mba論文面向零售業(yè)的關(guān)聯(lián)規(guī)則動態(tài)挖掘算法研究pdf

1、南京航空航天大學(xué)碩士學(xué)位論文面向零售業(yè)的關(guān)聯(lián)規(guī)則動態(tài)挖掘算法研究姓名：閆珍申請學(xué)位級別：碩士專業(yè)：計算機(jī)科學(xué)與技術(shù)指導(dǎo)教師：皮德常2010-12面向零售業(yè)的關(guān)聯(lián)規(guī)則動態(tài)挖掘算法研究 II Abstract With the development of barcode technology and the popularity of POS (Point Of Sells) systems, tremendous business op

2、portunities hold in daily transactions of the retail business. As one of the most active research method about data mining, association rule mining technique has been applied to the field of retailing. However, the tradi

3、tional association analysis algorithms cannot efficiently handle these retail data for its features such as complex structure, mass storage and dynamic update with the time. Therefore, it urgently needs to design a targe

4、ted data mining algorithms to analyze retail dataset. In this paper, we research the processing of complex data types, assessing of different modes and analyzing of mining results, thus solved the problems in transaction

5、 data. The main contributions of this dissertation are as follows: Firstly, to deal with the problem that the Apriori algorithm can handle sparse itemset which contain huge short models but runtime is not efficiency, a f

6、requent itemsets mining algorithm based on High-dimensional Sparse dataset named FIHS is proposed and meantime derive a new structure to store frequent itemsets based on Apriori algorithm. FIHS only scan the dataset once

7、 and can avoid generating infrequent candidate itemsets through optimizing the operation of connection and pruning. According to theoretical analysis and experiments, FIHS algorithm enjoys many advantages aiming at high-

8、dimensional sparse dataset, such as quick mining, less memory space, etc. Secondly, to solve the problem of efficiently maintenance frequent itemsets while the data dynamically update and the parameters have changed, a q

9、uick update algorithm named SWFIUA is proposed. The algorithm uses the concept of sliding time window to minimize the times of scanning dataset and reduce the number of candidate itemset. Experimental results show that,

10、SWFIUA algorithm improve the efficiency and also is simple, easy to maintain. Thirdly, the interestingness measure of Consine is introduced to improve the interest of mining rules based on the traditional framework of Su

11、pport-Confidence. Then, a mining algorithm IMAR based on the association rules of interestingness measures is proposed. The algorithm limits the format of the generated rules, redefines the conception of strong rule and

12、divides the rules into (positive) strong rules, (positive) weak rules and anti-rules. At the same time, to make better use of association rules to optimize business, “CompetItems model” and “MaxProfit model” are proposed