PAN Arabic Intrinsic Plagiarism Detection Shared Task Corpus

Evaluation corpus for ARAbic INtrinsic plagiarism detection (InAra Corpus) This corpus has been used in AraPlagDet 2015 shared task More details could be found in : https://araplagdet.misc-lab.org/ or https://pan.webis.de/fire15/pan15-web/index.html <strong>I. SYNOPSIS </strong> InAra corpus comprises 2048 documents; 80% of them contain passages borrowed from other documents to simulate documents that contain plagiarized fragments. The corpus involves 2 parts: Training and test. <strong>II. DESCRIPTION </strong> Each part of the corpus (training and test) consists mainly of 2 datasets: textual files and XML files. The textual files represent the suspicious documents i.e., the documents that contain artificial plagiarism; and the XML files are the plagiarism annotation i.e. they provide for each plagiarized passage its starting offset in the suspicious document and its length (offset and length are both expressed in characters). A suspicious document file and its plagiarism annotation file share the same name. <strong>III. PURPOSE </strong> The purpose of InAra corpus is to evaluate automatic plagiarism detection methods, notably methods of the intrinsic approach. This approach consists in uncovering the plagiarized passages on the basis of the writing style inconsistency in a given suspicious document. As opposed to the external approach, the intrinsic approach does not necessitate any comparison of the suspicious document against the potential sources of plagiarism. Hence, InAra corpus is not appropriate for the evaluation of the external plagiarism detection because the source of plagiarism are not provided. It should be noted that some documents in InAra corpus contain religious quotations (e.g., Quran and Hadith). These quotations have a peculiar writing style and then a simple intrinsic plagiarism detection software can consider them as plagiarism. However, quotations are not plagiarism, and they are not annotated in the XML files in InAra. Hence, it is an important feature for the plagiarism detection systems evaluated on InAra to not consider religious quotations as plagiarism cases unless they appear as part of a larger plagiarism case. <strong>IV. BUILDING METHODS </strong> The documents that compose InAra corpus do not contain actual plagiarism cases. They are rather artificial suspicious documents in which plagiarism was created automatically by a software that takes fragments of text from one or more sources documents and inserts them in another one according to a set of parameters, namely the percentage of plagiarism and the plagiarized passages lengths. This building method is the same used to construct PAN 2009-2011 corpora of plagiarism detection (see http://pan.webis.de for more information on PAN competition and its corpora). <strong>V. LANGUAGE AND ENCODING </strong> All the textual documents of this corpus are written in Arabic language and encoded in UTF-8 without BOM. <strong>VI. SOURCES OF TEXTS </strong> Texts used to build this corpus, either suspicious documents or the inserted passages, are taken mainly from the open library Arabic Wikisource (http://ar.wikisource.org), one of Wikimedia Foundation projects. A few numbers of documents were taken from other websites, namely: Create your own country blog: http://diycountry.blogspot.com Corpus of Classical Arabic (KSUCCA): http://ksucorpus.ksu.edu.sa Islamic book web site: http://www.islamicbook.ws <strong>VII. COPYRIGHT AND AVAILABILITY </strong> We were very careful to build the corpus with copyright-free texts only, to be able to make it publicly available without any sort of problems with texts owners. <strong>VIII. HOW TO CITE THE CORPUS ?</strong> If you publish a paper about your experimentations using InAra corpus, please cite the following paper: Bensalem, I., Boukhalfa, I., Rosso, P., Abouenour, L., Darwish, K., &amp; Chikhi, S.: Overview of the AraPlagDet PAN@FIRE2015 Shared Task on Arabic Plagiarism Detection. In P. Majumder, M. Mitra, M. Agrawal, &amp; P. Mehta (Eds.), Post Proceedings of the Workshops at the 7th Forum for Information Retrieval Evaluation (FIRE 2015), Gandhinagar, India, December 4-6, CEUR proceedings vol. 1587 (pp. 111–122). CEUR-WS.org (2015). We encourage you to compare your method tested on InAra with the methods of AraPlagDet competition described in the paper above. Additional information on the corpus building are in the papers: Bensalem, I., Rosso, P., Chikhi, S.: A New Corpus for the Evaluation of Arabic Intrinsic Plagiarism Detection. In: Forner, P., Müller, H., Paredes, R., Rosso, P., and Stein, B. (eds.) CLEF 2013, LNCS, vol. 8138. pp. 53–58. Springer, Heidelberg (2013). Bensalem, I., Rosso, P., Chikhi, S.: Building Arabic Corpora from Wikisource. 10th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA’13),May 27-30 Fes/Ifran, Morocco (2013).IEEE. You may wish to compare the results of your experiments with the result of the following papers that used InAra corpus: Bensalem I, Rosso P, Chikhi S (2019) On the use of character n-grams as the only intrinsic evidence of plagiarism. Language Resources and Evaluation 53:363–396. doi: 10.1007/s10579-019-09444-w Mahgoub AY, Magooda A, Rashwan M, et al (2015) RDI System for Intrinsic Plagiarism Detection (RDI_RID), Working Notes for PAN-AraPlagDet at FIRE 2015. In: Majumder P, Mitra M, Agrawal M, Mehta P (eds) Post Proceedings of the Workshops at the 7th Forum for Information Retrieval Evaluation (FIRE 2015), Gandhinagar, India, December 4-6, CEUR proceedings vol. 1587. CEUR-WS.org, pp 129–130 <strong>IX. WARNING </strong> It should be noted that the Arabic texts may contain quotations from the Quran and the Hadith; and due to the fact that text insertion is automatic and in random positions, it is possible that the plagiarized text is inserted unintentionally between Quranic verses or sentences of a Hadith cited in a document. Hence, the inserted passages may alter the meaning of the original text. For these reasons, this corpus must not be used outside the purpose for which it was built. Examples of the inappropriate use include using the corpus documents as a source of knowledge or distributing them without mentioning that they contain borrowed texts. If you are not interested in plagiarism detection and you are retaining the corpus because it contains books you want to read, then this corpus is not the right source. Please, you should refer to the sources mentioned in Section VI where you can find the original content of the books you are looking for. We emphasize that we are not responsible for the results of any use of this corpus other than the evaluation of the intrinsic plagiarism detection methods. <strong>X. CONTACT US</strong> We will be happy to hear from you about your experience in using InAra corpus. Please do not hesitate to contact us with the following email address: [e-mail] Imene Bensalem¹, Paolo Rosso², Salim Chikhi¹ ¹MISC Lab. Constantine 2 university, Algeria ²PRHLT, Universitat Politècnica de València, Spain

