Instance Based Authorship Attribution for Kannada Text Using Amalgamation of Character and Word N-grams Technique

Abstract

AbstractAuthorship Attribution is the task of identifying a true author of a given text from a set of suspected authors stylometry features play a vital role in recognizing the right author, it includes lexical and syntactic features. N-gram is one of the popular techniques used to extract syntactic features from the text. The main objective of this work is to use both lexical and syntactic features on a Kannada text and compare the performance of both approaches using different machine learning algorithms. The Kannada language is spoken by the Indian southern state Karnataka. Even though we can see major works in text processing, Authorship Attribution is in a tender state. Researches have been carried out on handwritten Kannada documents but not on digital text. Char n-gram, word n-gram and the combination of these two known as Amalgamation technique are used as syntactic features to extract the writing style of an author. The results show that Support Vector Machine algorithm outperform with 94% and 60% accuracy using N-grams and lexical features respectively.KeywordsAuthorship attributionDecision treeInstance based approachMachine learning algorithmsNaïve bayesProfile based approachRandom forest and support vector machine