A semantic followee recommender in Twitter using Topicmodel and Kalman filter

Abstract

The growing number of users in microblogging sites such as Twitter has created the problem of searching useful followees among millions of users in a reasonable time. One way to address this problem is using a recommender system, which is aimed at providing a list of useful followees in a reasonable time. Although Twitter provides a functionality what it calls 'Who to Follow', neither is it configurable by the user, nor its accuracy is of the highest level. Several approaches have been proposed in literature to recommend followees in Twitter. However, their accuracy and efficiency have been limited, given several Twitter-specific and natural language processing challenges. In this paper, we propose a semantic followee recommender in Twitter based on Topicmodel and Kalman filter, leveraging publicly available knowledge-bases. In particular, we aim to address the (1) wordsense disambiguation problem in tweets using Wikipedia and WordNet, (2) classify users in multiple-labels using Topicmodel and a modified Normalized Google Distance, and (3) remove noise and predict future multi-label classes using the results obtained in step (2) above using Kalman filter. As an application, we conduct a case study to evaluate the efficacy of our model to recommend followees in six predefined classes: politics, sports, business, entertainment, science, and travel. Preliminary analysis show that the model can effectively recommend useful followees in Twitter.