TwitAg: A Multi-agent Feature Selection and Recommendation Framework for Twitter

Abstract

With increasing number of users using social networking, there has been a considerable increase in user-generated content. Social networking is often used to connect with others, but in some domains such as microblogging social networking is primarily used to find interesting content. With the diversity of information sources available in microblogs, it becomes difficult for users to find interesting information sources. Recommendation engines have been developed to mitigate the problem of interesting content location in many domains, however recommendation engine research within the domain of microblogs has not been significantly explored. A key characteristic for any recommendation system is the ability to accurately classify users. Within the field of classification research feature selection is a widely used technique for improving classification accuracy. We demonstrate Unique Feature Selection (UFS), an agent based feature selection mechanism which parallelizes feature selection within the microblogging site Twitter. We show the effectiveness of UFS in both minimizing the feature space and improving classification results.