We discuss an approach for predicting microscopic (individual) and macroscopic (collective) user behavioral patterns with respect to specific trending topics on Twitter. Going beyond previous efforts that have analyzed driving factors in whether and when a user will publish topic-relevant tweets, here we seek to predict the strength of content generation which allows more accurate understanding of Twitter users' behavior and more effective utilization of the online social network for diffusing information. Unlike traditional approaches, we consider multiple dimensions into one regression-based prediction framework covering network structure, user interaction, content characteristics and past activity. Experimental results on three large Twitter datasets demonstrate the efficacy of our proposed method. We find in particular that combining features from multiple aspects (especially past activity information and network features) yields the best performance. Furthermore, we observe that leveraging more past information leads to better prediction performance, although the marginal benefit is diminishing.
& Sheth, A. P.
(2012). Prediction of Topic Volume on Twitter. Proceedings of the 4th International ACM Conference on Web Science, 397-402.
Additional FilesPredictionTopicVolume.pdf (111 kB)