Hierarchical User Intention-Preference for Sequential Recommendation with Relation-Aware Heterogeneous Information...

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Information Science

Background:

• Recommender systems typically use binary user-item relationships, overlooking user intentions and cognitive processes behind preferences.
• User preferences emerge from intentions, influenced by available options, and are often embedded within complex heterogeneous information networks (HINs).
• Extracting effective representations from HINs is crucial for understanding user intentions and improving recommendation accuracy.

Purpose of the Study:

• To propose a novel model, HIP-RHINE (hierarchical user's intention and preferences modeling for sequential recommendation based on relation-aware HIN embedding).
• To capture and model hierarchical user intentions and preferences within sequential recommendation tasks.
• To improve recommendation performance by leveraging relation-aware embeddings from HINs.

Main Methods:

• Constructing a multirelational semantic space for HINs to learn relation-specific node embeddings.
• Modeling user intentions and preferences using hierarchical tree structures.
• Utilizing structured decision patterns for preference learning and subsequent recommendations.

Main Results:

• The proposed HIP-RHINE model demonstrated significant improvements in recommendation performance.
• Experimental results showed enhanced Recall and Mean Reciprocal Rank metrics compared to baseline methods.
• The model effectively captures user intentions and preferences from heterogeneous information networks.

Conclusions:

• HIP-RHINE offers a more sophisticated approach to sequential recommendation by incorporating user intentions.
• The hierarchical modeling of user preferences based on relation-aware HIN embeddings is effective.
• This approach advances recommender systems by bridging cognitive insights with network representation learning.