CoHet4Rec: A recommendation for collaborative heterogeneous information networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Recommender Systems (RS) traditionally use user-item interactions to predict preferences.
• Graph Neural Networks (GNNs) improve RS by embedding graph data, but struggle with the cold-start problem.
• Social recommendation leverages user connections, yet existing models need richer relationship exploration.

Purpose of the Study:

• To propose CoHet4Rec, a novel recommendation model.
• To address limitations of existing recommender systems, including data sparsity and the cold-start problem.
• To enhance recommendation accuracy by incorporating diverse collaborative relationships beyond social networks.

Main Methods:

• Developed CoHet4Rec, a recommendation model utilizing GNNs.
• Constructed a Collaborative Heterogeneous Information Network (CHIN) with latent collaborative heterogeneous relation factors.
• Employed factorized representations to capture diverse user-item connections and incorporate external knowledge.

Main Results:

• CoHet4Rec demonstrated superior performance against 15 state-of-the-art (SOTA) recommendation techniques.
• Achieved significant improvements in key metrics: up to 31.88% for HR@5 and 38.39% for NDCG@5.
• Effectively alleviated data sparsity and the cold-start problem through enriched network information.

Conclusions:

• CoHet4Rec offers a robust solution for enhancing recommender systems.
• The model's flexibility allows for the integration of various knowledge sources.
• This approach significantly improves recommendation quality by capturing complex user-item relationships.