Automatic bug assignment has been a significant area of research in recent years, with textual bug reports playing a crucial role in aiding engineers in diagnosing and fixing software bugs. However, the presence of noise in these reports has presented challenges in utilizing classical Natural Language Processing (NLP) techniques effectively for bug assignment.
A research team led by Zexuan Li conducted a study published in Frontiers of Computer Science to investigate the impact of textual and nominal features on bug assignment. The team utilized an NLP technique, TextCNN, to assess the effectiveness of advanced techniques in handling textual features. Surprisingly, the results of the study indicated that textual features did not outperform nominal features even with the use of improved NLP techniques.
The research team delved into identifying influential features for bug assignment approaches and their significance in the bug-fixing process. Through the employment of the wrapper method and a bidirectional strategy, the team evaluated the importance of different groups of features by training classifiers. The findings suggested that nominal features, indicative of developers’ preferences, played a vital role in reducing the search scope of classifiers, ultimately leading to competitive results without the reliance on textual content.
Experimental Results
The study aimed to address three critical questions regarding bug assignment approaches. It evaluated the effectiveness of textual features when paired with deep-learning-based NLP techniques, explored the influential features for bug assignment, and assessed the extent to which selected features could improve bug assignments. The results indicated that improved NLP techniques resulted in limited enhancements, while the selected key nominal features achieved an accuracy range of 11-25% under popular classifiers like Decision Tree and SVM.
Moving forward, future research in this area could focus on incorporating source files to establish a knowledge graph linking influential nominal features with descriptive words. This could facilitate a better understanding and utilization of nominal features for bug assignments, potentially leading to more robust and accurate bug-fixing processes in software development.
Leave a Reply