A Novel Document Representation Approach for Authorship Attribution
Sreenivas Mekala, Raghunadha Tippireddy, Vishnu Vardhan Bulusu, JNTUH · International journal of intelligent engineering and systems · 2018
The rapidly growing data in the web result in stolen, unidentified and fraudulent data.Identification of such data is of a prime objective for forensic departments, researchers and governments.In this context, authorship analysis is very useful to reveal the truth by analyzing the text.Authorship analysis is observing the properties of a text to predict authorship of a document.Stylometry is the root for authorship analysis, which is a linguistic research field that exploits the machine learning techniques as well as knowledge of statistics.Authorship Attribution is a type of authorship analysis technique, which is aimed at recognizing the author of an anonymous text within a closed set of authors or subjects.Most of the researchers in Authorship Attribution approaches proposed various set of stylistic features to differentiate the authors based on style of writing.It was observed from the literature the accuracy of author prediction was not satisfactory with stylistic features.In this paper, the experimentation carried out with various stylistic features, feature selection measures and term weight measures identified in various text processing domains to predict the author of a new document.A new document representation approach is proposed to improve the prediction accuracy of author prediction.In the proposed approach the documents were represented with the weights of the documents specific to author group of documents.The results show that the proposed approach obtained good accuracies when compared with the results of stylistic features, feature section measures, term weight measures and most of the existing approaches.