Authorship identification of source codes

Source code authorship identification is an issue of authorship identification from documents, and it is to identify authors of source codes or programs based on source code examples of programmers. The main applications of authorship identification of source codes include software intellectual property infringement, malicious code detection and software maintenance and update. This paper proposes an approach of constructing author profiles of programmers based on a logic model of continuous word-level n-gram and discrete word-level n-gram, and a multi-level context model about operations, loops, arrays and methods. Further, we employ the technique of sequential minimal optimization for support vector machine training to identify authorship of source codes. The advantage of author profiles in this paper can discover explicit and implicit personal programming preference patterns of and between keywords, identifiers, operators, statements, methods and classes. Experimental results on programs from two open source websites demonstrate that our approach achieves a high accuracy and outperforms the baseline methods.