External and Intrinsic Plagiarism Detection Using Vector Space Models

Plagiarism detection can be divided in external and intrinsic methods. Naive external plagiarism analysis suffers from computationally demanding full nearest neighbor searches within a reference corpus. We present a conceptually simple space partitioning approach to achieve search times sub linear in the number of reference documents, trading precision for speed. We focus on full duplicate searches while achieving acceptable results in the near duplicate case. Intrinsic plagiarism analysis tries to find plagiarized passages within a document without any external knowledge. We use several topic independent stylometric features from which a vector space model for each sentence of a suspicious document is constructed. Plagiarized passages are detected by an outlier analysis relative to the document mean vector. Our system was created for the first PAN competition on plagiarism detection in 2009. The evaluation was performed on the challenge’s development and competition corpora for which we report our results.