The Model And Implementation Of Multi-label Software Behavior Learning

Software behavior learning is an important task in software engineering.With the software scale enlarging,increasing number of the developers is involved into the software developing.Through software behavior learning technology to complete the quality of software maintenance task and software feature discovery gradually become a hot topic in the current academia and industry.Software behavior is usually represented as information collection of program execution,the basic theory of software behavior learning is:the similar software behavior could reveal the similar software feature,which including the faults,the vulnerability of the software and so on.In order to reveal the software feature in the software,we introduce some machine learning technology to analyze the software behavior automatically,which could be used to assit the many important software engineering tasks,such as testing,debugging and failure report classification and so on.In practical,the introduction of software behavior learning technology has been proved has a great positive impact on the tasks.Current software behavior learning technology is mainly based on single label model,which is assumpt that similar software behavior can only reflect one feature of software.But in practice,the single label model usually cause some important information missed in the learning procedure.Therefore,this paper puts forward a multi-label software behavior learning model,which could better abstract the software engineering problems in the practical situation,thus improve performance of the software engineering methods which base on the software behavior model.We introduce the multi-label classification algorithm,MLkNN and the fuzzy clustering algorithm Fuzzy C-Means,to build up a multi-label software behavior learning model,which has a higher applicability and a better learning result.Further more,we present two application situation of the multi-label software behavior learning model,and we use five industrial programs as the target program to conduct the contrast experiment with the single label learning model.Experiment results show that,multi-label software behavior model has a better performance than the current widely used single label software behavior learning model.