收稿日期: 2015-03-13
修回日期: 2015-05-02
网络出版日期: 2015-05-20
基金资助
本文系广东省哲学社会科学"十二五"规划2013年度项目"基于情境和用户感知的知识推荐机制研究"(项目编号:CD13CTS01)研究成果之一。
Study on Spammer Detection Based on Reviewer-Specific Characteristics
Received date: 2015-03-13
Revised date: 2015-05-02
Online published: 2015-05-20
[目的/意义]重点探讨基于特征表现的虚假评论人的预测,目的在于揭示真实网络环境中"网络水军"的特点和行为规律,构建一个简洁清晰、可解释的评论人身份预测模型,为深层次的评论挖掘研究奠定基础。[方法/过程]结合实证分析和机器学习技术,对目标网站"大众点评网"的内部评价机制进行探索,利用因子分析提炼评论人属性及行为表现特征,并在此基础上构建基于Logistic回归的预测模型。[结果/结论]对于目标网站,模型对虚假评论人的分类预测精度达到73.8%,AUC指标达到80.9%。而评论人的贡献度、活跃度以及文字素养被验证与其身份有统计意义上的显著关系,但评论人的层级、情绪以及评价偏差则对其身份预测的影响不显著。实验结论和经验分析基本保持一致,模型能够被合理解释。
聂卉 , 吴毅骏 . 基于特征表现的虚假评论人预测研究[J]. 图书情报工作, 2015 , 59(10) : 102 -109 . DOI: 10.13266/j.issn.0252-3116.2015.10.015
[Purpose/significance]This paper mainly studies the problem of review spammer detection based on specific characteristics. It tends to reveal the characteristics and behavioral regularities of Water Army on the web, build a simple and reasonable explanation prediction model of identifying review spammers, to build a foundation for the deep reviews mining research.[Method/process]By integrating empirical analysis and Machine Learning (ML) techniques, it explores the inner evaluation strategies of the target website. Factor analysis is employed to extract the features and behavior-specific factors of reviewers, on the basis of which a Logistic Regression model is built to identify review spammer.[Result/conclusion]On the dataset built from the target website, the classification accuracy for spammer identification can achieve to 73.8%,and AUC measure is 80.9%. Additionally, three feature factors related with reviewers, contribution, activity and word- specific literacy are proved to be significantly associated with the identification of reviewer, while the opposite result is get for reviewers' level, emotion and rating deviation. Basically, it gets the same results by using ML based methods and empirical analysis and the reasonable explanation can be given for our prediction model.
Key words： review spammer; review spam; behavior characteristics
[1] 胡磊. 国外互联网信息可信性研究发展历程及特征分析[J]. 中国图书馆学报,2012(2):100-105.
[2] Jindal N,Liu Bing. Review spam detection[EB/OL]. [2015-05-01] http://www2007.wwwconference.org/posters/poster930.pdf.
[3] Jindal N, Liu Bing. Opinion spam and analysis[EB/OL].[2015-05-01] http://www.cs.uic.edu/~liub/FBS/opinion-spam-WSDM-08.pdf.
[4] Jindal N, Liu Bing, Lim E P. Finding unusual review patterns using unexpected rules[EB/OL].[2015-05-01]. http://www.cs.uic.edu/~liub/publications/CIKM-final-unexpected.pdf.
[5] Mukherjee A, Liu Bing, et al.Spotting fake reviewer groups in consumer reviews[EB/OL].[2015-05-01] http://www.cs.uic.edu/~liub/publications/WWW-2012-group-spam-camera-final.pdf.
[6] Ott M, Choi J Y, Cardie C, et al. Finding deceptive opinion spam by any stretch of the imagination[EB/OL].[2015-05-01]. http://homes.cs.washington.edu/~yejin/Papers/acl11_deception.pdf.
[7] Li Fangtao, Huang Minlie, Yang Yi, et al. Learning to Identify Review Spam[EB/OL].[2015-05-01].http://ijcai.org/papers11/Papers/IJCAI11-414.pdf.
[8] Jindal N,Liu Bing. Analyzing and detecting review spam[EB/OL].[2015-05-01]. http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4470288.
[9] Mukherjee A, Venkataraman V.What Yelp Fake review filter might be doing?[EB/OL].[2015-05-01]. http://www.cs.uic.edu/~liub/publications/ICWSM-2013-Arjun-Spam.pdf.
[10] Lim E P, Nguyen A V,Jindal N, et al. Detecting product review spammers Using rating behaviors[EB/OL].[2015-05-01]. http://www.cs.uic.edu/~liub/publications/cikm-2010-final-spam.pdf.
[11] Ghose A, Ipeirotis P G. Estimating the helpfulness and economic impact of product reviews: Mining text and reviewer characteristics [J]. IEEE Transactions on Knowledge and Data Engineering, 2011, 23(10): 1498-1512.
/
| 〈 |
|
〉 |
