Detecting Text Reuse with Modified and Weighted N-grams

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 54–58, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics Detecting Text Reuse with Modified and Weighted N-grams Rao Muhammad Adeel Nawab∗, Mark Stevenson∗ and Paul Clough† ∗Department of Computer Science and †iSchool University of Sheffield, UK. {r.nawab@dcs, m.stevenson@dcs, p.d.clough@} .shef.ac.uk Abstract Text reuse is common in many scenarios and documents are often based, at least in part, on existing documents. This paper reports an ap- proach to detecting text reuse which identifies not only documents which have been reused verbatim but is also designed to identify cases of reuse when the original has been rewrit- ten. The approach identifies reuse by compar- ing word n-grams in documents and modifies these (by substituting words with synonyms and deleting words) to identify when text has been altered. The approach is applied to a cor- pus of newspaper stories and found to outper- form a previously reported method. 1 Introduction Text reuse is the process of creating new docu- ment(s) using text from existing document(s). Text reuse is standard practice in some situations, such as journalism. Applications of automatic detection of text reuse include the removal of (near-)duplicates from search results (Hoad and Zobel, 2003; Seo and Croft, 2008), identification of text reuse in journal- ism (Clough et al., 2002) and identification of pla- giarism (Potthast et al., 2011). Text reuse is more difficult to detect when the original text has been altered. We propose an ap- proach to the identification of text reuse which is intended to identify reuse in such cases. The ap- proach is based on comparison of word n-grams, a popular approach to detecting text reuse. However, we also account for synonym replacement and word deletion, two common text editing operations (Bell, 1991). The relative importance of n-grams is ac- counted for using probabilities obtained from a lan- guage model. We show that making use of modified n-grams and their probabilities improves identifica- tion of text reuse in an existing journalism corpus and outperforms a previously reported approach. 2 Related Work Approaches for identifying text reuse based on word-level comparison (such as the SCAM copy de- tection system (Shivakumar and Molina, 1995)) tend to identify topical similarity between a pair of doc- uments, whereas methods based on sentence-level comparison (e.g. the COPS copy detection sys- tem (Brin et al., 1995)) are unable to identify when text has been reused if only a single word has been changed in a sentence. Comparison of word and character n-grams has proven to be an effective method for detecting text reuse (Clough et al., 2002; Cede˜no et al., 2009; Chiu et al., 2010). For example, Cede˜no et al. (2009) showed that comparison of word bigrams and tri- grams are an effective method for detecting reuse in journalistic text. Clough et al. (2002) also applied n-gram overlap to identify reuse of journalistic text, combining it with other approaches such as sentence alignment and string matching algorithms. Chiu et al. (2010) compared n-grams …

Content truncated to 500 words