A Stylistic Features Based Approach for Author Profiling

Abstract

Author profiling is the area of data-driven computational linguistics which is used to predict the demographic profiles such as age, gender, location, nativity language of authors by processing the textual content of their written text. Author profiling is so popular in recent days because of its potential applications such as marketing analysis studies, forensic linguistics, security and literary research. The majority of approaches for Author Profiling tend to focus on the content of the texts. Some researchers argue that focusing on structure of a text is more effective than content of a text. A popular research area called Stylometry is used by these researchers. Stylometry is used for finding the stylistic differences among the texts. In Author Profiling, various researchers proposed several types of stylistic features for differentiating the style of writing of the authors. The main concentration of this work is on feature engineering, the development, evaluation, and application of the feature set in the context of machine learning techniques to author profiling. In this work, we focused on gender prediction of the authors of reviews dataset. It was observed that stylistic features achieved best accuracies for predicting gender when compared with state-of-the-art approaches in many different scenarios, especially when combined with other features.