Intrinsic Plagiarism Detection using N-gram Classes

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1459–1464, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Intrinsic Plagiarism Detection using N-gram Classes Imene Bensalem MISC Lab Constantine 2 University, Algeria bens.imene@gmail.com Paolo Rosso NLE Lab PRHLT Research Center Universitat Politècnica de València, Spain prosso@dsic.upv.es Salim Chikhi MISC Lab Constantine 2 University, Algeria slchikhi@yahoo.com Abstract When it is not possible to compare the suspi- cious document to the source document(s) plagiarism has been committed from, the evi- dence of plagiarism has to be looked for in- trinsically in the document itself. In this pa- per, we introduce a novel language- independent intrinsic plagiarism detection method which is based on a new text repre- sentation that we called n-gram classes. The proposed method was evaluated on three pub- licly available standard corpora. The obtained results are comparable to the ones obtained by the best state-of-the-art methods. 1 Introduction and Related Works Intrinsic plagiarism detection is an essential alternative in situations where the plagiarism source does not have a digital version, e.g. an old book, or the plagiarized text was directly written by another author without copying from any source, e.g. the case of a student who asked someone else to write for him parts of his essay or thesis. Hence, the task of detecting plagiarism intrinsically is to identify, in the given suspicious document, the fragments that are not consistent with the rest of the text in terms of writing style. The automatic analysis of the writing style is an important component of many NLP applica- tions. For some of them, when analyzing the style, a document is considered as a whole, which is the case of the authorship identification (Stamatatos, 2009a) and the authorship verifica- tion (Koppel and Seidman, 2013). For other ap- plications, a document is perceived as a set of fragments, for each of them the writing style needs to be analyzed individually. Examples of such applications include: paragraph authorship clustering (Brooke and Hirst, 2012), authorial segmentation of multi-author documents (Akiva and Koppel, 2013), detection of stylistic incon- sistencies between consecutive paragraphs (Graham et al., 2005) and plagiarism direction identification (Grozea and Popescu, 2010). For intrinsic plagiarism detection, it is crucial to analyze the writing style at fragments level. However, the majority of methods tend to ana- lyze the whole document writing style as well. Indeed, intrinsic plagiarism detection puts to- gether, in one research problem, many difficul- ties that are not present, or present separately, in the aforementioned related problems. Its main difficulties are listed below. In contrast to multi-author documents related problems, the number of authors in the suspi- cious documents is unknown, i.e., it might be one author if the document is plagiarism-free or many unknown authors if it contains plagiarism. Unlike the authorship attribution and verifica- tion, where the examined text and the potential author text are separate (and hence their writing styles could be readily characterized and com- pared), these two parts are both merged …

Content truncated to 500 words