基于问题类别自动分类的参与者推荐

Participant Recommendation Based on Automatic Classification of Issue Categories

在开源社区中,参与者积极参加问题解决过程对于推进开源社区的发展具有积极意义.在本文中,我们选择了部分Github流行仓库中的18215个问题,根据ISO/IEC 14764规范并将问题分为纠正性、适应性、完善性或预防性维护,然后使用了深度学习模型对问题的类型进行了分类,并分析问题类型对开发人员参与评论问题积极性的影响,我们的分析表明,部分开发人员对问题类型具有敏感性,即他们更偏向于参与特定类型的问题解决过程.基于此认识,我们提出了一个对问题进行自动分类的参与者推荐方法.在该方法中,利用了Atten-3CNN深度学习模型进行问题分类,达到了较高的准确度.在问题分类的基础上,我们构建了开发者的问题偏好模型,并计算问题与开发者的匹配程度从而实现回答者推荐.在实际数据上的实验结果表明,加入问题类别到问题特征向量后,显著提高了问题类型敏感人群参与问题推荐效果.

Onopen source society, participants’ active involvement in issue resolution process has a positive effect to it’s development.In this paper,we surveyed 18215 issues from some popular Github repositories, and divided parts of them into corrective,adaptive,perfective, or preventative maintenance based on ISO/IEC 14764 standard.Then,we used a deep learning model to classify the issue and analyzed the influence of maintenance type on developers’ activities for commenting on the issue.The results depicted that some developers tended to attend the resolution process of issues of some types.Therefore,we proposed a participant recommendation approach based on issue type classification.We used an Atten-2 CNNdeep learning model to classify the types of issues and obtained satisfying result.Based on the issue types,we built the features of developers and calculated the matching degree between issues and developers so that developers with high matching degrees are recommended.The experiment results show that after adding the issue types into the issue features,the recommendation results can be greatly improved for issue type sensitive persons.