Source Code Author Identification Method Combining Semantics and Statistical Features

Abstract

The globalization of information sharing has made copying easier and easier. The endless duplication of plagiarism has aroused wide attention in academic circles, and the related research in plagiarism detection has become a hot topic in recent years. Taking deep learning-based plagiarism detection modeling as the research object and improving the performance of the plagiarism detection system as the research objective, this paper conducts an in-depth study on the task of source code author identification in internal plagiarism detection. In the task of source code author identification, text features based on pre-trained language model can be used to model the semantic information of code fragments. However, it still lacks in the representation of complex full-text statistical features at the level of text granularity. Therefore, this paper proposes a source code author identification method that integrates semantic and full-text statistical features and combines statistical and semantic features to build a source code writing style model to realize accurate identification of author identity. Experimental results on AI-SOCO datasets show that the proposed modeling method is superior to the statistical and single semantic feature models.