The Research On Sequential Pattern Mining Algorithm And Its Application In Business Process Design

Data mining, one of the very active leading research fields, is a technology to discover the important information hided behind the large-scale data. Sequential pattern mining, one of the key research areas in data mining, has attracted more and more attention from database practionaers and researchers because of its promising high applicability. However, since the mining process may have to generate a combinationally explosive number of intermediate subsequences, it's difficult for existing sequential pattern mining algorithms to handle the large-scale data efficiently.Based on indepth analysis on existing researches, this thesis not only presents a sequential pattern mining algorithm which can handle large-scale sequence data in an efficient way but also apply the sequential pattern mining tool to a new field.Concerning to the algorithm research, sequential pattern mining algorithm based on Coded Frequent Pattern tree is presented. This algorithm can not only mine the frequent pattern from a large-scale database efficiently but also handle both 1-dimensional and multi-dimensional sequence data. The unified simple linear structure, catering for 1-dimensional as well as multi-dimentional data, was built based on the concept of Item Relation Flag. Besides, the unified compact data structure, named as Coded Frequent Pattern tree, was built on coding technique and ID-queue linkage. The cost of memory and CPU time was highly saved since no intermediate subsequences are recursively generated during mining. Experiments show great performance gain over existing sequential pattern mining algorithms, especially for large database.Concerning to the application, an automatically built Business Process Template was proposed, which not only apply the sequential pattern mining tool to a new field, but also provide a very useful approach to business process design. Business process design is vital for the success of enterprise business process built and reengineer. However, at present, business process design heavily depends on individual expert's experience and expertise, which leads to expensive and time consuming. The automatically built configurable Business Process Template, which utilizes the sequential pattern mining tool to analyze existing business process, was presented. Built on Websphere Business Integration Modeler, the system is developed and runned well. This system, based on which the industrial experience and knowledges can be accumulated from existing cases in a systematical way, provides a very valuable reference for business process design.