Department of Computer Science, College of Computer Sciences and Information Technology, King Faisal University, Al-Ahsa 31982, Saudi Arabia.
Sensors (Basel). 2021 Nov 7;21(21):7401. doi: 10.3390/s21217401.
The automatic localization of software faults plays a critical role in assisting software professionals in fixing problems quickly. Despite various existing models for fault tolerance based on static features, localization is still challenging. By considering the dynamic features, the capabilities of the fault recognition models will be significantly enhanced. The current study proposes a model that effectively ranks static and dynamic parameters through Aggregation-Based Neural Ranking (ABNR). The proposed model includes rank lists produced by self-attention layers using rank aggregation mechanisms to merge them into one aggregated rank list. The rank list would yield the suspicious code statements in descending order of the rank. The performance of ABNR is evaluated against the open-source dataset for fault prediction. ABNR model has exhibited noticeable performance in fault localization. The proposed model is evaluated with other existing models like Ochiai, Fault localization technique based on complex network theory, Tarantula, Jaccard, and software-network centrality measure concerning metrics like assertions evaluated, Wilcoxon signed-rank test, and Top-N.
软件故障的自动定位在帮助软件专业人员快速解决问题方面起着至关重要的作用。尽管基于静态特征的容错有各种现有模型，但故障定位仍然具有挑战性。通过考虑动态特征，可以显著增强故障识别模型的能力。本研究提出了一种通过基于聚合的神经排序（ABNR）有效对静态和动态参数进行排序的模型。所提出的模型包括使用排序聚合机制的自注意力层生成的排序列表，以将它们合并为一个聚合的排序列表。排序列表将按降序排列可疑代码语句。ABNR 模型在故障预测的开源数据集上进行了性能评估。ABNR 模型在故障定位方面表现出了显著的性能。针对诸如断言评估、Wilcoxon 符号秩检验和 Top-N 等指标，该模型与其他现有模型（如 Ochiai、基于复杂网络理论的故障定位技术、Tarantula、Jaccard 和软件网络中心度度量）进行了评估。
