Full Citation

Title: MlNING MULTI-LEVEL ASS0CIATI0N RULES USING DATA CUBES AND MlNING N-MOST INTERESTING lTEMSETS

Citation Type: Dissertation/Thesis

Publication Year: 2000

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: Most of previous studies on mining association rules are single dimensional ones and are at single concept level. However, the more useful or interesting associa- tion rules can be found with the multi-levels using data cube, which is one of the popular structures in OLAP. A new algorithm, is proposed for mining multi-level association rules using data cubes. This algorithm is called as mL-layered-search and its performance is also studied. MoreoVer, two thresholds, the minimum support and minimum confidence thresholds, are necessary to mine association rules in most previous studies. But it is very difficult for users to set these two thresholds to obtain the result they required. If these thresholds are set too small, too many rules will be mined. It is difficult to select the information. If these thresholds are set too large, there may not be any result. Users would not have much idea about how large the thresholds should be. Here we study an approach where the user can set a threshold on the amount of results instead of the thresholds. Since the most step in the mining of association rules is the discovery of large itemsets, and Most of previous studies on mining association rules are single dimensional ones and are at single concept level. However, the more useful or interesting associa- tion rules can be found with the multi-levels using data cube, which is one of the popular structures in OLAP. A new algorithm, is proposed for mining multi-level association rules using data cubes. This algorithm is called as mL-layered-search and its performance is also studied. MoreoVer, two thresholds, the minimum support and minimum confidence thresholds, are necessary to mine association rules in most previous studies. But it is very difficult for users to set these two thresholds to obtain the result they required. If these thresholds are set too small, too many rules will be mined. It is difficult to select the information. If these thresholds are set too large, there may not be any result. Users would not have much idea about how large the thresholds should be. Here we study an approach where the user can set a threshold on the amount of results instead of the thresholds. Since the most step in the mining of association rules is the discovery of large itemsets, and Most of previous studies on mining association rules are single dimensional ones and are at single concept level. However, the more useful or interesting associa- tion rules can be found with the multi-levels using data cube, which is one of the popular structures in OLAP. A new algorithm, is proposed for mining multi-level association rules using data cubes. This algorithm is called as mL-layered-search and its performance is also studied. MoreoVer, two thresholds, the minimum support and minimum confidence thresholds, are necessary to mine association rules in most previous studies. But it is very difficult for users to set these two thresholds to obtain the result they required. If these thresholds are set too small, too many rules will be mined. It is difficult to select the information. If these thresholds are set too large, there may not be any result. Users would not have much idea about how large the thresholds should be. Here we study an approach where the user can set a threshold on the amount of results instead of the thresholds. Since the most step in the mining of association rules is the discovery of large itemsets, and Most of previous studies on mining association rules are single dimensional ones and are at single concept level. However, the more useful or interesting associa- tion rules can be found with the multi-levels using data cube, which is one of the popular structures in OLAP. A new algorithm, is proposed for mining multi-level association rules using data cubes. This algorithm is called as mL-layered-search and its performance is also studied. MoreoVer, two thresholds, the minimum support and minimum confidence thresholds, are necessary to mine association rules in most previous studies. But it is very difficult for users to set these two thresholds to obtain the result they required. If these thresholds are set too small, too many rules will be mined. It is difficult to select the information. If these thresholds are set too large, there may not be any result. Users would not have much idea about how large the thresholds should be. Here we study an approach where the user can set a threshold on the amount of results instead of the thresholds. Since the most step in the mining of association rules is the discovery of large itemsets, and other problems such as mining correlations or subspace clustering also depend on this step, mining itemsets is an interesting problem. We assume users would specify N, the number of itemsets to be mined and we shall return the N itemsets with the greatest supports. Two new algorithms, which are Itemset-Loop and Itemset-iLoop, are proposed to mine a set of interesting itemsets how many the users required. They find A^itemsets and use the iterative loop-back approach for avoiding any missing itemsets with enough support. The results are the N-most interesting itemset.

Url: https://core.ac.uk/download/pdf/48535110.pdf

User Submitted?: No

Authors: Kwong, Wang-Wai

Institution: Chinese University of Hong Kong

Department:

Advisor:

Degree:

Publisher Location:

Pages: 123

Data Collections: IPUMS USA

Topics: Other

Countries: United States