Analysis Of Software Behavior Pattern Based On Sequence Pattern Mining

With the increasing scale of software,the analysis of the behavior of complex software systems has become a hotspot in the field of data mining.The operation of the software corresponds to an execution trace,which indicates the behavior of the software.Mining the behavior patterns that people are interested in from a large number of software execution traces is important to help solve the problem such as software defect location,software anomaly detection,test cases selection and reduction.From the data mining point of view,the execution trace of the software can be regarded as a software execution sequence.Therefore,combining the sequence pattern mining,we analyze the software behavior patterns from a large number of dynamic software execution traces.The related work is as follows.Firstly,we extract the software execution sequence from the software execution trace,and from the dynamic and static two angles,combining with the distance matching and statistical analysis method,we propose a method to measure the key functions.In this way,not only can it identify the key functions of the software,but also can provide a reference for the following study.Secondly,a frequent pattern mining algorithm based on PT-tree is proposed.The algorithm compresses the software execution sequence database into a tree structure and stores the itemsets contained in each node of the PT-tree through the data structure of FNodesets.The set-enumeration tree is used as the search space.Based on the super equivalence,a pruning strategy is proposed to improve the efficiency of the algorithm.Thirdly,a high utility path pattern mining algorithm FHUPPM based on function call path sequence is proposed in combination with the key function metric and the call relation between functions.According to the rank of key functions,we assign the proportion for the external utility of each function.The PIUL structure is designed to store both the utility information and location information of each pattern.The UCMS matrix structure is proposed and a high utility pruning strategy is designed based on it,which can be also used as the basis of extending the adjacent path patterns.Finally,the mining algorithms proposed in this paper are implemented on the Windows platform,using java language.The running time,memory usage and the scalability of the proposed algorithms are verified with the constrast algorithms.