Plagiarism Detection in Obfuscated Documents Using an N-gram Technique

Plagiarism is considered as a major problem these days especially in case of academic institutions. Very often students present someone else’s work as their own and they are given credit for it. Therefore, we have to integrate plagiarism checking into the submission process which includes usage of plagiarism detection software. In this paper we focus on vulnerabilities of this software from the view of different obfuscations. We briefly introduce our statistical approach that should quite well detect obfuscated documents in case larger portion of the document is obfuscated. We show limits of this method and test effectiveness of n-gram technique used for detecting obfuscated documents. By playing with various parameters when dividing text into n-grams, we test success of n-gram method on similarity detection. We try to find the best parameters for the n-gram technique in order to achieve best detection results. We present the carried experiments and derive the conclusions.