Predicting the Spread of a New Tweet in Twitter

Abstract

Online social network services serve as vehicles for users to share user-generated contents (e.g. blogs, tweets, videos etc.) with any number of peers. Predicting the spread of such contents is important for obtaining latest information on different topics, viral marketing etc. Existing approaches on spread prediction are mainly focused on content and past behavior of users. However, not enough attention is paid to the structural characteristics of the network. In this paper, we propose topic based approach to predict the spread of a new tweet from a particular user in online social network namely in Twitter based on latent content interests of users and the structural characteristics of the underlying social network. We apply Latent Dirichlet Allocation (LDA) model on users’ past tweets of learn the latent topic distribution of the users. Using word-topic distribution from LDA, we next identify top-k topics relevant to the new tweet. Finally, we measure the spread prediction of the new tweet considering its acceptance in the underlying social network by taking into account the possible effect of all the propagation paths between tweet owner and the recipient user. Our experimental results on real dataset show the efficacy of the proposed approach.