Author detection: Analyzing tweets by using a Naïve Bayes classifier

Abstract

In the context of digital social media, where users have multiple ways to obtain information, it is important to have tools to detect the authorship within a corpus supposedly created by a single author. With the tremendous amount of information coming from social networks there is a lot of research concerning author profiling, but there is a lack of research about the authorship identification. In order to detect the author of a group of tweets, a Naïve Bayes classifier is proposed which is an automatic algorithm based on Bayes’ theorem. The main objective is to determine if a particular tweet was made by a specific user or not, based on its content. The data used correspond to a simple data set, obtained with the Twitter API, composed of four political accounts accompanied by their username and tweet identifier as it is mixed with multiple user tweets. To describe the performance of the classification model and interpret the obtained results, a confusion matrix is used as it contains values like accuracy, sensitivity, specificity, Kappa measure, the positive predictive and negative predictive value. These results show that the prediction model, after several cases of use, have acceptable values against the observed probabilities.