Transfer Learning for One-Class Recommendation Based on Matrix Factorization

MPhil Thesis Defence


Title: "Transfer Learning for One-Class Recommendation Based on
Matrix Factorization"

By

Mr. Ruiming XIE


Abstract

One Class Recommender System aims at predicting users' future behaviors 
according to their historical actions. In these problems, the training 
data usually contain only binary data reflecting the behavior is happened 
or not. Thus, the data is sparser than traditional rating prediction 
problems. Recently, there are two ways to tackle the problem. 1, using 
knowledge transferred from other domains to mitigate the data sparsity 
problem. 2, providing methods to distinguish negative data and unlabeled 
data. However, it's not easy to transfer knowledge simply from source 
domain to target domain since their observations may be inconsistent. And 
without data from external source, distinguishing negative and unlabeled 
data is sometimes infeasible.

In this paper, we propose a novel matrix tri-factorization method to 
transfer the useful information from source domain to target domain. Then 
we embed this method to a cluster-based SVD (singular value decomposition) 
framework. In several real-world datasets, we show our method achieve 
better prediction precision than other state-of-the-art methods. The 
cluster-based SVD method has been online for 2 months in an online 
shopping site, and its Performance is among the best.


Date:			Thursday, 12 February 2015

Time:			9:30am - 11:30am

Venue:			Room 3494
 			Lifts 25/26

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


**** ALL are Welcome ****