Why and what happened? Aiding bug comprehension with automated category and causal link identification

When a new bug report is assigned to developers, they first need to understand what the bug report expresses (what) and why this bug occurs (why). To do so, developers usually explore different bug related data sources to investigate whether there are historical bugs with similar symptoms and causes related to the bug at hand. Automatic bug classification with respect to what and why information of bugs would enable developers to narrow down their search of bug resources and improve the bug fixing productivity. To achieve this goal, we propose an approach, BugClass, which applies a deep neural network classification approach based on Hierarchical Attention Networks (HAN) to automatically classify the bugs into different what and why categories by exploiting the bug repository and commit repository. Then, we explore the causal link relationship between what and why categories to further improve the accuracy of the bug classification. Experimental results demonstrate that BugClass is effective to classify the given bug reports into what and why categories, and can be also effectively used for identifying the why category for new bugs based on the causal link relations.