The 31st IEEE/ACM International Conference on Program Comprehension (ICPC 2023) was held in Melbourne, Australia in May 2023, where the research paper, Implant Global and Local Hierarchy Information to Sequence-based Code Representation Models, was honored with the ACM SIGSOFT Distinguished Paper Award. The authors include Kechi Zhang (PhD Student, School of Computer Science, Peking University), Zhuo Li (Master Student, School of Computer Science, Peking University), and Ge Li (Professor of PKU’s School of Computer Science), and Zhi Jin (Professor of PKU’s School of Computer Science).
The ACM SIGSOFT Distinguished Paper Award is an award established by ACM SIGSOFT to reward no more than 10% of outstanding papers presented at top software engineering conferences.
The paper introduces an outstanding method for program representation, offering a robust foundational model for addressing various downstream tasks in software engineering. This work incorporates structural information into the program representation model of code sequences, proposing a tree-based Transformer model. By analyzing the hierarchical structure information on abstract syntax trees, a hierarchical structure model is embedded within the sequence model to handle the code's structural information and integrate it into vector representations of code sequences. Extensive experiments are conducted across 8 mainstream datasets encompassing 3 different downstream tasks, such as code classification, clone detection, and method name generation. The results indicate that this approach, owing to the incorporation of hierarchical structure information, demonstrates superior performance and training stability, with minimal extra training costs and parameter overhead.
Notably, this is the 3rd time that Prof. Zhi Jin and Ge Li's team have received the ACM SIGSOFT Distinguished Paper Award at the ICPC conference. Previously, the articles, "Deep Code Comment Generation" by Dr. Xing Hu and, "A Self-Attentional Neural Architecture for Code Completion with Multi-Task Learning" by Dr. Fang Liu, were honored with the ACM SIGSOFT Distinguished Paper Award at ICPC 2018 and ICPC 2020, respectively.
