On bootstrapping web service recommendation

Abstract

We present a novel framework to bootstrap Web Service recommendation. Service recommendation has become an effective means to achieve personalized service selection. It leverages past user-service interaction information to accurately predict user preference on previously unknown services. However, one key impediment has been the incompetence of current service recommendation systems in dealing with new users and services. Since a recommendation system has no knowledge about new users and services, it may completely fail to provide any recommendation or provide very poor ones. The proposed framework uses an agile interview process to quickly profile new users and services. The interview is structured by a decision tree that enables adaptive, intuitive, and rapid querying of users or services. We propose to exploit Non-negative Matrix Tri-Factorization (NMTF) to simultaneously partition users and services into a set of user and service groups. The group structure helps estimate the missing interaction information and also provides class labels to construct decision trees for both users and services, which will be used in the interview process. We conduct extensive experiments to assess the effectiveness of the proposed framework for bootstrapping service recommendation.