Automatic assignment of integrators to pull requests: The importance of selecting appropriate attributes
•Introduction of new attributes for the integrator prediction task.•Adoption of attribute selection strategies for choosing the best attribute set.•A novel classifier evaluation method tailored to the pull requests scenario.•Experiments with 32 open-source projects.•Normalized improvements 54% highe...
Gespeichert in:
| Veröffentlicht in: | The Journal of systems and software Jg. 144; S. 181 - 196 |
|---|---|
| Hauptverfasser: | , , , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
Elsevier Inc
01.10.2018
|
| Schlagworte: | |
| ISSN: | 0164-1212, 1873-1228 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | •Introduction of new attributes for the integrator prediction task.•Adoption of attribute selection strategies for choosing the best attribute set.•A novel classifier evaluation method tailored to the pull requests scenario.•Experiments with 32 open-source projects.•Normalized improvements 54% higher than the state-of-the-art.
In open-source projects that adopt the pull-based development workflow, a core developer needs to analyze the contribution received via pull requests and decide on integrating it or not in the repository. However, this process is time-consuming, leading to an increasing number of pull requests left to be analyzed. Consequently, the assignment of suitable integrators to pull requests becomes an important step in the pull-based development workflow. Classification methods have already been used to recommend integrators, based on different sets of predictive attributes. The main contribution of this paper is to identify a set of attributes that can improve the performance of the integrator prediction task reported in the literature. To do so, we first evaluate different sets of attributes used by previous studies with different classification algorithms. Besides, we explore attribute selection strategies on an extended set of attributes composed not only by the attributes already used in the literature but also new attributes we consider relevant to the problem. Experiments with 32 open-source projects evidenced that after applying attribute selection strategies and, consequently, identifying a more suitable set of attributes, the recommendation has achieved normalized improvements 54% higher than the state-of-the-art. |
|---|---|
| ISSN: | 0164-1212 1873-1228 |
| DOI: | 10.1016/j.jss.2018.05.065 |