NOVEL CLASSIFICATION FEATURE SETS FOR SOURCE CODE PLAGIARISM DETECTION OF JAVA FILES

Abstract

In programming learning environments, the pressure of delivering many assignments makes plagiarism become the easiest solution. This problem of plagiarism threatens the learning process and obstructs the evaluation fairness. Therefore, fast, automatic and accurate detection of source code plagiarism becomes of the essence. This research proposes novel classification feature sets to detect whether a Java file is plagiarized. The proposed feature sets are based on using histograms to summarize the similarity matrix of function signatures and comparing the lexical code similarity of each individual class pair. For testing the effectiveness, a source code plagiarism dataset that consists of 12K Java files was used. The results show a 4% improvement in F-Measure. A re-annotation to the dataset is performed and improves F-Measure by 7.5%.