A Text Alignment Corpus for Persian Plagiarism Detection Fatemeh - PowerPoint PPT Presentation

Persian Plagdet 2016 A Text Alignment Corpus for Persian Plagiarism Detection Fatemeh Mashhadirajab, Mehrnoush Shamsfard Fatemeh Shafiee, Chakaveh Saedi Razieh Adelkhah, NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

2 /29 Outline  Introduction  Text Alignment Corpus Construction  Strategies For Plagiarisms Types  Dataset Statistics  Conclusions  References NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

3 /29 Introduction A taxonomy of plagiarism [2]

4 /29 Text Alignment Corpus Construction  Data Source Preparation  Documents Clustering  Set of Suspicious and source Documents  source and suspicious document pairs selection  Source Documents Segmentation  Segment Extraction  Segment Obfuscation  Obfuscated Segment Insertion NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

5 /29 Text Alignment Corpus Construction Data Source Preparation   articles or theses in the fields of computer science and engineering & electrical engineering o 4,500 documents from Wikipedia articles o 1,500 documents from articles and theses available from online stores NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

5 /29 Text Alignment Corpus Construction Data Source Preparation   Our corpus contains 11,089 documents NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

6 /29 Text Alignment Corpus Construction  Documents Clustering NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

7 /29 Text Alignment Corpus Construction  Set of Suspicious and source Documents NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

8 /29 Text Alignment Corpus Construction Suspicious Documents Source Documents  Set of Suspicious and source Documents They are randomly selected from each cluster NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

9 /29 Text Alignment Corpus Construction  source and suspicious document Suspicious Documents Source Documents pairs selection NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

9 /29 Text Alignment Corpus Construction  source and suspicious Suspicious Documents Source Documents susp document pairs selection Similarity Detection system if the similarity < 50% NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

9 /29 Text Alignment Corpus Construction  source and suspicious Suspicious Documents Source Documents susp src document pairs selection Similarity Detection system if the similarity < 50% NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

10/29 Text Alignment Corpus Construction  Source Documents Segmentation Suspicious Documents Source Documents src Similarity Detection system if the similarity < 50% NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

11/29 Text Alignment Corpus Construction  Segment Extraction Suspicious Documents Source Documents Similarity Detection system src if the similarity < 50% NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

12/29 Text Alignment Corpus Construction  Segment Obfuscation Suspicious Documents Source Documents Similarity Detection system src if the Segment similarity < 50% Obfuscation NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

13/29 Text Alignment Corpus Construction  Obfuscated Segment Insertion Suspicious Documents Source Documents susp Similarity Detection system src if the Segment similarity < 50% Obfuscation NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

14/29 Strategies For Plagiarisms Types  Exact Copy  Near Copy  Modified Copy  Text Manipulation (Paraphrasing)  Text Manipulation (Summarizing)  Automatic Translation  Manual Translation  Cyclic Translation  Idea Adoption (semantic-based meaning) NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

15/29 Strategies For Plagiarisms Types  Exact Copy Segment Obfuscation src susp NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

16/29 Strategies For Plagiarisms Types  Near Copy Segment Obfuscation src susp Insertion deletion substitution sentence split or join NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

17/29 Strategies For Plagiarisms Types  Modified Copy Segment Obfuscation src susp the Persian sentence understanding and generation system introduced by Adelkhah et al. [7] semantic representation sentence production based on semantic (sentence understanding) representation (sentence generation) NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

18/29 Strategies For Plagiarisms Types  Text Manipulation (Paraphrasing) Segment Obfuscation src susp the Persian sentence understanding and generation system introduced by Adelkhah et al. [7] Each word is replaced with a synonym retrieved from FarsNet or FavaNet NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

19/29 Strategies For Plagiarisms Types Text Manipulation (Summarizing)  Segment Obfuscation src susp Persian summarizer introduced by Shafiee et al. [6] NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

20/29 Strategies For Plagiarisms Types  Automatic Translation Segment Obfuscation src susp Google translate Spell checker Hunspell Persian to English Persian English NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

21/29 Strategies For Plagiarisms Types  Manual Translation Segment Obfuscation Translateion src susp Persian English NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

22/29 Strategies For Plagiarisms Types  Cyclic Translation Segment Obfuscation Negar spell checker src susp Google translate Hunspell English Persian Google translate NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

23/29 Strategies For Plagiarisms Types  Idea Adoption (semantic-based meaning) Segment Obfuscation src susp NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

24/29 Dataset Statistics documents 11089 plagiarism cases 11603 Document purpose languages fa source documents 48% suspicious documents with plagiarism 28% w/o plagiarism 24% Document length short (<10 pages 1 ) 64% medium (10-100 pages) 35% long (>100 pages) 1% Plagiarism per document hardly (<20%) 25% medium (20%-50%) 20% much (50%-80%) 26% entirely (>80%) 29% Case length short (<1k characters) 37% medium (1k-3k characters) 55% long (>3k characters) 8% NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

25/29 Conclusion  This article describes a methodology for building a Persian corpus for evaluating plagiarism detection systems.  This corpus is in PAN format.  This corpus is a variety of plagiarism types in large volume are created.  To produce this corpus, the focus is on the simulation of different types of plagiarism  Different strategies are employed to create obfuscation in each plagiarism category NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University

26/29 References 1. Asghari, H., Mohtaj, S., Fatemi, O., Faili, H., Potthast, M. and Rosso, P. 2016. Overview of the PAN@FIRE2016 Shared Task on Persian Plagiarism Detection and Text Alignment Corpus Construction, Notebook Papers of FIRE 2016, FIRE-2016, CEUR-WS.org. 2. Alzahrani, M., Salim, N. and Abraham, A. 2012. Understanding plagiarism linguistic patterns, Textual features, and detection Methods. IEEE Trans. SYSTEMS, MAN, AND CYBERNETICS — PART C: APPLICATIONS AND REVIEWS , vol. 42, no. 2. 3. Potthast, M., Stein, B. and et.al. 2010. An Evaluation Framework for Plagiarism Detection. Proceedings of the 23rd International Conference on Computational Linguistics, COLING 2010 Beijing,_c ACL . 4. Shamsfard, M., and Kiani, S., and Shahedi, Y. STeP-1: Standard Text Preparation for Persian Language. CAASL3 Third Workshop on Computational Approaches to Arabic Script- Languages . 5. Makrehchi, M. and Kamel, M. 2004. A fuzzy set approach to extracting keywords from abstracts. North American Fuzzy Information Processing Society- NAFIPS 2003 , Banf, Canada. 6. Shafiee, F. and Shamsfard, M. 2015. The automatic Persian summarizer. The 20st Computer Society of Iran computer conference. NLP Research Lab, Faculty of Computer Science and Engineering, Shahid Beheshti University