点击次数：

论文类型：期刊论文

发表时间：2011-09-15

发表刊物：EXPERT SYSTEMS WITH APPLICATIONS

收录刊物：Scopus、SCIE、EI

卷号：38

期号：10

页面范围：13378-13385

ISSN号：0957-4174

关键字：Association rule; Search engine; Semantic annotation; Genetic Network Programming

摘要：In this paper, we propose an evolutionary method for directly mining interesting association rules. Most of the association rule mining methods give a large number of rules, and it is difficult for human beings to deal with them. We study this problem by borrowing the style of search engine, that is, searching association rules by keywords. Whether a rule is interesting or not is decided by its relation to the keywords, and we introduce both semantic and statistical methods to measure such relation. The mining process is built on an evolutionary approach, Genetic Network Programming (GNP). Different from the conventional GNP based association rule mining method, the proposed method pays more attention to generate the GNP individuals carefully, which will mine interesting association rules efficiently. After the rules are generated, they will be ranked and annotated by meaningful information, such as similar rules and representative transactions, in order to help the user to understand the rules better. We also discuss how to mine generalized interesting association rules, describing more abstract level of information than the common association rules. In the simulation section, we give some demonstrations of the proposed method using a census data set, which shows a promising way to find the interesting association rules. Crown Copyright (C) 2011 Published by Elsevier Ltd. All rights reserved.