面向豆瓣网影评的水军识别研究  

Research on the identification of water army facing douban.com film critics

作　　者：王亚[1] WANG Ya(College of Information Engineering,Xuchang University,Xuchang Henan 461000,China)

机构地区：[1]许昌学院,河南许昌461000

出　　处：《智能计算机与应用》2020年第6期218-220,224,共4页Intelligent Computer and Applications

基　　金：许昌学院校级科研项目(2020YB017);河南省高等院校重点科研项目(19A520036)。

摘　　要：豆瓣网影评中存在很多恶意刷高分或恶意贬低电影的水军用户,这些用户的存在严重干扰了正常用户使用豆瓣网进行信息的获取与交流。本文通过分析豆瓣网水军用户与正常用户的区别,提取用户静态行为特征和动态行为特征,采用逻辑回归的二分类方法对这两类特征进行模型训练,并将这两类特征融合进行模型训练,以最大程度的提高识别影评水军用户的有效性。实验表明,本文抽取的这两类特征对水军用户识别的准确率最高可达近67.8%。There are many spam users who maliciously brush high scores or maliciously degrade movies in Douban website.The existence of these users seriously interferes with normal users'use of Douban for information acquisition and communication.This paper analyzes the differences between Douban spam users and normal users,extracts user static behavior characteristics and dynamic behavior characteristics,uses logistic regression binary classification to perform model training on these two types of features,and fuses the two types of features for model training,which maximize the effectiveness of identifying film review spam users.Experiments show that the accuracy of the two types of features extracted in this paper for spam users is up to 67.8%.

关 键 词：水军用户 特征提取 逻辑回归 

分 类 号：TP305[自动化与计算机技术—计算机系统结构]