The Algorithm Research Of Mining Shared Emerging Patterns

Pattern mining is an important part of the field of data mining, it is thefoundation for the task of data mining that including classification, clustering,association rules and so on. Emerging patterns(EPs) is a new kind of knowledgepatterns, because it comes from two classes that support changes greatly, it has goodperformance of classification. The study of emerging patterns is focus on one dataset.however, the study of SEPs is focus on two datasets. SEPs are the same and similarityemerging patterns from two datasets. It is has great potential for shared emergingpatterns in transfer learning and analogy. In this paper, the main contents andcontributions are as following:(1) Research the application of SEPs, proposed one application of SEPs thatuses SEPs to measure the similarity of datasets. A definition of datasets similarity isproposed, combined the average quality and quantity of SEPs to get the contribution,aggregate the contribution of SEPs to measure the similarity of datasets. Theclassification experimental results show that the classifier accuracy is high whenusing the high similarity datasets to be auxiliary data.(2) In order to solve the coverage of the SEPs，we propose a new method tomeasure patterns’ similarity. Combining the edit distance measure of string similaritypresents a similarity measure to strengthen the new definition of SEPs, use thedistance to measure the similarity of patterns. The experimental results show that thenew similarity measure method can get three times patterns than before.(3) In order to improve the performance of the algorithm, we propose a miningalgorithm besed on OSP-tree. The algorithm uses the ordered pattern tree to store thedatasets, when insert dataset into tree we can reduce the time. And in the process ofmining add the pruning strategies to reduce the depth of recursion. Take the strongdiscriminate of jumping emerging patterns into account, we combine the jumpingemerging patterns and shared emerging patterns to propose the concept of sharedjumping emerging patterns. The experiment results show that the time of OSP-treemining algorithm is mostly2/3than sp-tree.