The 34rd IEEE/ACM International Conference on Automated Software Engineering （ASE 2019）was held in San Diego from November 10th to November 15th. The paper on Automatic Generation of Pull Request Descriptions by the Software Engineering Team of the School of Computer Science won the ACM SIGSOFT Distinguished Paper Award. A total of 445 papers were submitted to ASE 2019. After rigorous screening, 85 submissions were accepted, and only 6 were selected as Distinguished Papers.
On code sharing cites (such as GitHub), high-quality Pull Request (PR) descriptions may reduce the possibility of the PR being ignored or rejected and can also help subsequent software maintenance. However, developers sometimes neglect to write descriptions for PRs. To alleviate this problem, this award-winning paper proposed the automatic generation of PR descriptions and creatively put forward it as a text summarization problem. That is, integrate the code submission log of a PR and the comment added by the developer into an article, treat the description of this PR as the summary of the article and attempt to automatically generate a summary based on the article.
In order to solve this special text summarization problem, the authors adopted a novel seq2seq model. By integrating the pointer generator and a special loss function based on reinforcement learning, the authors dealt with the problem of out-of-vocabulary words in software artifacts and bridged the gap between the training loss function of the sequence-to-sequence model and human evaluation. The authors also constructed the first automatically generated PR dataset to validate their approach, and the evaluation results showed that this approach outperforms two baselines by significant margins.
The primary author of this award-winning paper is Liu Zhongxin, a Ph.D. student at the School of Computer Science, Zhejiang University. His main co-authors are his mentor, Professor Li Shanping, and Xia Xin from Monash University. This is the second time Liu Zhongxin has won the ACM SIGSOFT Distinguished Paper Award.
The software engineering team of the ZJU School of Computer Science has achieved world-class research results in software engineering theories, especially software analytics, software warehouse mining and empirical software engineering. In recent years, they have published more than 40 papers in CCF A journals and conferences in the field of software engineering. This is the third time they won the ACM SIGSOFT Distinguished Paper Award after ICSE 2018 and ASE 2018.