阿拉伯语内在式抄袭检测评测语料库（InAra Corpus）。该语料库曾用于AraPlagDet 2015共享任务。更多详细信息可访问：https://araplagdet.misc-lab.org/ 或 https://pan.webis.de/fire15/pan15-web/index.html

**一、数据集概况**
InAra语料库共包含2048份文档，其中80%含有从其他文档中摘抄的片段，以模拟存在抄袭片段的待检测文档。该语料库分为训练集与测试集两个部分。

**二、语料描述**
语料库的每个部分（训练集与测试集）均主要包含两类数据集：文本文件与XML文件。其中，文本文件即为待检测文档——即含人工模拟抄袭内容的文档；XML文件则为抄袭标注文件，用于为每一处抄袭片段提供其在待检测文档中的起始偏移量与长度（偏移量与长度均以字符数计量）。单份待检测文档文件与其对应的抄袭标注文件文件名完全一致。

**三、评测用途**
InAra语料库的核心用途为评测自动抄袭检测方法，尤其是内在式抄袭检测（intrinsic plagiarism detection）方法。该类方法通过分析单份待检测文档的写作风格不一致性，来识别其中的抄袭片段。与外在式抄袭检测方法不同，内在式方法无需将待检测文档与潜在抄袭源文本进行比对。因此，InAra语料库不适用于外在式抄袭检测方法的评测，因为语料未提供抄袭源文本。

需注意的是，InAra语料库中的部分文档包含宗教引用内容（如《古兰经》与圣训）。这类引用具有独特的写作风格，因此简易的内在式抄袭检测工具可能会将其误判为抄袭内容。但宗教引用并非抄袭内容，且InAra语料库的XML文件未对其进行标注。因此，基于InAra语料库评测的抄袭检测系统需具备一项重要特性：不得将宗教引用判定为抄袭，除非其作为更大范围抄袭片段的一部分出现。

