Forecasting high tide: Predicting times of elevated activity in online social media

Abstract

Social media provides a powerful platform for influencers to broadcast content to a large audience of followers. In order to reach the greatest number of users, an important first step is to identify times when a large portion of a target population is active on social media, which requires modeling the behavior of those individuals. We propose three methods for behavior modeling: a simple seasonality approach based on time-of-day and day-of-week, an autoregressive approach based on aggregate fluctuations from seasonality, and an aggregation-of-individuals approach based on modeling the behavior of individual users. We test these methods on data collected from a set of users on Twitter in 2011 and 2012. We find that the performance of the methods at predicting times of high activity depends strongly on the tradeoff between true and false positives, with no method dominating. Our results highlight the challenges and opportunities involved in modeling complex social systems, and demonstrate how influencers interested in forecasting potential user engagement can use complexity modeling to make better decisions.