An intrinsic authorship verification technique for compromised account detection in social networks

The proliferation of social networks resulted in a remarkable increase in their popularity empowering users to create, share and exchange content for interaction and communication among them. However, these have also opened new avenues for malicious and unauthorized use. This paper presents an intrinsic profiling-based technique for the assessment of authorship verification and its application toward detection of compromised accounts. For the same, efficiency of different textual features has been examined. Four categories of features, namely content free, content specific, stylometric and folksonomy, are extracted and evaluated. Experiments are performed with 3057 twitter users taking 4000 latest tweets for each user. Various feature selection techniques are used to rank and select optimal features for each user which are further combined using a popular rank aggregation technique called BORDA. The problem of authorship verification in this paper is studied as a unary classification problem. Performance of various one-class classifiers, namely Local Outlier Factor, Isolation Forest and One-Class SVM, is analyzed on the basis of different evaluation metrics. Experimental results depict that OCC-SVM with rbf kernel outperformed other one class classifiers attaining an average F-score of 87.29% and Matthews Correlation Coefficient of 74.42% under varied parameter settings.