Authorship Attribution of Source Code Based on Programmers Coding Style

Source code authorship attribution is an important problem for resolving plagiarism and copyright issues in the programming field. Modern solutions for authorship identification use features specific to a particular language and can not be easily applied to another one. Also, existing works did not test solutions on large amounts of data due to the lack of an appropriate dataset. In this work, we present a tool for mining the history of project changes that can be used to collect datasets for authorship identification from projects with multiple authors. Considering history instead of static snapshot results in a significantly higher amount of data collected from a single project. Then, we collect 7 datasets from IntelliJ IDEA project to test different aspects of authorship detection models. Also, we design two models achieving state-of-the-art accuracy in authorship attribution for Java, Python and C++ code. The models do not depend on features of a particular language and can be applied to any programming language without modification. Keywords: source code authorship identification, stylometry, abstract syntax tree, machine learning, path-based representations, random forest.