Comparative Convolutional Dynamic Multi-Attention Recommendation Model.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Attention mechanisms improve recommender systems by focusing on pivotal user interests.
• Current models often fail to mine user preferences comprehensively, limiting recommendation accuracy.

Purpose of the Study:

• To propose a novel recommendation model, Comparative Convolutional Dynamic Multi-Attention (CCDMA), for more accurate user interest mining.
• To address the limitations of existing methods in thoroughly understanding user preferences.

Main Methods:

• Utilizing multi-attention-based convolutional neural networks for dynamic extraction of user and item latent feature vectors.
• Implementing both self-attention (internal) and cross-attention (mutual) mechanisms.
• Introducing an optimized comparative learning framework to mine ternary relationships between users and item pairs.

Main Results:

• The proposed CCDMA model demonstrates superior performance compared to state-of-the-art baselines.
• Experiments on real-world datasets validate the effectiveness of the CCDMA model across various evaluation metrics.

Conclusions:

• CCDMA offers a more accurate approach to represent user and item features for enhanced recommendations.
• The model's ability to mine ternary relationships and utilize multi-attention significantly improves preference understanding.