Feature Interaction Dual Self-attention network for sequential recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Sequential recommendation systems leverage item features for improved accuracy.
• Current methods often use basic attention mechanisms, neglecting feature interactions and integrated representations.
• There is a need for models that capture complex feature relationships within sequential data.

Purpose of the Study:

• To propose a novel Feature Interaction Dual Self-attention network (FIDS) for sequential recommendation.
• To enhance the modeling of feature interactions and sequential transition patterns.
• To improve the accuracy of next item prediction in recommendation systems.

Main Methods:

• Developed a Feature Interaction Dual Self-attention network (FIDS) utilizing dual self-attention mechanisms.
• Employed multi-head attention to model feature interactions and create higher-order feature representations.
• Utilized two independent self-attention networks to capture item sequence and integrated feature sequence transitions.
• Incorporated stacked self-attention blocks with residual connections for robust pattern extraction.

Main Results:

• The FIDS model effectively captures both feature interactions and sequential transition patterns.
• Experiments on two real-world datasets demonstrated superior performance compared to state-of-the-art recommendation models.
• The proposed method shows significant improvements in recommendation accuracy.

Conclusions:

• The FIDS model offers a more effective approach to sequential recommendation by integrating feature interactions.
• Dual self-attention mechanisms are crucial for capturing complex patterns in sequential recommendation.
• The findings suggest a promising direction for future research in personalized recommendation systems.