Research And Implementation Of Software Bug Localization Based On Learning To Rank

The finding and fixing of software bug is one of the core tasks in software engineering.Software bug tracking system is widely used in complex software projects,by reporting bugs,assigning tasks,and bug localization to fix bugs.However,it is very labor-intensive and time-consuming to complete the whole process manually.The purpose of this study is to use computer technique to automate the software bug localization process.The current research focused on the Java language data sets,and there is still room for accuracy improvement.In this thesis,natural language processing technique is used for preprocessing,information retrieval technique is used to compute the similarity,and machine learning technique is used to recommend the results.A data set of Apache Spark is constructed.The query set consists of bug reports of version 1.6.0 and 1.6.1,and the document set consists of the source code files of version 1.5.2.A report-file correspondence table is obtained through manual checking.This thesis measures the similarity between bug report and file from three dimensions of text,identifiers and components.Different from the traditional linear weighted method,Ranking SVM algorithm is used to rank the files.Finally,a simple prototype system of bug localization named RSLocator(Ranking SVM Locator)is shown in the thesis.The proposed scheme is compared with the classic VSM model and BM25 model on Spark data set,and precision,recall and MRR are used as indicators.The result of combination three dimensions of text,identifier,and component is better than using single dimension.In addition,the Ranking SVM algorithm performs better than the linear weighting algorithm on result recommendation.