Entropy-Enhanced Attention Model for Explanation Recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Existing deep learning recommendation systems often rely on Recurrent Neural Networks (RNNs).
• RNNs are time-consuming and struggle to capture long-range dependencies in user comments.
• Sentiment analysis and information entropy can improve user interest modeling.

Purpose of the Study:

• To develop a more efficient and effective recommendation system.
• To improve the capture of user interests through sentiment analysis and information entropy.
• To enhance the interpretability of recommendation models.

Main Methods:

• Proposed a self-attention recommendation model incorporating entropy regularization.
• Utilized sentiment analysis to understand user comment polarity and characteristics.
• Employed information entropy to filter noise and refine user interest construction.
• Introduced a multi-head self-attention network to model mixed user interactions.
• Incorporated a loss function for recommendation system interpretability.

Main Results:

• The proposed model demonstrated superior performance compared to baseline methods.
• Achieved higher Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (NDCG) scores across multiple datasets.
• The model successfully provided good interpretability for the recommendation process.

Conclusions:

• The self-attention model with entropy regularization offers significant improvements over RNN-based systems.
• Integrating sentiment analysis and information entropy effectively models user interests and reduces noise.
• The model provides a more interpretable and accurate approach to personalized recommendations.