An Investigation on Repost Activity Prediction for Social Media Events

Repost activity provides a way to measure the rate of information propagation about an event on a microblog service and is a key concept to understand its success. In this paper we deal with the repost prediction challenge proposed on the WISE 2012 conference, which required us to predict repost activities for 33 posts of 6 events within a period of 30 days. To achieve this objective, we propose the construction of a representative model based on semantic relationship between the events within the dataset. Next, we use two state of the art data-mining approaches when estimating the reposting of messages: (i) the Logistic Regression and (ii) the Conditional Random Fields. We also present a novel simulation framework which best uses the characteristics of our semantic data model in predicting results.