The Influence of Text Pre-processing on Plagiarism Detection

International Conference RANLP 2009 - Borovets, Bulgaria, pages 55–59 The Influence of Text Pre-processing on Plagiarism Detection Zdenek Ceska Computer Science and Engineering Faculty of Applied Sciences University of West Bohemia Pilsen, 306 14, Czech Republic zceska@kiv.zcu.cz Chris Fox School of Computer Science and Electronic Engineering University of Essex Colchester, CO4 3SQ, United Kingdom foxcj@essex.ac.uk Abstract This paper explores the influence of text pre- processing techniques on plagiarism detection. We examine stop-word removal, lemmatization, number replacement, synonymy recognition, and word generalization. We also look into the in- fluence of punctuation and word-order within N-grams. All these techniques are evaluated according to their impact on F1-measure and speed of execution. Our experiments were per- formed on a Czech corpus of plagiarized docu- ments about politics. At the end of this paper, we propose what we consider to be the best com- bination of text pre-processing techniques. Keywords Plagiarism, Copy Detection, Natural Language Processing, Stop-words, Lemmatization, Synonymy, WordNet, Thesaurus. 1 Introduction Recently, there has been much interest in automatic plagiarism. Written-text plagiarism is a wide-spread problem which many organizations have to deal with. Various methods are used in this field, such as SCAM [9] and Kopi [5]. SVDPlag [1] is another technique whose performance outperforms all these other meth- ods. Text pre-processing can have a significant influence on the performance of many Natural Language Pro- cessing (NLP) tasks, including plagiarism detection. Although many studies on pre-processing techniques have been performed for applications such as text cat- egorization [10], it is appropriate to look at such pre- processing techniques again when when considering a new application. Plagiarism detection is a distinct field that should be given particular attention, as it may be appropriate to apply a wide range of pre- processing techniques. Various pre-processing have different effects, some improve the accuracy, some just decrease the time requirements, and some do both. This paper aims to clarify the influence of text pre- processing on this task when using the SVDPlag method. 2 Pre-processing Techniques Plagiarism detection can employ various pre- processing techniques in order to improve the accuracy or decrease the number of features that need to be processed. Figure 1 shows the text pre-processing step-by-step. The most essential block is Tokenization, which ex- tracts single words from the structured text. Punctu- ation marks can be extracted if they are required by other processes. The other blocks represent optional techniques that can be applied if the user wishes. Stop-word removal (STR) Lemmatization (LM) Synonym recognition (SYR) Word�generalization (WG) Number replacement (NMR) Plagiarism Detection WordNet Lemma dictionary Corpus Stop-word dictionary Tokenization Fig. 1: Text pre-processing scheme Below we describe each technique in detail. The im- pact of these techniques on accuracy and processing time is given in Section 4, where we also explore the impact of maintaining word-order, and the boundaries between sentences and phrases, as marked by punctu- ation. 2.1 Stop-word Removal (STR) Stop-word removal is a fundamental pre-processing approach that removes common words. Its primary use is to prevent the following processing being …

Content truncated to 500 words