Research
A Multi-Objective Search Over 22,000 Java Bug Reports Localizes 88.5% Within the Top 10 Classes
Rather than maximizing lexical similarity alone, this system treats bug localization as a multi-objective search that maximizes similarity while minimizing the number of suggested faulty files, ranked at class level. Applying SPEA-2 to six open-source Java projects covering more than 22,000 bug reports beat both NSGA-II and MOEA/D and single-objective baselines on precision and recall, identifying the buggy class or file for 88.5% of reports within the top 10 recommendations and 94% within the top 20. It also transferred to an industrial Kotlin Android project, which is the language-portability claim most prior work fails.
↳ Follow the thread