Hierarchical fine-grained multi-behavior recommendation with behavior-aware contrastive learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Real-world recommendation systems utilize diverse user interactions like clicks, adds-to-cart, and purchases.
• Existing multi-behavior recommendation models face limitations in inferring true user intent and integrating complex relational data.

Purpose of the Study:

• To propose a novel framework, Hierarchical Fine-grained Multi-behavior Recommendation with Behavior-aware Contrastive Learning (HFCL), to address limitations in current recommendation systems.
• To enhance the understanding of multi-dimensional user preferences by disentangling user intents behind different behaviors.

Main Methods:

• Developed a framework that disentangles user intents to learn fine-grained preferences and incorporates intent influence into multi-behavior semantics.
• Designed dual-level graph convolutional networks (GCNs) to integrate high-order relations at specific and global levels.
• Introduced a behavior-aware contrastive learning approach to balance consistency and uniqueness across behavior embeddings.

Main Results:

• The proposed HFCL framework demonstrated superior performance compared to state-of-the-art baseline methods in recommendation tasks.
• HFCL effectively addresses the limitations of existing models in inferring user intent and leveraging high-order relationships.

Conclusions:

• HFCL offers a significant advancement in multi-behavior recommendation by accurately modeling user intent and relationships.
• The framework provides a more nuanced understanding of user preferences, leading to improved recommendation quality.