ABSTRACT
This paper presents a systematic literature review of deep learning (DL)-based software refactoring, which involves restructuring and simplifying code without altering its external functionality. The study analyzed 17 primary works and found that CNN, RNN, MLP, and GNN are commonly used DL models for code refactoring, with MLP performing the best. However, current research efforts primarily focus on Java code, method-level refactoring, and single language refactoring with varying evaluation methods. The review also highlights the limitations and challenges of DL-based software refactoring and suggests future research directions.
Disclosure statement
No potential conflict of interest was reported by the author(s).