Modeling the neural network architecture to identify the author of the source code

Abstract: The paper proposes new hybrid architectures of neural networks to solve the problem of identifying the author of the source code. Models based on popular unidirectional and bidirectional convolutional-recurrent architectures are considered. The most effective model achieves an accuracy of 97% and demonstrates independence from the language in which the author programs, which makes it better in comparison with analogues.

Keywords: model, machine learning, source code, identification, deep neural networks