A feature-based regression algorithm for cold-start recommendation

Abstract

Recommender systems are widely used to help user select relevant online information. A key challenge of recommender systems is to provide high-quality recommendations for cold-start users or cold-start items. We propose a feature-based regression algorithm with baseline estimates to cope with three types of cold-start problems: cold-start system, cold-start users, and cold-start items. We consider all available information of users and items to solve the cold-start problems and take into account the user and item effects that exist in collaborative filtering systems. Compared to some existing algorithms, our algorithm is effective on the 100 k MovieLens data-set for cold-start recommendation.

Keywords:

• cold-start problems
• user and item effects
• baseline estimates
• collaborative filtering
• content-based filtering

Acknowledgments

This work was under Grand by the Natural Science Foundation of China (61202441), the Talent Science Research Start-up Foundation of Dalian University of Technology (No. 1600-852018).