Fighting plagiarism: metrics and methods to measure and find similarities among source code of computer programs In VPL

Abstract

Plagiarism in academia is becoming a major problem; this can be seen in the increasing number of cases that come to light. Increasing use of information and communication technologies seems to be an important factor to facilitate plagiarism, although it can be part of the solution too. Assignments where students should develop computer programs to submit their source code are no strangers to this problem, being even one of the areas in which more cases of plagiarism are reported. The most common form of plagiarism is produced by making changes in an original source file to get a different version while maintaining the behavior of the new program as in the original. In this paper, we show a tool that uses different metrics and methods to look for and show the most similar source files in a set. Similarity among files is directly related to the possibility that they are the outcome of a process of plagiarism. The metrics used are three: two proposed by the authors and a third commonly used. The reason for using three metrics is that each one is sensitive to different forms of systematic changes in source code files, so the combination of them increases the capability of discover plagiarism attempts. Searching for the most similar files requires a preprocessing consisting of: a lexical analysis, a filtering and a normalization of expressions, to get a signature for each file. These signatures will then be compared using the proposed metrics. The search process is optimized to run using a minimum memory and little time. As result of the process, we get a list of the most similar pairs of files sorted from highest to lowest similarity, besides a list of clusters of the most similar files. Both lists are shown using a gradation of colours to express the similarity levels in a friendly manner; the numeric results of the applied metrics are shown as well. This interface is designed to facilitate taken appropriate decisions. The proposed tool is part of VPL, a Virtual Programming Lab module for Moodle, a popular Learning Management System distributed under GNU/GPL license. The anti-plagiarism tool offers a user-friendly interface allowing compare files from VPL activities among them or against external sources, with online response.