Font Size: a A A

Research On The Concise Association Rule Representation Model

Posted on:2015-11-29Degree:MasterType:Thesis
Country:ChinaCandidate:H ChenFull Text:PDF
GTID:2308330473959337Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Association analysis is a classical research field of data mining. Its purpose is to reveal the complex correlation between variables through quantitative relationships mined from dataset. The main shortcoming of association analysis is that generally there are too many association rules in a typical mining process, which damages their interpretability and applicability and also results in great resources pressure while mining, storing and transmitting. A variety of concise representatives have been suggested for representing the whole raw association rule set. But there more or less exist deficiencies in these models, such as poor compactness, information loss or complex recovery algorithm and etc. In this thesis we would give up the MDL principle, the basis of nearly all classical association rule representatives, and try to find a new concise association rule representative based on some specfic association rule having strong description ability.The main research works of this thesis are as follows:(1) As MDL principle is the basis of nearly all classical association rule representatives, shortcomings in representation of relationship of association rules based on MDL principle has been studied. Then, Basic Association Rules (BAR), which don’t follow MDL principle but have better association representation, is proposed, and on the basis of it a new lossless association rule representative, BAR set, has been designed. Some properties of the BAR set are analyzed and are proved. According to the lattice structural features among association rules, basic association rule set is divided into the approximate BAR set and the exact BAR set. Algorithms for mining BAR’s two part are designed and are analyzed respectively. Experiments show that the proposed association rule representaive based on BAR is more concise than the existing ones.(2) For recovering the whole original association rule set from BAR set, the method based on the following steps has been proposed. Firstly, divide BAR set into the approximate BAR set and the exact BAR set. Secondly, extend the approximate BAR set and the exact BAR set repectively according to their association representative characteristics to recover the approximate rule set and the exact rule set. Finally, combine the two sets to get the whole original association rule set. Algorithms for the recoverings are designed and are analyzed. Experiments for verification have been done.
Keywords/Search Tags:data mining, association analysis, association rule, concise representation, basic association rules
PDF Full Text Request
Related items