Recognition on Online Social Network by user's writing style

Résumé

Compromising legitimate accounts is the most popular way of disseminating fraudulent content in Online Social Networks (OSN). To address this issue, we propose an approach for recognition of compromised Twitter accounts based on Authorship Verification. Our solution can detect accounts that became compromised by analysing their user writing styles. This way, when an account content does not match its user writing style, we affirm that the account has been compromised, similar to Authorship Verification. Our approach follows the profile-based paradigm and uses N-grams as its kernel. Then, a threshold is found to represent the boundary of an account writing style. Experiments were performed using two subsampled datasets from Twitter. Experimental results showed the developed model is very suitable for compromised recognition of Online Social Networks accounts due to the capacity of recognizing user styles over 95% accuracy for both datasets.