**四、语料构建方法**
InAra语料库中的文档均不含真实抄袭案例，而是通过软件自动生成的人工待检测文档：该软件从一份或多份源文档中提取文本片段，并根据一组参数（即抄袭占比与抄袭片段长度）将其插入到目标文档中，以此模拟抄袭内容。该构建方法与PAN 2009-2011抄袭检测语料库的构建方法一致，如需了解PAN竞赛及其语料库的更多信息，请访问http://pan.webis.de。

**五、语言与编码格式**
本语料库的所有文本文档均为阿拉伯语，编码格式为无BOM的UTF-8。

**六、文本来源**
用于构建本语料库的文本，包括待检测文档与插入的抄袭片段，主要来自阿拉伯语维基文库（Arabic Wikisource，http://ar.wikisource.org）——维基媒体基金会的项目之一。少量文档来自以下其他网站：“创建你的国家”博客（http://diycountry.blogspot.com）、古典阿拉伯语语料库（KSUCCA，http://ksucorpus.ksu.edu.sa）以及伊斯兰图书网站（http://www.islamicbook.ws）。

**七、版权与获取方式**
本语料库的构建仅使用无版权的文本，以确保可公开获取且不会与文本所有者产生版权纠纷。

**八、语料引用规范**
若您发表使用InAra语料库开展实验的论文，请引用以下文献：
Bensalem, I., Boukhalfa, I., Rosso, P., Abouenour, L., Darwish, K., & Chikhi, S.: 《阿拉伯语抄袭检测PAN@FIRE2015共享任务概述》，发表于第七届信息检索评估论坛（FIRE 2015）工作坊论文集，印度甘地讷格尔，12月4-6日，CEUR会议论文集第1587卷，第111-122页，CEUR-WS.org（2015年）。

我们鼓励您将基于InAra语料库测试的方法与上述论文中提及的AraPlagDet竞赛参赛方法进行对比。关于语料库构建的更多信息可参考以下论文：
1. Bensalem, I., Rosso, P., Chikhi, S.: 《用于阿拉伯语内在式抄袭检测评测的新型语料库》，发表于Forner, P., Müller, H., Paredes, R., Rosso, P., and Stein, B. (eds.) CLEF 2013, LNCS, vol. 8138. 第53-58页，Springer出版社，海德堡（2013年）。
2. Bensalem, I., Rosso, P., Chikhi, S.: 《从维基文库构建阿拉伯语语料库》，发表于第10届ACS/IEEE国际计算机系统与应用会议（AICCSA’13），5月27-30日，摩洛哥菲斯/伊夫兰（2013年），IEEE出版社。

您可将基于InAra语料库的实验结果与以下使用该语料库的论文结果进行对比：
1. Bensalem I, Rosso P, Chikhi S (2019) 《仅使用字符n元语法作为内在式抄袭检测依据的研究》，《语言资源与评价》第53卷，第363-396页。doi: 10.1007/s10579-019-09444-w
2. Mahgoub AY, Magooda A, Rashwan M, et al (2015) 《用于内在式抄袭检测的RDI系统（RDI_RID）》，发表于FIRE 2015工作坊PAN-AraPlagDet论文集，收录于第七届信息检索评估论坛（FIRE 2015）工作坊论文集，印度甘地讷格尔，12月4-6日，CEUR会议论文集第1587卷，CEUR-WS.org，第129-130页。

**九、使用警告**
需注意的是，本语料库中的阿拉伯语文本可能包含《古兰经》与圣训的引用；由于文本插入为自动操作且插入位置随机，抄袭片段可能会被意外插入到《古兰经》经文或文档中引用的圣训语句之间。因此，插入的片段可能会改变原文本的含义。

基于上述原因，本语料库仅可用于其设计的评测用途，不得挪作他用。不当使用的示例包括将语料库文档作为知识来源，或在分发文档时未提及其中包含摘抄内容。若您并非用于抄袭检测，仅因语料库包含您想阅读的书籍而保留该语料库，则本语料库并非合适的来源。请参考第六部分提及的来源，以获取您所需书籍的原始内容。

我们特此声明：除用于内在式抄袭检测方法评测外，本语料库的任何其他使用所产生的后果均与我们无关。

**十、联系方式**
我们期待收到您关于使用InAra语料库的经验反馈，可通过以下邮箱联系我们：[e-mail]

伊梅内·本萨勒姆（Imene Bensalem）¹，保罗·罗索（Paolo Rosso）²
¹阿尔及利亚君士坦丁第二大学MISC实验室
²西班牙瓦伦西亚理工大学PRHLT实验室