Exploiting gated graph neural network for detecting and explaining self-admitted technical debts

Self-admitted technical debt (SATD) refers to a specific type of technical debt that is introduced intentionally in the software development and maintenance processes. SATD enables practitioners to take some temporary solutions instead of making comprehensive decisions, which will lead to the high complexity of the software. However, most existing studies relied on manual methods for detecting SATDs. A recent study proposed a method HATD that used a hybrid attention-based method to automatically detect SATDs and it achieved the state-of-the-art performance. However, HATD mainly focused on the locality of the comment instances and lacked of the relationship between long-distance and discontinuous comment instances. To address such an issue, in this work, we propose a novel approach named GGSATD. Specifically, GGSATD first builds the graph for comment instances and then employs the gated graph neural network to iteratively update node representation. The global representation can be obtained by the soft attention mechanism and pooling operation. Experiments on 10 projects show that our GGSATD method obtains promising performance against five baseline methods in both within-project and cross-project scenarios. Extended experiments on seven real-world projects illustrate the effectiveness of our GGSATD method.