采用集成特征选择的网络书写纹识别研究  

Research on Online Writeprint Identification Method Using Ensemble Feature Selection

作　　者：孙建文[1] 刘三(女牙)[1] 杨宗凯[1] 王佩[2] 

机构地区：[1]华中师范大学国家数字化学习工程技术研究中心,武汉430079 [2]武汉大学信息管理学院,武汉430072

出　　处：《小型微型计算机系统》2012年第5期1108-1112,共5页Journal of Chinese Computer Systems

基　　金：国家"十二五"科技支撑计划项目(2011BAK08B03)资助;国家"八六三"高技术研究发展计划项目(2008AA01Z131)资助;华中师范大学中央高校基本科研业务费项目(CCNU09A02006)资助

摘　　要：网络书写纹识别技术通过分析用户在网络留言中的书写风格识别其身份.针对网络书写纹特征具有高维、冗余的特点,提出一种基于集成特征选择的识别方法.采用基于Filter与Wrapper的混合式特征选择算法对特征空间进行划分,训练出与待识别用户数相同的基分类器,使每一个基分类器成为代表一个用户的专家.通过对20个校园BBS用户的实验研究,结果表明该方法有效提高了网络书写纹的识别性能与识别模型的可理解性.Online writeprint identification is a technique to identify individuals based on textual identity cues people leave behind online messages.To deal with the high-dimensional and redundant feature problems,an ensemble feature selection based writeprint identification approach is proposed in this study.This method employs a Filter and Wrapper combined model to construct the ensemble base on feature space partition.There is a one to one correspondence between the authors and base classifiers which are trained to make each one of them become an expert for the associated author.Experiments are conducted on a test bed encompassing hundreds of campus BBS posts from 20 users.The results show the proposed approach improves the performance significantly,and also increases the interpretability of the identification model.

关 键 词：网络书写纹 n-gram字符 集成特征选择 RELIEF 遗传算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]