融合多特征的垃圾评论检测模型  

Spam Review Detection Model Fusing Multiple Features

作　　者：原福永[1] 刘宏阳 王领 冯凯东 黄国言[1] YUAN Fu-yong;LIU Hong-yang;WANG Ling;FENG Kai-dong;HUANG Guo-yan(School of Information Science and Engineering,Yanshan University,Qinhuangdao 066004,China)

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《小型微型计算机系统》2020年第3期539-543,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61772451)资助.

摘　　要：在线评论已经是影响用户决定是否购买该产品或者服务的重要因素,因而一些不法分子会创建虚假、恶意的评论,对用户和商家造成不良的影响,所以能够快速准确的检测垃圾评论是一个很急迫的需求.已有的研究主要是针对评论文本进行分析,忽略了其它的外部特征并且在准确性上有待提高.本文在评论文本的基础上,考虑了评论者的特征和评论的商品的特征,提出了一种融合多特征的垃圾评论检测模型将三个特征统一考虑进行垃圾评论的检测.首先,使用融入全局-局部注意力机制的卷积神经网络构建评论特征提取模型;其次,分别使用神经网络及卷积神经网络构建评论者及商品特征提取模型;最后,将三个特征模型融合,构成垃圾评论检测模型.通过在真实的数据集上测试证明了本模型的有效性.Online review is already an important factor affecting users’ decision to purchase the product or service. Therefore,some criminals will create false or malicious review,which will have adverse effects on users and businesses. Therefore,it is urgent to detect spam review quickly and accurately. The existing research mainly analyzes the review text,ignores other external features and needs to be improved in accuracy. On the basis of the review text,this paper considers the characteristics of the reviewer and the characteristics of the product of the review,and proposes a spam review model that combines multiple features for the detection of spam review. Firstly,the feature extraction model is constructed by convolutional neural network integrated with the global-local attention mechanism.Secondly,the neural network and convolutional neural network are used to construct the reviewer and product feature extraction model respectively. Finally,the three feature models are merged to form a spam review model. The validity of the model is demonstrated by testing on real data sets.

关 键 词：多特征 卷积神经网络 全局-局部注意力机制 垃圾评论检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]