Automatic bug assignment has been a topic of interest for researchers over the past decade. Bug reports play a crucial role in helping engineers identify and fix bugs in software systems. However, the presence of noise in textual bug reports can pose challenges for automatic bug assignment systems, often due to the limitations of traditional Natural Language Processing (NLP) techniques.
A recent study led by Zexuan Li, published in Frontiers of Computer Science, aimed to explore the effects of textual and nominal features on bug assignment approaches. The research team implemented an NLP technique, TextCNN, to evaluate the impact of improved NLP techniques on bug assignments using textual features. Surprisingly, the results showed that textual features did not outperform nominal features, even with the use of an advanced technique.
The research team further analyzed the influential features for bug assignment approaches and found that nominal features, which reflect the preferences of developers, played a significant role in achieving competitive results. By employing a statistical perspective, the team demonstrated that nominal features could effectively contribute to bug assignments without the reliance on textual content.
The study addressed three main questions. Firstly, the effectiveness of textual features in combination with deep-learning-based NLP techniques was explored by comparing them to nominal features using TextCNN. Secondly, the research aimed to identify influential features for bug assignment approaches and provided insights into why these features were crucial for the classifiers. The team speculated that nominal features could assist in narrowing down the search scope of the classifier, a hypothesis that was validated through statistical analysis.
Experimental Results
The experimental results indicated that while improved NLP techniques had some impact on bug assignments, the selected key nominal features were able to achieve an accuracy range of 11-25% under popular classifiers like Decision Tree and SVM. The research was conducted on five diverse projects, varying in size and type, to ensure the generalizability of the findings.
The study suggests that future research could explore the integration of source files to establish a knowledge graph connecting influential nominal features with descriptive words. This approach could enhance the embedding of nominal features and improve bug assignment accuracy in software systems. By considering the interplay between textual and nominal features, researchers can further enhance automatic bug assignment techniques in the future.
Leave a Reply
You must be logged in to post a comment.