| Today,we are in a society full of data,and we can’t live without big data in every aspect of our clothing,food,housing and transportation.We have been enveloped by the big data,and the government,enterprises,and public welfare organizations are using these big data to provide a variety of convenience for our life.The realization of these conveniences is attributed to the continuous development of data mining techniques.With the development of data mining techniques,the itemset mining technique based on utility has also become one of the important development directions of data mining today according to the efforts of researchers.However,with the development of the technology,its research and development also faces challenges in various aspects such as basic theory,the extension application of concept and engineering application.For example,for the current mining algorithm,how to further improve the execution efficiency of the algorithm and save the user’s waiting time.For different real-life application scenarios,how to enable users to use this technology more simply and efficiently,without constant trial and error,to get the desired results.And for the limitations of the current utility measurement,how to expand its theoretical connotation and application extensions,better solve the problem of technology adaptability,and facilitate the rapid implementation of technology in commercial applications.Therefore,based on the innovation of basic theoretical research,this paper focuses on solving the research problems of high itemset mining in terms of theoretical connotation and its extension.The specific research contents are as follows: Therefore,this paper takes the optimization of the basic algorithm as the starting point of the research,focuses on the performance improvement of the basic high utility itemset mining algorithm,and realizes the iterative upgrade of the basic theory research.Then,the extended application research of high utility itemset mining are carried out.The specific research work is as follows:For the basic theoretical research of high average utility itemsets mining,this paper proposes an efficient high average itemset mining algorithm(EMAUI)based on efficient pruning strategy and novel list structure.The algorithm designs a utility list structure called KFAU-List(Key Feature Average Utility list),which is used to obtain the key feature information of each item in each transaction,and then uses the proposed utility overestimation model to calculate a tighter average utility upper bound of itemset;Secondly,the algorithm also proposes an effective pruning strategy for search space,which estimates the tighter itemset utility upper bound through the proposed concept of Relative Maximum Average Utility Upper Bound(RMUB);At the same time,in order to further avoid the generation of irrelevant candidate itemsets,when the itemset is judged as hopeless in the process of construct the utility list,the Early Abandoning(EA)strategy is used to give up the KFAU-List construction in advance,thus further saving time.To solve the application problem that the traditional high average itemset mining technology needs to determine an exact minimum average utility threshold before starting the mining task,an Efficient Top-k high Average Utility Itemset mining(ETAUI)algorithm is proposed.Aiming at the threshold enhancement of the core problem of Top-k,this algorithm designed a Mixed Average Utility Matrix(MAUM)to compressively store the average utility information of co-occurrence itemsets and continuous itemsets.Then based on this matrix,two minimum average utility threshold raising strategies that can deal with databases of different densities are designed.Thus,the average utility value of all possible itemsets is obtained maximally before iterative mining,and the initial utility threshold is also maximally raised.At the same time,the algorithm also uses the EPBF(Exploring the most Promising Branches First)strategy,which always effectively raises the utility threshold in the process of iterative mining.Through experimental verification,this algorithm can significantly improve the utility threshold before iteration compared with existing algorithms,and improve the practicability of the Top-k algorithms.For the applicability of high average utility itemset mining,in order to improve the accuracy of itemset mining,reveal the internal utility information of itemsets,and avoid users from making wrong decisions,an efficient mining high average utility quantitative itemsets(EMAUQI)algorithm is proposed.This algorithm adopts three methods of quantitative itemset merging,which can output different high average quantitative itemset results according to user selection during execution.At the same time,the algorithm also uses the KFAU-List structure.The internal utility information of the itemset is stored in the header of the utility list,and then uses the utility overestimation model to avoid generating additional candidate itemsets.Secondly,the algorithm also adopts RMUB strategy and EA strategy for pruning the hopeless itemset and giving up the construction of invalid utility list in advance.Finally,the MAUI algorithm combined with quantization method is used as comparison algorithm.The performance of the algorithm is verified in terms of running time,memory consumption and join operations by testing five standard datasets.And with 2 of the datasets,experimental tests with different merging methods were conducted to illustrate the impact of the merging methods on the output results. |
