A New Document Representation Approach for Gender Prediction Using Author Profiles

Abstract

Author Profiling is used to predict the demographic characteristics like age, gender, country, nativity language, and educational background of anonymous text by analyzing their style of writing. Several researchers proposed different types of features like lexical, character based, content based, syntactic, topic specific, structural features, and readability features to discriminate the style of writing of the authors for Author Profiling. The representation of a document with extracted features is one of the important tasks in Author Profiling. In Author Profiling approaches, most of the researchers used the bag-of-words model for document representation. This paper concentrates on the alternative document representation to increase the performance of Author Profiling system. In this work, a new document representation model is proposed and compared the proposed model with existing document representation models like BOW and SOA. The proposed model is evaluated on the reviews dataset for predicting the gender of the authors using various machine learning classifiers. The proposed approach results were promising than most of the existing approaches for Author Profiling.