A Hierarchical Attention Network for Cross-Domain Group Recommendation

Area of Science:

• Recommender Systems
• Artificial Intelligence
• Data Science

Background:

• Online services require personalized recommendations for user groups.
• Data sparsity and complex user-group relationships challenge group recommender systems (GRSs).
• Existing cross-domain recommender systems (CDRSs) are designed for individuals, not groups.

Purpose of the Study:

• To alleviate data sparsity in GRSs.
• To propose a novel cross-domain group recommendation method.
• To improve personalized recommendations for both individual users and groups in data-scarce domains.

Main Methods:

• Developed a hierarchical attention network-based cross-domain group recommendation (HAN-CDGR) method.
• Constructed a hierarchical attention network to model individual and group preferences, considering member interactions and relationships.
• Employed adversarial learning for effective knowledge transfer from a source to a target domain.

Main Results:

• HAN-CDGR effectively leverages data from a source domain to improve recommendations in a sparse target domain.
• The method accurately generates recommendations for both individual users and user groups.
• Experimental results demonstrate the effectiveness and superiority of the proposed HAN-CDGR method.

Conclusions:

• The proposed HAN-CDGR method successfully addresses data sparsity in group recommendation.
• Hierarchical attention networks and adversarial learning are effective for cross-domain group recommendation.
• This approach offers accurate and personalized recommendations for diverse online group activities.