摘要
随着电商经济发展迅猛,各类电商平台为提升服务品质非常重视对用户评论信息的收集、分析和利用.但各电商平台都存在“刷评论”现象,虚假评论、默认好评等因素导致平台无法获取用户的真实情感,严重影响服务质量的提升.为了更加准确地获取用户的真实情感,提出一种融合多元评论信息的用户情感分类方法.首先,对评论中的文本和图片进行分类,构建去除虚假评论的图文数据集;其次,对原始和追加评价文本进行分割和重构;最终,将预处理后的原始评论、原始图片、追加评论、追加图片等多元评论信息输入到多通道卷积神经网络中,经训练后得到用户情感分类模型.经对比实验验证,融合多元评论信息的用户情感分类方法准确率可以达到96%,优于现有主流情感分类方法.
In recent years,the e-commerce economy has developed rapidly,and various e-commerce platforms attach great importance to the collection,analysis,and utilization of user comment information in order to improve service quality.However,all e-commerce platforms have the phenomenon of“swipe comments”.Factors such as false comments and acclaimed reviews cause the platform to fail to obtain the true emotions of users and seriously affect the improvement of service quality.In order to obtain the user′s true emotions more accurately,a user emotion classification method based on multiple comment information is proposed.First,classify the text and pictures in the comments to construct a graphic data set that removes false comments;second,segment and reconstruct the original and additional evaluation texts;finally,the pre-processed original comments,original pictures,and appends Multiple comment information such as comments and additional pictures are input into the multi-channel convolutional neural network,and the user emotion classification model is obtained after training.It has been verified by comparative experiments that the accuracy rate of user sentiment classification methods incorporating multiple comment information can reach 96%,which is better than the existing mainstream sentiment classification methods.
作者
徐红艳
黄法欣
冯勇
XU Hong-yan;HUANG Fa-xin;FENG Yong(College of Information,Liaoning University,Shenyang 110036,China)
出处
《辽宁大学学报（自然科学版）》
CAS
2021年第2期97-107,F0002,共12页
Journal of Liaoning University：Natural Sciences Edition
基金
辽宁省档案科技项目(L-2016-R-7)
辽宁省社会科学规划基金项目(L18AGL007)
吉林大学符号计算与知识工程教育部重点实验室项目(93K172018K01)
赛尔网络下一代互联网技术创新项目(NGII20190301)。
关键词
情感分类
评论
图片分类
文本分割
CNN
Sentiment classification
reviews
image classification
text segmentation
CNN
