Alleviating the Sparsity Problem of Collaborative Recommender Systems using Scalable Transfer Learning

Prof. Lior Rokach, Dept. of Information System Eng., Ben-Gurion Univ. of the Negev

Abstract:

The collaborative filtering (CF) approach plays a central role within many recommender systems.

The lack of ratings data can pose a major challenge to collaborative filtering algorithms. One approach to address this sparsity problem is to utilize data from other domains. In this talk I will present a transfer learning algorithm that extracts knowledge from multiple dense domains (e.g., movies and music) in order to boost the model's generation in a sparse target domain (e.g., games). The proposed algorithm learns the relatedness between the different source domains and the target domain, without requiring common users or items. Experiments with several datasets show that, using multiple sources and the relatedness between domains improves the predictive performance of the recommender system. In addition, I will present a distributed version of our algorithm that can scale well and efficiently process extremely large rating data on commodity hardware.

ההרצאה תתקיים ביום שלישי, .25.12.12, בשעה 14:00 בחדר 206, בנין וולפסון הנדסה, הפקולטה להנדסה, אוניברסיטת תל-אביב