Han, J., Fu, Y.: Mining multiple-level association rules in large databases. Knowledge and Data Engineering 11 (1999) 798--804  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Association Rule Mining Some transaction databases may contain data with a hierarchical structure. Users are interested in generalized association rules that span different levels of the hierarchy since, sometimes, more interesting rules can be derived by taking the hierarchy into account [HF99, SA95]. For example, High level rules, such as 80 of customers who purchase milk may also purchase bread. Low level rules, such as 70 of people buy wheat bread if they buy 2 milk. We can select the uniform minimum support threshold for different levels. However, one problem will arise. If ....

....with the class label attribute as the consequent. Those rules obtained can then be constructed into a classifier. The performance of the consequent constraint pruning will be given later in this chapter. 4.4.3. Bit based Pruning for Multi level Rules There may be interest in multi level rules [HF99], which means the different itemsets in the rule can have different precision in our case. A bit based pruning technique can be applied for mining multi level rules. The basic idea of bit based pruning is that, if 67 Aset [1,1] 2 (the interval [1,1] in band 2) is not frequent, then Asets [10,10] ....

J. Han and Y. Fu, "Mining Multiple-Level Association Rules in Large Databases," IEEE Transactions on Knowledge and Data Engineering, Vol. 11, No. 5, September/October 1999, pp. 798-805.

....of attributes; abstraction from brands gives generalized rules, that are often more informative, intuitive and exible. As mentioned before, also non boolean attributes lead to natural hierarchies. Since the number of generated rules increases enormously, a notion of interestingness, cf. [8, 16], is necessary to describe them. It might for instance be informative to know that people often buy milk early in the day; on a more detailed level one might detect that people who buy low fat milk often do so between 11 and 12 o clock. The more detailed rule is only of interest if it deviates ....

....are complementary to the underestimated ones. If necessary, the con dence can be used to turn the list of interesting itemsets into a list of interesting rules, further decreasing the number of interesting rules. Note that ancestors of frequent itemsets are automatically frequent, unless as in [8] di erent support thresholds are speci ed at di erent tree levels (if, e.g. fMilk ; Breadg is frequent, fDairy ; Breadg should be frequent too in order to compute the support deviation) The run time of the algorithms may as usual be long when the number of records is large and the minimum ....

J. Han and Y. Fu. Mining multiple-level association rules in large databases. IEEE Transactions on Knowledge and Data Engineering, 11:798-804, 1999.

....of the products is available. In this setting association rules may involve categories of products; abstraction from brands gives generalized rules, that are often more informative, intuitive and exible. Since in this case the number of rules increases enormously, a notion of interestingness, cf. [5, 8], is necessary to describe them. It might for instance be informative to know that people who buy a history book also tend to buy a crime novel; on a more detailed level one might nd that people who buy The Rise and Fall of the Roman Empire often also buy The Hound of the Baskervilles . The ....

....occurs. In the example in Section 1 we would get 50= 100 0:5) 1:0 through parent C and 50= 50 0:5) 2:0 through parent B. If this support deviation is higher than the interestingness threshold, the itemset is called interesting. Note that the ancestors are automatically frequent, unless as in [5] di erent support thresholds are speci ed at di erent tree levels. The frequent itemsets can be ordered with respect to support deviation: the higher this ratio, the more interconnection occurs between the products involved. In fact, the assumption concerning the independence between lifted and ....

J. Han and Y. Fu. Mining multiple-level association rules in large databases. IEEE Transactions on Knowledge and Data Engineering, 11:798-804, 1999.

....total number of items, to be 1000, L, the number of potentially frequent itemsets, to be 2000, and D, the total number of transactions, to be 100,000. The scale up tests on total number of items I , average size of transactions T , and total number of tuples D, showed satisfactory results as well [13]. Two databases settings are used, DB1, with average size (# of items) of potentially frequent itemsets of 4 and average transaction size (# of items) of 10, and DB2, with average size of potentially frequent itemsets of 6 and average transaction size of 20. Each transaction database is ....

....are excluded. Two algorithms are implemented to find cross level rules, ML T2LA C and ML T1LA C, which are revised version of ML T2LA and ML T1LA, respectively. Our experiments show both algorithms are an order of magnitude slower than their counterparts for minimum supports from 1 to 5 [13]. This is because there are many more frequent itemsets at high levels and the support computation is more complex. 9 6 Presentation of Interesting Association Rules Not every strong rule so discovered (i.e. passing the minimum support and minimum confidence thresholds) is interesting enough ....

[Article contains additional citation context not shown here]

J. Han and Y. Fu. Mining multiple-level association rules in large databases. In Technical Report, University of Missouri-Rolla, http://www.umr.edu/ yongjian/pub/ml.ps, 1997.

....number of items, to be 1,000; L, the number of potentially frequent itemsets, to be 2,000; and D, the total number of transactions, to be 100,000. The scale up tests on total number of items, I; average size of transactions, T ; and total number of tuples, D, showed satisfactory results as well [13]. Two database settings are used, DB1, with average size (the number of items) of potentially frequent itemsets of 4 and average transaction size (the number of items) of 10, and DB2, with average size of potentially frequent itemsets of 6 and average transaction size of 20. Each transaction ....

....Two algorithms are implemented to find cross level rules, ML T2LA C and ML T1LA C, which are revised versions of ML T2LA and ML T1LA, respectively. Our experiments show both algorithms are an order of magnitude slower than their counterparts for minimum supports from 1 percent to 5 percent [13]. This is because there are many more frequent itemsets at high levels and the support computation is more complex. 6 PRESENTATION OF INTERESTING ASSOCIATION RULES Not every strong rule so discovered (i.e. passing the minimum support and minimum confidence thresholds) is interesting enough to ....

[Article contains additional citation context not shown here]

J. Han and Y. Fu, Mining Multiple-Level Association Rules in Large Databases, technical report, Univ. of MissouriRolla, URL: http:// www.umr.edu/~yongjian/pub/ml.ps, 1997.

No context found.

Han, J., Fu, Y.: Mining multiple-level association rules in large databases. Knowledge and Data Engineering 11 (1999) 798--804

No context found.

Han, J., Fu, Y.: Mining multiple-level association rules in large databases. Knowledge and Data Engineering 11 (1999) 798--804

No context found.

Han , J. and Fu, Y., "Mining multiple-level association rules in large databases," IEEE Trans. Knowledge and Data Engineering Vol. 11, No. 5, Sept. 1999 , pp.788-805.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC