Issue-Commit Traceability Datasets

<p dir="ltr">Traceability dataset that consists of issue commit pairs</p><p>This dataset contains a collection of issue–commit pairs from software projects, annotated for the presence or absence of traceability links. These links indicate whether a specific commit is associated with resolving a particular issue, as commonly tracked in systems such as GitHub, Jira, or Bugzilla. The dataset combines samples from three sources:</p> <ul> <li>LinkFormer Dataset (<a href="https://zenodo.org/records/6524460" target="_blank"><u>LinkFormer: Automatic Contextualised Link Recovery of Software Artifacts in a Cross-Project Setting</u></a>)<br> </li> <li>20-MAD Dataset (<a href="https://osf.io/kvxr4/" target="_blank"><u>OSF | 20-MAD: Mozilla Apache Dataset</u></a>)<br> </li> <li>Carikado Dataset (a manually curated small-scale dataset from the author's past projects)</li> </ul> <p>Each data entry includes metadata and textual features extracted from:</p> <ul> <li>Issues: summary, description, type, status, creation date</li> <li>Commits: commit message, diff, file names, authoring date</li> </ul> <p>Labels are binary:</p> <ul> <li>1 indicates that a traceability link exists between the issue and the commit.</li> <li>0 indicates no such link.</li> </ul> <p>This dataset was used to fine-tune several pretrained language models (e.g., BERT, RoBERTa) for binary classification and was further analyzed using Explainable AI techniques (LIME and SHAP) to interpret feature importance.</p>