An Approach to Detecting Similar Programs using Instruction Embeddings of Java Programs

With recent advancements in information technology, data processing and analysis technologies have become essential in fields such as artificial intelligence and big data.Furthermore, the importance of software in information technology is increasing, and data processing techniques for understanding and utilizing software characteristics have various applications in the development of applied services.This paper proposes an instruction embedding method to effectively represent software characteristics and utilize it in similar program detection methods.The instruction embedding is a data representation method that reflects the characteristics of the software, and is designed to express the frequency of the instructions as vectorized data so that the characteristics of the instructions included in software can be reflected.In this paper, we propose an instruction embedding method for similar program detection and evaluate its performance through a similar program detection model through a neural network.In benchmark experiments using neural network models, the similar program detection model achieved 99.6% accuracy with data using instruction embedding.Additionally, sensitivity and specificity were recorded at 88.0% and 99.7%, respectively.The experimental results confirmed that program analysis data using instruction embedding is an effective method for detecting similar programs.It is anticipated that program analysis and embedding methods through instruction embedding can be widely utilized to enhance performance in software analysis and artificial intelligence applications.