An effective text plagiarism detection system based on feature selection and SVM techniques

Text plagiarism has greatly spread in the recent years, it becomes a common problem in several fields such as research manuscripts, textbooks, patents, academic circles, etc. There are many sentence similarity features were used to detect plagiarism, but each of them is not discriminative to differentiate the similarity cases. This causes the discovery of lexical, syntactic and semantic text plagiarism types to be a challenging problem. Therefore, a new plagiarism detection system is proposed to extract the most effective sentence similarity features and construct hyperplane equation of the selected features to distinguish the similarity cases with the highest accuracy. It consists of three phases; the first phase is used to preprocess the documents. The second phase is depended on two paths, the first path is based on traditional paragraph level comparison, and the second path is based on the computed hyperplane equation using Support Vector Machine (SVM) and Chi-square techniques. The third phase is used to extract the best plagiarized segment. The proposed system is evaluated on several benchmark datasets. The experimental results showed that the proposed system obtained a significant superiority in the performance compared to the systems with a higher ranking in the recent years. The proposed system achieved the best values 89.12% and 92.91% of the Plagdet scores, 89.34% and 92.95% of the F-measure scores on the complete test corpus of PAN 2013 and PAN 2014 datasets, respectively.