec, including its reliance on manual meta-path selection and scalability challenges, and suggest directions for future work, including automated meta-path discovery, scalability improvements, and temporal modeling.