A New Term Weight Measure for Gender Prediction in Author Profiling

Abstract

Author profiling is used to predict the demographic characteristics such as gender, age, native language, location, and educational background of the authors by analyzing their writing styles. The researchers in author profiling proposed various features such as character-based, word-based, structural, syntactic, and semantic features to differentiate the writing styles of the authors. The existing approaches in author profiling used the frequency of a feature to represent the document vector. In this work, the experimented carried with various features with their frequency and observed that only frequency is not suitable to assign better discriminative power to the features. Later, a new supervised term weight measure is proposed to assign suitable weights to the terms and analyzed the accuracies with various machine learning algorithms. The experimentation carried out on review domain and the proposed supervised term weight measure obtained good accuracy for gender prediction when compared to existing approaches.