Author Profiling Approach for Location Prediction

Abstract

Author Profiling is a type of text categorization technique which is used to predict the profiling characteristics of the authors like gender, age, and location by observing their writing styles in the text. Various researchers proposed several types of techniques to predict the demographic characteristics like gender and age of the authors from different types of datasets. Few researchers show interest on location prediction of the texts. In this work, we concentrated on prediction of the location of the authors. Most of the existing approaches in Author Profiling concentrated on extraction of different types of stylistic features to discriminate the writing style of the text. In this work, the experimentation carried out with content-based features like most frequent terms and a document representation technique profile specific document weighted approach is used to predict the location of the authors. In this approach, a new term weight measure is proposed to compute the term weights of the features. These term weights of most frequent terms were used to compute the document weights specific to every location profile group. These document weights were exploited to represent the document vectors for generating the classification model. The obtained results were good when compared with existing approaches for location prediction.