Author Gender Identification from Text

ABSTRACT: The identification of an author's gender from a text has become a popular research area within the scope of text categorization. The number of users of social network applications based on text, such as Twitter, Facebook and text messaging services, has grown rapidly over the past few decades. As a result, text has become one of the most important and prevalent media types on the Internet. This thesis aims to determine the gender of an author from an arbitrary piece of text such as, for example a journal article or email. This field of research has garnered the interest of the researchers for the reason that some people fake their gender in text-based Internet forensics. The psychology of linguistic indicates how closely the words and writing styles people use correlate with their gender. Various feature sets have been used by researchers in recent decades to identify the gender of an author; however, identifying feature sets remains a research obstacle. In this dissertation, five feature sets were selected to prepare a feature space for the gender identification problem. The features in these sets included character-based features, word-based features, syntactic-based features, structure-based features and the function words that an author used in a text. Two state-of-the-art machine learning algorithms were considered for the author gender identification problem, based on the proposed feature space in this thesis. Weka (data mining software) was used to design a support vector machine classifier and a Bayesian logistic regression classifier. The reason for choosing these two classifiers was that support vector machine and Bayesian logistic regression are the most powerful classifiers for text mining. An Enron email dataset, which is available to researchers on the Internet, was used in the training and testing phases during experiments to provide sufficient data for the classification process. Keywords: Machine Learning, classifier, psychology linguistic, Support Vector Machine, Bayesian logistic regression, gender identification. …………………………………………………………………………………………………………………………