Author profile prediction using pivoted unique term normalization

Author profiling is a text classification technique, which is used to predict the demographic characteristics of the authors by analyzing their written texts. Author Profiling became popular in several information technology enabled applications such as marketing, forensic analysis, psychology and entertainment. In reviews domain, most of the authors write reviews on several products without specifying their details. In this context, Author Profiling is helpful to know about the characteristics of the authors like gender, age, native language, educational background, location and personality traits by analyzing their written texts.  Most of the approaches for Author Profiling used various features like lexical features, content based features, structural features, syntactic features and semantic features to differentiate the writing style of the authors. These approaches of Author Profiling suffer from high dimensionality of features and fail to capture the relationship between the features. In this paper, a new approach is proposed to address the high dimensionality feature space problem by aggregating the term weights to find the weight of a document against the profile of the authors. The proposed approach was experimented on reviews domain to predict the gender and age group of the authors using accuracy as a measure.