Cross-Domain Recommendation Based on Sentiment Analysis and Latent Feature Mapping.

Area of Science:

• Artificial Intelligence
• Data Science
• Information Retrieval

Background:

• Cross-domain recommendation systems enhance target domain accuracy using source domain data.
• Existing methods often overlook latent sentiment information in user reviews for cross-domain feature mapping.
• User reviews contain subjective insights into preferences and attribute sentiments.

Purpose of the Study:

• To propose a novel cross-domain recommendation algorithm (CDR-SAFM) that utilizes sentiment analysis and latent feature mapping.
• To address the cold-start problem in recommendation systems by effectively transferring user sentiment features across domains.
• To improve recommendation accuracy by capturing implicit sentiment information within user reviews.

Main Methods:

• Sentiment analysis of user reviews, categorizing sentiment into positive, negative, and neutral using three-way decision ideas.
• Latent Dirichlet Allocation (LDA) for modeling user semantic orientation and generating latent sentiment review features.
• Multilayer Perceptron (MLP) for creating a cross-domain non-linear mapping function to transfer sentiment features.

Main Results:

• The proposed CDR-SAFM framework demonstrates effectiveness in cross-domain recommendation scenarios.
• The algorithm successfully maps latent sentiment features between different domains.
• Performance was validated against existing recommendation algorithms on the Amazon dataset.

Conclusions:

• Integrating sentiment analysis into cross-domain recommendation significantly enhances performance.
• The CDR-SAFM algorithm provides a robust solution for leveraging user review sentiment across domains.
• The method effectively mitigates the cold-start problem by utilizing rich sentiment data.