Fault centrality: boosting spectrum-based fault localization via local influence calculation

摘要

Spectrum-Based Fault Localization (SBFL) is a widely investigated heuristic approach and a lightweight but efficient technique. Recently, to discover useful latent information for accelerating fault localization, more attention has been paid to the research of the fault-relevant correlation among software entities. Considering that the interactive behaviors among software entities may imply some fault patterns, we introduce the fault influence of interactive entities and develop a novel synthetical fault localization approach based on the software network. (1) In line with the intuition that the entity correlated with more suspicious entities is more likely to be faulty, we firstly construct a directed, node-weighted, and link-weighted Software Fault Network (SFN). As a characterization solution, SFN maps methods into nodes and method-call relations into links. Then, SFN initializes the node weights by the raw suspiciousness score of a specific SBFL formula and assigns the link weights by the faulty similarity Fault Influence Coefficient (FIC). (2) A suspiciousness measure criterion named Fault Centrality (FC) is proposed based on SFN. This approach calculates the final suspiciousness score by aggregating the faulty influence of local interactive methods. We conduct the experiments on 349 faults of Defects4J and separately apply 33 existing SBFL formulas. According to the results, this approach can boost almost all the performance of the existing formulas. The outcomes of acc@1, acc@3, and acc@5 are better by 9.43%, 17.40%, and 18.92%, respectively.