A multi-intent based multi-policy relay contrastive learning for sequential recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Sequential recommendation systems aim to capture dynamic user preferences but struggle with sparse data.
• Contrastive learning (CL) shows promise in addressing data sparsity by improving item representations.
• Learning influential latent intents is crucial for understanding sequence evolution in user behavior.

Purpose of the Study:

• To develop a novel approach for sequential recommendations that effectively handles data sparsity.
• To enhance the learning of user representations by focusing on multiple latent intents.
• To improve the cooperation and effectiveness of multiple contrastive learning tasks.

Main Methods:

• A novel multi-intent self-attention module is proposed to decompose user behavior sequences into distinct latent intents.
• A multi-policy relay training strategy is introduced to manage and optimize multiple contrastive learning tasks in stages.
• The model is evaluated on four public recommendation datasets.

Main Results:

• The proposed multi-intent self-attention module effectively captures diverse user preferences from sparse data.
• The multi-policy relay training strategy enhances the performance of contrastive learning by enabling better cooperation between different data augmentations.
• Experimental results demonstrate the superiority of the proposed model over existing methods.

Conclusions:

• The novel approach significantly improves sequential recommendation performance, especially in sparse data scenarios.
• The combination of multi-intent self-attention and multi-policy relay training offers a robust solution for data sparsity challenges.
• This work provides a valuable contribution to the field of sequential recommendations and contrastive learning.