Plagiarism Detection Algorithm for Source Code in Computer Science Education

2. Existing Methods And Shortcomings

Back in the 1970s, researchers started research of the similarity detection technology against source code. Halstead (1975) proposed the first algorithm named property counting method. The algorithm counted the operators and operands statistics appeared in the source program, and used the results as main basis of detecting. Ottenstein (1976) implemented the first source code near-duplicates detection system for Fortran by using properties counting method. Since the attribute notation doesn’t remain the program structure information, the method cannot meet practical requirements of short program due to high false alarm rate (definition in section 4).

In the mid-1990s, Verco and Wise (1996) added vector dimension technology to the properties counting method, but the effect is still not satisfactory. Damashek (1995) proposed structural measure approach, used program control flow as metrics, such methods are usually applicated with attribute notation. Such methods work well in checking large programs, because in handling complex problems, different programmers often have different ideas, probability of identical program control flow is extremely low, so the false alarm rate is relatively low, but experiments proved that when such algorithms applying on program designing jobs, it has a relatively high false alarm rate. Because programming as common work is simple and the fundamental knowledge is quite similar, so the students’ main concepts of solving the problems are similar, thus the control flow of the program will be basically alike.