Predicting future popularity trend of events in microblogging platforms

Abstract The fast information sharing on Twitter from millions of users all over the world leads to almost real‐time reporting of events. It is extremely important for business and administrative decision makers to learn events' popularity as quickly as possible, as it can buy extra precious time for them to make informed decisions. Therefore, we introduce the problem of predicting future popularity trend of events on microblogging platforms. Traditionally, trend prediction has been performed by using time series analysis of past popularity to forecast the future popularity changes. As we can encode the rich Twitter dynamics using a rich variety of features from microblogging data, we explore regression, classification and hybrid approaches, using a large set of popularity, social and event features, to predict event popularity. Experimental results on two real datasets of 18382 events extracted from ∼133 million tweets show the effectiveness of the extracted features and learning approaches. The predicted popularity trend of events can be directly used for a variety of applications including recommendation systems, ad keywords bidding price decisions, stock trading decisions, dynamic ticket pricing for sports events, etc.