The Research Of Dynamic Mining Technology Based On Granular Computing

With the rapid development of computer technology, network technology and communication technology, the traditional dynamic data mining technology are unsuitable for the constantly updated dynamic database and real-time database, to make the idea of divide and conquer to reduce the complexity of dynamic context, granular computing is applied to dynamic data mining, which hides or reveals the details of the mined object by changing the size of information granule to discover different levels of information. Under the mining context of dynamic data, it is very important research significance to construct the data mining model based on granular computing to adaptively deal with dynamic data.For the problem of frequent pattern mining over data streams, this thesis discusses the technology of frequent pattern mining over data streams based on granular computing from the research of frequent closed itemset based on sliding window over data streams.Firstly, this thesis constructs the composite granules, which provides the theoretical basis for granular computing to adaptively deal with dynamic data. This thesis proposes the methods of granular computing and granular transformation to make the granule transform each other among the different problem spaces, after it constructs the composite object granules, the composite attribute granules and the composite structure granules from the information window.And, this thesis proposes a method of generating frequent closed itemset. The method uses the mixed radix to map the search space of generating the fuzzy frequent closed itemset, and uses the object granular computing to generate the fuzzy frequent closed itemset, and uses the attribute granular computing to discover frequent closed itemset.Then, this thesis constructs an embedded granular computing model. The model can adaptively construct different levels of granules accordingly to the diverse characters of data, and uses the method of granulating a super state into some substates to reduce the complexity of mining context, namely, on the one hand, it uses granular computing to discover the fuzzy frequent closed itemset from the different problem space via the transformation between the structure granule and the object granule; onthe other hand, it discovers frequent closed itemset from the different problem space via the transformation from the object granule to the attribute granule.Finally, this thesis studies the technologies with reading data for the first time and updating data, and proposes a frequent pattern mining model based on the embedded granular computing over data streams accordingly to the constructed embedded granular computing model. The model uses the embedded granular computing to real-time mine frequent closed itemset on the two phases of reading data for the first time and updating data. Comparing it with the traditional mining algorithms, the experiments indicate that it shows the good mining efficiency on the different characters of datasets, and the used memory become relatively stable in the dynamic mining, especially, the usage memory is less than the other when the number of frequent itemsets is more.To extend the mining model, this thesis uses the frequent pattern mining model based on the embedded granular computing over data streams to discover the maximal frequent itemsets for the landmark window based on the time decay. The experiments indicate that the model has the good mining efficiency and the space utilization, and it also has the good commonality for mining over the other window model.