Automating discussion structure re-organization for GitHub issues

Shuotong Bai

doi:10.1016/j.eswa.2023.120024

2023-09-01

PRE

Shuotong Bai

Lei Liu

刘华虓 (Huaxiao Liu) *

DOI：10.1016/j.eswa.2023.120024

原文链接

原文求助

摘要

En 中文

As a popular social code hosting platform, GitHub encourages developers to discuss and leave opinions on issues. However, the linear format of GitHub issue discussions makes popular discussions difficult for developers to organize and extract useful information effectively. In this paper, we propose an issue discussion re-organization approach, aiming at converting an issue discussion with the linear structure into a discussion tree with key information. First, we conduct a motivational study to investigate the current situation of issue discussions in GitHub. Further, to re-organize discussion structures, we employ a Transformer-based model with transfer learning to predict the response relationship between comments for re-building structures and utilize TF-IDF to extract key information from the content with different topics. The experimental results show that our approach outperforms other baselines, and achieves an average improvement of 14.54% on metrics in the task of predicting response relationships, as well as getting an average improvement of 27.19% in terms of metrics of the re-organizing task. To investigate our re-organized results from actual perspectives, we also conduct a human evaluation. The results show that our approach can predict the accurate response relationships for 80.74% of comments from actual perspectives and 63% of topics extracted by our approach are highly rated. Moreover, 90.00% of newcomers from the open-source community approve of re-organized discussion structures.

Keyword:

Discussion re-organization

GitHub

Issue

Deep learning