Transfer Learning in Recommendations with Matrix Factorization Models

PhD Thesis Proposal Defence


Title: "Transfer Learning in Recommendations with Matrix Factorization Models"

by

Mr. Bin CAO


ABSTRACT:

Machine learning problems on the Web usually are not traditional supervised 
learning problems. User Adaptation (i.e. personalization) and Domain Adaptation 
are two common problems on the Web. For example, for a recommender system like 
Amazon, it may need to provide personalized movie recommendations for a user 
based his feedback from multiple domains including books and clothing. For a 
search engine like Google, it may need to serve personalized ads based on 
users' browsing history. The quality of recommendations is one of the key 
factors to the revenue for these service providers. Therefore, it is critical 
to provide high quality recommendations. However, this is challenging due the 
limited information provided by the user.

In this proposal, we consider the problem of using transfer learning to improve 
recommendations. More specifically, we ask three questions. Firstly, how to 
transfer knowledge across users or items? Secondly, how to transfer knowledge 
across user groups or item domains? Thirdly, what models could be used to solve 
the problems? To answer these questions, we propose a matrix factorization 
based collaborative and transfer learning framework for solving the problems. 
We discuss specific matrix factorization models that could handle user 
adaptation and domain adaptation separately. Furthermore, we propose a unified 
model that could handle them at the same time.


Date:                   Friday, 29 April 2011

Time:                   10:00am - 12:00noon

Venue:                  Room 3501
                         lifts 25/26

Committee Members:      Prof. Qiang Yang (Supervisor)
                         Prof. Nevin Zhang (Chairperson)
 			Dr. James Kwok
 			Dr. Raymond Wong


**** ALL are Welcome ****