Abstract

Twitter has become the most widely used microblogging service nowadays, where people tells and spread, with short messages, what are they feeling or what it is happening at that moment. For this reason, having an insight of the behavior of the messages stream inside the social network could be of great help to support difficult challenges such as event detection, credibility analysis, and marketing, among others problems in social network analysis. A massive amount of data is generated in this context, and a simple idea that might be useful for every challenging mining task consists of predicting the amount of messages (stream volume) that will be emitted in some specific time span. In this work we model the messages’ stream volume as a time series by counting the number of collected messages during a time interval. Moreover, computational intelligence techniques are applied to identify the most influential regressors or lags, and a nonlinear autoregressive model is adjusted to this time series. Simulation experiments were conducted for a sample of over 900K collected tweets in a specific geographic area. With this methodology, an attempt to answer some questions about the behavior of the stream volume will be made.