摘要
虚假评论的检测与治理,对净化网络环境具有重要的意义。针对现有的虚假评论检测方法收敛速度慢、准确率低的问题,提出一种融合评论情感特征的虚假评论检测方法。首先,基于预先构建的情感词典,设计情感特征抽取方法抽取情感特征;其次,引入Transformer模型对融合情感特征后的嵌入表示提取特征向量;最后,将特征向量送入全连接层并利用softmax函数实现真实评论与虚假评论分类。采用Amazon数据集,设计实验验证了基于情感词典所提情感特征的有效性,在融合情感特征后准确率提升了1.19百分点;同时与深度学习LSTM方法相比,该方法检测准确率提高0.59百分点。
The detection and governance of fake reviews is of great significance to network environment purification.Considering there exist slow convergence and low accuracy problems in the existing fake review detection methods,this paper proposes a method of fake review detection fusing sentiment features of reviews.Firstly,based on the preconstructed emotional dictionary,a sentiment feature extraction method is designed to extract sentiment features.Secondly,the trans-former model is introduced to extract the feature vector from the embedded representation after the fusion of sentiment features.Finally,the feature vector is sent to the full connection layer,and the softmax function is used to classify real reviews and fake reviews.With Amazon dataset,experiments are designed to verify the effectiveness of the proposed sentiment features based on the emotional dictionary.After the fusion of sentiment features,the accuracy is improved by 1.19 percentage point.Compared with the LSTM method,the accuracy is improved by 0.59 percentage point.
作者
曹东伟
李邵梅
陈鸿昶
张建朋
张桥
CAO Dongwei;LI Shaomei;CHEN Hongchang;ZHANG Jianpeng;ZHANG Qiao(Zhongyuan Network Security Research Institute,Zhengzhou University,Zhengzhou 450001,China;Information Engineering University,Zhengzhou 450001,China)
出处
《信息工程大学学报》
2021年第3期326-330,共5页
Journal of Information Engineering University
基金
国家自然科学基金资助项目(61521003)
国家自然基金青年基金资助项目(62002384)。
