CEAI: CCM-based email authorship identification model

Abstract

In this paper we present a model for email authorship identification (EAI) by employing a Cluster-based Classification (CCM) technique. Traditionally, stylometric features have been successfully employed in various authorship analysis tasks; we extend the traditional feature set to include some more interesting and effective features for email authorship identification (e.g., the last punctuation mark used in an email, the tendency of an author to use capitalization at the start of an email, or the punctuation after a greeting or farewell). We also included Info Gain feature selection based content features. It is observed that the use of such features in the authorship identification process has a positive impact on the accuracy of the authorship identification task. We performed experiments to justify our arguments and compared the results with other base line models. Experimental results reveal that the proposed CCM-based email authorship identification model, along with the proposed feature set, outperforms the state-of-the-art support vector machine (SVM)-based models, as well as the models proposed by Iqbal et al. (2010, 2013) [1,2]. The proposed model attains an accuracy rate of 94% for 10 authors, 89% for 25 authors, and 81% for 50 authors, respectively on Enron dataset, while 89.5% accuracy has been achieved on authors’ constructed real email dataset. The results on Enron dataset have been achieved on quite a large number of authors as compared to the models proposed by Iqbal et al. [1,2].

Keywords

• CCM
• Email authorship analysis
• Feature selection
• Machine learning
• Stylometric features