Code Visualization for Plagiarism Detection

The use of deep convolutional neural networks is very common in computer graphics. With this, methods for exploiting knowledge in other fields are also developing. Finding plagiarism among student source codes is challenging, especially when students have the same assignment. In this case, we try to find differences between two semantically identical codes at the level of syntax, approach, or just style. This paper aims to visualize binary codes and verify if it is possible to detect plagiarism using deep convolution neural networks. Using the siamese network, we trained a neural network to evaluate the similarity between the two programs. The training data for our network are the ICPC competition submissions for which we can be confident of their authorship. The overall success rate of our model consistently reaches 75 to 80 % accuracy, which mainly shows that the visualization of inherently non-graphical entities (like source code) can be useful in the application of neural networks designed primarily for graphical purposes.