USING MACHINE LEARNING METHODS IN SOFTWARE CLONE DETECTION
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The article is devoted to the analysis of methods for determining the similarity of software code fragments, particularly at the level of binary representations. The highest efficiency is achieved by hybrid methods that combine analysis at multiple levels of abstraction. All approaches are classified based on the features used: syntactic and semantic. Syntactic features include low-level elements such as byte sequences, assembly instructions, statistical measures, and control flow graphs (CFG). Their analysis is close to the byte level and is effective for finding syntactic clones. Semantic features, on the contrary, describe relationships within the code and often represent vector or graph models built using machine learning algorithms, including NLP methods. The key problem is the selection and combination of optimal feature sources. Using only low-level data leads to a "lack of semantics," while relying solely on high-level representations increases the number of false positives. However, combining approaches often reduces overall effectiveness by increasing the frequency of both type I and type II errors. As a solution, the sequential extraction and joint use of low-level binary code features and high-level semantic features extracted based on Abstract Abstract Syntax Trees (AAST) is proposed to optimize the search for semantic clones.
Fields of Study & Subjects
Software Engineering Research
Physical Unclonable Functions (PUFs) and Hardware Security
Academic integrity and plagiarism
Publication details
- Journal / source
- Bezopasnost informacionnyh tehnology
- Volume
- 33
- Issue
- 2
- Pages
- 28–37
- Publication date
- 1 March 2026
- Language
- EN