Novel authorship verification model for social media accounts compromised by a human

Abstract

Social media networks usage is spreading but accompanied by a new shape of the social engineering attacks in which users’ accounts are compromised by attackers to spread malicious messages for different purposes. To overcome these attacks, authorship verification, a classification problem for classifying a text, whether it belongs to a user or not, is needed. Moreover, the verification must be accurate and fast. Herein, an authorship verification model proposed. The model uses XGBoost, as a preprocessor, to discover functional features of the text message, which ranked using MCDM methods to build a classification model. Twitter messages are used to test the model; however, any social media’s data might be used. The suggested model was evaluated against a crawled dataset from Twitter composed of 16124 tweets with 280 characters. The proposed method achieved F-score over 0.94.