Project description

[azhe825]

Better self-admitted technical debt identification with adaptive CNN.

• Incorporate a CNN model in the framework of a human-in-the-loop self-admitted technical debt identification system.
• Explore ways to efficiently update the CNN model with incrementally obtained labels.
• Make the prediction model of CNN adaptive to the target software project.

Improving the efficiency of software technical debt identification with CNN. I have a TSE paper under major revision right now (https://arxiv.org/abs/2002.11049), which focuses on the effectiveness of a human-in-the-loop system for technical debt identification. The current system utilizes traditional machine learning algorithms (random forest and SVM). Recently, there is evidence that in the supervised learning manner, CNN significantly outperforms random forest and SVM in identifying technical debt. Therefore, I plan to write a new paper exploring whether the performance could be further improved when replacing the traditional machine learning algorithms with CNN.

[azhe825]

Neural network-based detection of self-admitted technical debt: From performance to explainability

[azhe825]

https://github.com/ai-se/Jitterbug

[azhe825]

DATA: https://github.com/ai-se/Jitterbug/tree/master/data