Enhancing collaborative filtering by user interest expansion via personalized ranking.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommender systems rely on user behavior but often neglect latent user interests.
• Existing collaborative filtering methods primarily use user-system interaction data, leaving user interests underexplored.

Purpose of the Study:

• To propose a novel collaborative filtering recommender system, iExpand, that leverages latent user interests for improved recommendations.
• To develop an item-oriented, model-based collaborative filtering framework that incorporates user interests.

Main Methods:

• iExpand utilizes a three-layer user-interests-item representation scheme inspired by topic models.
• The method focuses on user interest expansion via personalized ranking to enhance recommendation accuracy.
• It strategically addresses overspecialization and cold-start problems inherent in traditional collaborative filtering.

Main Results:

• Experimental results on three benchmark datasets demonstrate iExpand's superior ranking performance.
• iExpand achieves more accurate recommendations with reduced computational cost.
• The proposed representation scheme aids in understanding user-item-interest interactions.

Conclusions:

• iExpand offers a significant improvement over state-of-the-art recommender systems.
• The approach effectively incorporates latent user interests, enhancing recommendation quality.
• iExpand provides a robust solution for common challenges in collaborative filtering.