Authorship Attribution with Temporal Data in Reddit

Abstract

Context: The practicality brought by the use of smartphones has resulted, in recent years, in greater interaction through online social networks. Problem: Social networks can influence users both positively and negatively, one of the negative impacts is the spread of fake news. In this context, identifying the correct source of information or whether the information is true becomes an extremely relevant activity. Solution: This paper presents an approach for authorship attributions that combines text mining and temporal analysis techniques. IS Theory: This work is under the Social Network Theory, in particular, the user interaction through a forum network model, in which each post creates a comment thread and the user can reply or not inside the thread. Method: This work is a controlled experiment and it aims to extend a previous case study that used a classification between two and ten authors. The results were validated through a quantitative approach. Summary of Results: Among 10 authors, classification results had more than 97% of accuracy with chars feature having more than 99% of accuracy, among 100 authors all features presented more than 70% of accuracy. Contributions and Impact in the IS area: The main contribution of this works is to validate the authorship attribution in a big data context, using significant features and a robust classifier model.