基于粒计算的商业数据流概念漂移特征选择  

Granular computing based concept drift features selection for business data streams

作　　者：琚春华[1] 帅朝谦[1] 封毅[1] 

机构地区：[1]浙江工商大学计算机与信息工程学院,现代商贸研究中心,杭州310018

出　　处：《南京大学学报（自然科学版）》2011年第4期391-397,共7页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(71071141,60905026);浙江省自然科学基金(Z1091224,Y1091164);浙江省研究生创新科研项目

摘　　要：商业数据流具有动态性、漂移性等特性,概念漂移特征选择是数据流挖掘的重要工作之一.本文从数据流的特点和概念漂移特性出发,提出了数据流的概念形式化分析流程和基于粒计算构建数据流的概念形式化描述模型;商业数据流的概念漂移实际上取决于其概念外延的变化,文中使用包括外延偶合度和内涵偶合度在内的概念偶合度来描述概念间的相似性;经过粒化的数据流由概念格来表示,进而通过概念格对的松弛偶合度来分析数据流特征;结合概念偶合分析和数据流变化特征,阐述了一种基于数据流概念格对的松弛匹配偶合度算法,并据此分析概念格对来选择数据流的漂移特征.通过实例验证、评价了特征选择,证明其有效性.Concept drift features selection is an important aspect of data streams mining as business data streams are dynamic and easy to drift.This paper describes the characteristics and the concept drift of data streams,and proposes a work flow of formal concept analysis and a formal concept description model of streaming data based on granular computing.Concept drift in business data streams is actually decided by the changes upon the extension of the concept.Then,we describe concept coincidence,including coincidence on extent,intent,and concept.Because concept-lattices can be expressed in terms of the granulated data streams,we analyze coincidence of concept lattice pairs instead of continued and non-formal streaming data.Furthermore,we propose a concept lattice pairs-based concept relaxation-matching coincidence degree algorithm;the feature selection method is also described.Finally,experiments and analyses are presented in order to explain and evaluate the method.

关 键 词：商业数据流 概念漂移 粒计算 形式化概念分析 特征选择 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]