Software fault localization based on eigenvector centrality in complex network theory
Wentao Wu, Shihai Wang, Yuanxun Shao, Wandong Xie · 2024
Software debugging plays a crucial role in fault localization tasks, and spectrum-based fault localization (SBFL) is a hot topic in software automation debugging research. However, existing SBFL technologies are generally limited by tie within ranks, where a large number of elements share the same suspiciousness, which in turn severely limits the performance of SBFL. To this end, we propose an SBFL model based on eigenvector centrality (FLEC). This algorithm first utilizes the statement coverage information in the program spectrum to construct a statement network, and adopts the correlation between statements as edge weights. Then, FLEC takes statement suspiciousness as node weight. On this basis, the algorithm utilizes the eigenvector centrality to calculate the weighted suspiciousness of each statement while considering both node importance and correlation between nodes. Finally, FLEC conducted experimental validation on 3 datasets of Defects4J, and the results showed an average improvement of 10.7% in ACC@N compared to the optimal baseline.