Pride: uma ferramenta de detecção de similaridade em código-fonte

Abstract

Plagiarism among students of introductory programming courses has been increasing over time. The ease of exchange of information brought by the Internet can be the factor responsible for this increase. In many cases, students try to disguise the plagiarism making some modifications to the source code. However, some masking techniques are extremely complex to be detected and may not be seen with the naked eye. In this dissertation, detection techniques were analyzed and, on this basis, was developed a system able to detect plagiarism in source code. This system is based on the representation code as an abstract syntax tree and Karp-Rabin Greedy String Tiling algorithm. The system was evaluated using a source-code base of students of programming disciplines. Oracle based comparison was performed to compare the system with others. The oracle was created from the manual analysis of the teacher of the subject, which was marked if there was plagiarism or not in each pair of source code. To represent the results, ROC curves and confusion matrices were used. The same procedure was applied to existing systems, allowing direct comparison of results. More specifically, we use the value of the area under the curve and the minimum distance to point (0, 1) of the ROC space, since these figures represent the classification performance. The analysis of results shows that, for the sample used, the developed system obtained higher area under the curve and also the shortest distance to the point (0, 1) of the space ROC. However, we find that the choice of similarity detection tool in source code will depend on conservative or liberal profile of teaching.