Rating knowledge sharing in cross-domain collaborative filtering.

Area of Science:

• Computer Science
• Machine Learning
• Recommender Systems

Background:

• Cross-domain collaborative filtering (CF) leverages shared data across related domains to improve recommendation accuracy.
• Existing CF methods often struggle with data sparsity and modeling dynamic user preferences over time.

Purpose of the Study:

• To propose a unified framework for cross-domain CF using a novel site-time coordinate system.
• To develop a generative model, Ratings Over Site-Time (ROST), for predicting ratings across multiple domains.
• To address data sparsity and model user-interest drift using ROST extensions.

Main Methods:

• A unified framework is proposed, conceptualizing CF domains within a 2-D site-time coordinate system.
• A generative model, ROST, is developed to share group-level rating patterns and impose user/item dependence.
• Two ROST extensions are introduced: ROST (sites) for sparsity alleviation and ROST (time) for user-interest drift modeling.

Main Results:

• ROST (sites) effectively alleviates rating sparsity, significantly improving prediction performance.
• ROST (time) successfully tracks and visualizes user-interest drift over time.
• Both ROST extensions demonstrate the efficacy of the unified framework.

Conclusions:

• The proposed unified framework and ROST models offer a powerful approach to cross-domain collaborative filtering.
• The framework effectively handles data sparsity and dynamic user preferences, enhancing recommendation systems.
• ROST provides a versatile solution for various cross-domain recommendation challenges.