ISSTA2024

HECS: A Hypergraph Learning-Based System for Detecting Extract Class Refactoring Opportunities

Luqiao Wang, Qiangqiang Wang, Jiaqi Wang, Yutong Zhao, Minjie Wei, Zhou Quan, Di Cui, Qingshan Li

被引用 2 次

摘要

HECS is an advanced tool designed for Extract Class refactoring by leveraging hypergraph learning to model complex dependencies within large classes. Unlike traditional tools that rely on direct one-to-one dependency graphs, HECS uses intra-class dependency hypergraphs to capture one-to-many relationships. This allows HECS to provide more accurate and relevant refactoring suggestions. The tool constructs hypergraphs for each target class, attributes nodes using a pre-trained code model, and trains an enhanced hypergraph neural network. Coupled with a large language model, HECS delivers practical refactoring suggestions. In evaluations on large-scale and real-world datasets, HECS achieved a 38.5% increase in precision, 9.7% in recall, and 44.4% in f1-measure compared to JDeodorant, SSECS, and LLMRefactor. These improvements make HECS a valuable tool for developers, offering practical insights and enhancing existing refactoring techniques.