Research On Mining And Dynamic Maintenance Of Frequent Patterns

Mining frequent pattems,the core of association rules mining,has been studied popularly in data mining research.Most of the current studies adopt a frequent pattern growth approach which does not bring the candidate generation.However,the cost of repeatedly scanning database and traveling structures is still high,and very little work has been done on maintenance of frequent patterns mined.Based on the frequent patterns mining,this paper studies the mining and dynamic maintenance of(utility)frequent patterns mining.The main contents are in the following.1 Research on frequent patterns mining based on the sharing and inheriting strategyThe sharing and inheriting strategy,based on the lattice theory,is proposed to efficiently discover the frequent patterns.Based on the share table structure,the sharing and inheriting strategy provides a way to share most of the results from the previous mining process instead of separating them distinctively,thereby dramatically reducing the cost of scanning database and traversing complex structures.Furthermore,various optimization techniques used sufficiently improve the efficiency.2 Utility frequent patterns miningFrequent patterns mining only discovers frequent patterns without considering the different utility values of individual items.In this paper, we introduce a utility frequent pattern mining model based on the sharing and inheriting strategy to find the combination of items that can continually generate high profit.This model rather satisfies the practice and develops the task of frequent pattern mining.3 Frequent equivalent patterns miningThe definition of frequent equivalent patterns,a new compression technology of frequent patterns,is proposed to compress the size of frequent patterns.The number of frequent equivalent patterns is rather smaller than the one of the frequent patterns,but includes the whole information about frequent patterns.Through the comparison with frequent closed patterns,it has higher compression ratio.4 Interactive mining of frequent patternsBased on the share table structure,we present an interactive mining method of frequent patterns,which is flawless integrated with the sharing and inheriting strategy.When the minimum support threshold is changed, it provides a way to share most of the results from the last mining process but not from the very beginning,thereby dramatically reducing the cost of repeatedly scanning database and traversing complex structures.5 Incremental mining of frequent patternsBased on the analysis of incremental mining,we propose an incremental technique for efficient maintenance of frequent patterns discovered when new transaction data are added to the transaction database.It only need scan the original database once and provides a way to maintain the frequent patterns mined,thereby dramatically reducing the repeated original database scan and calculation.6 Application of frequent patterns in association rules miningThe problem of association rules mining is based on the frequent patterns already mined.In this paper,we design the algorithm to generate association rules.The test result validates the practical value of frequent patterns mining.Moreover,we extend the interactive and incremental mining of frequent patterns mining to the utility frequent patterns mining.We use real world and synthetic data for our performance study,and compare our algorithms with various algorithms well known.The performance studies show that the sharing and inheriting strategy is efficient for mining frequent patterns and utility frequent patterns.At the same time,when the database and thresholds are changed,test results also show that the strategy can be integrated into interactive and incremental mining problem,improving the performance.