Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: Ph.D. Dissertation
- Language: Farsi
- Document No: 54182 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Heydarnoori, Abbas
- Abstract:
- As one of the most important software artifacts, release reports include most essential changes of software systems in each release. These reports are useful for documenting team activities, improving communication between team members and facilitating decision makings. They are often generated manually by a team member, which can be a tedious and time-consuming task. For each release, one should go through hundreds of issue reports, changed files, source code, and other related documents to determine what has changed and why. Then they need to identify the most important and relevant changes. Therefore, automated generation of release reports can be highly desirable. Despite the importance of release reports, current studies suffer from several shortcomings including neglecting the context of projects, and the objective and priority of issue reports. They are also dependent on human intervention either for retrieving and selecting the change information or writing the final report.We propose a Machine-Leaning-based approach to automatically generate prioritized reports. We first extract the main context of repositories (objective, functionalities, language, domain, etc.) using their textual information and files and multi-label classification. Then, we manage issue reports by categorizing them into groups of bug reports, enhancement, and documentation and support, and prioritizing them. Next, issues are mapped to their corresponding commits using two textual and non-textual components, and the commit message is generated. Finally, a hierarchical report of important changes with different levels of granularity is constructed. We first evaluated each component of the proposed approach separately. Results indicate that our approach outperforms the competing techniques in all components. We also assess the resultant report by designing a user study. The results show that experts deem the performance of the model regarding two qualitative measures of completeness and conciseness, successful. The proposed approach scored 4.1 and 4.4 (5 star-scale) on average regarding completeness and conciseness
- Keywords:
- Mining Software Repositories ; Transformer Network Optimization ; Software Evaluation ; Machine Learning ; Commit ; Source Coding ; Release Note ; Collaborative Coding ; Issue
Digital Object List
-
محتواي کتاب
- view
Bookmark