Research On Location Of Software Defects Based On Frequent API Usage Pattern Mining

With the development of computer and network, the computer software plays anincreasingly significant role in our daily life. There are still a lot of bugs and flaws insoftware because of its complex structure and cumbersome development process. So it iscrucial to test the software and locate the bugs for guaranteeing the quality of software.The use of correct programming rules to detect software bugs has become an importantresearch direction in software testing, with its short time testing and high accuracy.The method of defect localization based on mining rules is to find rule violationsthrough the analysis of source code. API usage patterns is one way that applys staticanalysis to source code, it could provide the way to use the APIs correctly for softwaredevelopers. This paper presents an improved method based on frequent pattern mining,which locates the defects in the software through three steps, including extracting thepattern graphs, mining frequent API usage patterns and anomaly detections. The extractionof pattern graph is based on the standardizing node, the main idea of this step is to reducethe semantic differences resulting from the different programming styles of developers, bysimplifying and abstracting the complexity statement. The extraction of pattern graphneeds two steps, firstly merging complex structures into temporary pattern graph, thencreating final pattern graph by analysing temporal usage order and data relevance amongall the nodes. Frequent pattern mining employes the node expansion and works out thesubgraph isomorphism problem with pattern feature vectors. By checking the minedpatterns and their occurrences, estimating whether the expansion process is abnormal ornot, therefor the anomaly testing can locate software defects. The experiment shows thatthe improved method is effective, and it can mine multi-object API usage patterns.