Detecting High Obfuscation Plagiarism: Exploring Multi-Features Fusion via Machine Learning

Providing effective methods of identification of high-obfuscation plagiarism seeds presents a significant research problem in the field of plagiarism detection. The conventional methods of plagiarism detection are based on single type of features to capture plagiarism seeds. But for high-obfuscation plagiarism detection, these single type features are not sufficient for identifying the plagiarism seeds effectively because of the varied plagiarism methods used in high-obfuscation plagiarism. This paper presents a multi-features fusion method for the highobfuscation plagiarism seeds identification. This method exploits Logical Regression model to integrate lexicon features, syntax features, semantics features and structure features which extracted from suspicious document and source document. A multi-feature fusion classifier based on Logical Regression model is proposed to decide whether a text fragment pair can be regarded as plagiarism seeds or not. Experimental results on the PAN@CLEF2013 summary-obfuscation corpus show that the fusion of different types of features produces more accurate results.