Aaditeshwar Seth, Jie Zhang, Robin Cohen
We propose a method to determine the credibility of messages that are posted in participatory media (such as blogs), of use in recommender systems designed to provide users with messages that are considered to be the most credible to them. Our approach draws from theories developed in sociology, political science, and information science -- this results in a method for evaluating the credibility of messages that is user-specific and sensitive to the social network in which the user resides. Our methodology rests on Bayesian learning, integrating new concepts of context and completeness of messages inspired by the strength of weak ties hypothesis from social network theory. We show that our credibility evaluation model can be used to significantly enhance the performance of collaborative filtering recommendation. Experimental validation is done using a dataset obtained from digg.com, a knowledge sharing website where users indicate their satisfaction with messages that are provided to them. Our results reinforce the value of using sociological insights in recommender system design.
Subjects: 1.10 Information Retrieval; 7.2 Software Agents
Submitted: May 7, 2008