Time sensitivity-based popularity prediction for online promotion on Twitter

Abstract Currently, companies and individuals tend to use social media to publish information for promoting products and options. Although an increasing body of research has focused on promotional skills on social media, the role of post-publication time is rarely studied. However, publication time of a post plays an important role in its popularity. To select suitable publication times, an effective approach is to predict popularity values of a post when it is published at a series of future time points. However, this task is not trivial because, (i) except for publication time, all the features are inefficient, as they are the same, and (ii) the new model needs to output multiple popularity values for each input post. To address these challenges, we introduce a latent factor model to build a time sensitivity-based predictive model that can predict posts’ popularity values when they are published at various times. In this model, to alleviate data sparsity, we decompose posts into syntactic units that are derived from dependency parsing results. To take advantage of auxiliary information, we exploit the features of temporal information and neighborhood influence. Experiments with Twitter data demonstrate that the proposed model significantly outperforms state-of-the-art methods.