Improved plagiarism detection with collaboration network visualization based on source-code similarity

摘要

Plagiarism detection is a serious problem in higher education. Teachers use similarity (plagiarism) detection systems, which highlight similarities between student documents, to help them find plagiarism. Most systems are built for text but there are special systems to find similarities between source-code files. In most cases the results are presented in table form showing similarities between pairs of documents in descending order by similarity, and then a teacher is responsible for confirming which similar documents represent cases of plagiarism. While most systems present their results in the form of tables, only few of them present the results as a graph. Some studies indicate that using clustering algorithms to represent such data graphically can improve the speed and accuracy of finding potential instances of plagiarism in large collections of source-code files. The purpose of the study is to answer the following research questions. Can visualization of student solutions (of source-code similarities) in collaboration networks form help identify new cases of plagiarism? What are the steps to do so? The study was designed in a form of two case studies where one was performed on a graduate level university course and one on a course in professional studies. The article presents empirical results describing two cases where a collaboration network (based on source-code similarity) representation has been used. The article argues that the graphical presentation is able to identify new clusters of plagiarised source-code files that would have been missed using existing tabular presentation of data.