Predicting information cascade on Twitter using support vector regression

Twitter is one of the very popular micro-blogging platforms for people to share content and information. Information propagates through the interaction between users with many different ways, such as retweet, mention or reply. With those abilities, Twitter has become one of the medium for advertisers to perform the marketing campaign. Sometimes in their campaign, advertisers hire several buzzers to make the campaign activity running more organically. In this paper we will discuss how to predict information cascades, in term of number interaction over the network that will happen just after buzzer doing its campaign. We formulate the task into a regression problem and define a feature-set, then extract the features on initial interaction data to build a model for prediction. Our experiment shows that Support Vector Regression (SVR) is better than Linear Regression (LR) algorithm. SVR has Mean Absolute Error (MAE) ranged from 1.54 to 33.93. We also found that the optimal setting of initial interaction data is 2 hours time lag which hit the lowest MAE 1.54.