Article Preview
Top1. Introduction
Plagiarism is a challenging problem for the academic community. A large collection of documents is freely available on the Web and modern search engines, for example Google, enable authors to easily find texts on different topics from the Web. With the help of sophisticated word processing tools, authors can easily reuse such existing texts in their own work. The diverse sources of information and ease of text reuse make it difficult to detect manually whether the author of a document has plagiarized or not. To account for this problem, different software systems, for example Turnitin1 and iThenticate2, have been developed to automatically detect plagiarism. Such systems can easily detect verbatim copy of text, but identifying plagiarism in heavily-paraphrased documents (or the cases of copying others ideas) is still a challenging task. Different plagiarism detection techniques have been proposed in literature, which are broadly divided into extrinsic (also called external) and intrinsic approaches. The extrinsic plagiarism detection approach compares the similarity of a given suspicious document against a reference collection (Yerra & Ng, 2005; Ceska, 2008; Elhadi & Al-Tobi, 2009). This approach uses different document-similarity measures, including for example, cosine similarity (Zechner, Muhr, Kern, & Granitzer, 2009; Murugesan, Jiang, Clifton, Si, & Vaidya, 2010), Jaccard similarity (Barrón-Cedeño, Basile, Esposti, & Rosso, 2010) and edit distance (Cohen, Ravikumar, & Fienberg, 2003). The intrinsic approach, on the other hand, analyses the suspicious document in isolation, without considering any source collection, to find plagiarism cases (Stein, Lipka, & Prettenhofer, 2011; Eissen, Stein, & Kulig, 2007). The latter makes use of writing-style analysis, including readability scores, to uncover anomalies in the text.
Bulk of the existing work on plagiarism detection focuses on English mostly disregarding other languages, for example, Arabic. This is an important gap because Arabic is a widely spoken language in the world and it constitutes a major portion of documents on the Web. More importantly, a paradigm shift in different Arabic-lead educational institutions towards e-learning systems in the recent years has made the plagiarism problem ever challenging. This is because in an e-learning setup, students normally have an open access to a large repository of documents on the Web thereby increasing the chances of simple 'copy and paste' approach in their work. To address this issue, we developed an automatic plagiarism detection system for Arabic language, particularly for students' essays. The performance of the overall system is evaluated using an indigenous corpus, which was developed specifically for this purpose. In this article, we describe the proposed plagiarism detection system, the developed corpus and evaluation results. The work presented here has two main contributions to the existing work on plagiarism detection and similarity computation. First, a labelled corpus for plagiarism detection in Arabic is built, which uses the actual cases of plagiarism unlike others where plagiarism cases are artificially created; see for example, Clough and Stevenson (2011). This corpus can be used, for example, in a shared task to test and compare different approaches/systems for plagiarism detection. Second, the plagiarism detection system could be used in an academic setting to identify plagiarism cases, for example, in students' work on Arabic essays or assignments.