Toward an Effective Bug Triage System Using Transformers to Add New Developers

Abstract

As defects become more widespread in software development and advancement, bug triaging has become imperative for software testing and maintenance. The bug triage process assigns an appropriate developer to a bug report. Many automated and semiautomated systems have been proposed in the last decade, and some recent techniques have provided direction for developing an effective triage system. However, these techniques still require improvement. Another open challenge related to this problem is adding new developers to the existing triage system, which is challenging because the developers have no listed triage history. This paper proposes a transformer-based bug triage system that uses bidirectional encoder representation from transformers (BERT) for word representation. The proposed model can add a new developer to the existing system without building a training model from scratch. To add new developers, we assumed that new developers had a triage history created by a manual triager or human triage manager after learning their skills from the existing developer history. Then, the existing model was fine-tuned to add new developers using the manual triage history. Experiments were conducted using datasets from well-known large-scale open-source projects, such as Eclipse and Mozilla, and top-k accuracy was used as a criterion for assessment. The experimental outcome suggests that the proposed triage system is better than other word-embedding-based triage methods for the bug triage problem. Additionally, the proposed method performs the best for adding new developers to an existing bug triage system without requiring retraining using a whole dataset.

Automatic Component Prediction for Issue Reports Using Fine-Tuned Pretrained Language Models

“…They also adopted the contribution-based downsampling method that Mani et al [7] used to address dataset problems. Through the experiments, they achieved better topk accuracy results than Lee et al [21] and Mani et al [7] Thus, this finding indicates that the proposed fine-tuned pretrained language model can predict components correctly because the component prediction dataset has fewer classes, which were developers in Zaidi et al [39] study.…”

“…Each study has proposed a different wordbased method, but the studies use the same information from the issue report dataset: the title and description. In addition, various approaches toward automatic bug triage [21], [26], [39] using deep learning methods have been actively conducted recently. Thus, we also reviewed automatic bug triage research and found that it uses the title and description information from the issue report repository to predict a developer (assignee).…”

“…Thus, these results are evidence of the proposed second task method, which downsamples the dataset by deleting the components with few issues. Zaidi et al [39] adopted the bidirectional encoder representations from transformers (BERT) [11] model, which is a pretrained language model to predict the developer using the title and description information extracted from the work by Mani et al [26] and data collected from Mozilla 4 and NetBeans 5 . They also adopted the contribution-based downsampling method that Mani et al [7] used to address dataset problems.…”

Developer Assignment Method for Software Defects Based on Related Issue Prediction

“…With the continuous development of machine learning technology, some researchers utilized traditional machine-learning and deep-learning algorithms to solve the developer assignment task [14,[18][19][20][21][22][36][37][38][39]. Jonsson et al [19] proposed a defect assignment method based on ensemble learner, using TF-IDF technology to extract features from the title information and description information of defect reports and combining Naive Bayes, Support Vector Machines, KNN, and Decision Tree classifiers to create a stacked generalization classifier for improving the prediction accuracy of automatic developer assignment.…”

A Survey on Bug Deduplication and Triage Methods from Multiple Points of View

“…Experimental results showed that DeepCrash achieved an F score of 80.72%. Zaidi et al proposed using a transformer (BERT) to recommend developers for a given bug [117]. Structural features were extracted from bug report descriptions, and the unstructured text was tokenized and fed into a fine-tuned BERT model.…”