Research On Mining Representative Software Behavior Pattern From Software Execution Traces

"Internet +" era of arrival,the size of computer software has become huge.Software development engineers spend a lot of time working to maintain the software.We hope that the data generated from the implementation of the software to dig out a representative of the knowledge,so that the latter part of the software will be more convenient and accurate,for the analysis of user behavior,software fault location provides a complete basis.In this paper,from the sequence point of view,the sequence pattern mining is applied to the software execution trace,and mining the behavior pattern of software representation.Firstly,the tracing method is proposed to track the execution of the software into a sequence of software execution sequences,which is constructed into a sequence database.There are many contiguous repetitive patterns in the execution sequence of the software,and these sequences are simplified and the repeated repetition patterns are removed.The filter contiguous repetitive patterns algorithm is proposed to preprocess the software execution sequence and analyze the software representative behavior patterns.Secondly,we propose a method to mine the maximal sequence pattern algorithm from the software execution sequence.The algorithm is based on the vertical database and the three optimization strategies are designed.The first is to filter the non-maximal sequence pattern strategy,design the data structure according to the length of the packet storage;the second is the maximal extended detection strategy,through the support to determine;the third is based on EMAP storage structure design pruning strategy.Thirdly,we propose a algorithm for mining high-utility contiguous patterns from the software execution sequence.The algorithm uses the maximal utility calculation method to simplify the utility calculation method,and then uses a listed index table UL-list data structure,used to record the utility value and location information,and use the index table to propose a backward extension strategy and the new remaining utility upper bound model makes the algorithm faster to find contiguous patterns.Finally,the proposed algorithm uses the Java language to develop and run under Windows systems.The performance of the algorithm is verified by comparisonexperiment,running time,memory occupation,scalability and so on.