首页|广告虚假好评和返利虚假好评的识别:兼有情绪与文本双重特征的模型框架

广告虚假好评和返利虚假好评的识别:兼有情绪与文本双重特征的模型框架

扫码查看
在线商品广告虚假好评和返利虚假好评干扰了数字经济的良性发展.本文目的是建立兼有情绪与文本双重特征的模型框架,以识别两类虚假好评和真实好评.通过网络爬虫收集真实评论,依据对评论撰写者的调研实践提取标注规则,建立两类虚假好评与真实好评的中文数据集.引入PAD 情绪理论解构评论蕴含的情绪,结合情绪强化调节影响,构造评论的情绪特征.利用n-gram分词和TF-IDF向量化评论文本,运用Boruta 方法构建多维文本特征.采用18 种主流分类算法构建多类好评的分类模型.实验与对比分析显示:两类虚假好评与真实好评在情绪强化调节影响分布与PAD三维度上存在显著差异,据此提出一种评价好评情绪的可行建模方法;提取的情绪与文本特征使得分类算法均可有效识别三类好评,体现所构建模型框架对分类算法的较低依赖性;引入多维情绪特征能够显著提升分类算法对三类好评,尤其可增强对隐蔽性较强的返利虚假好评的辨识力,体现情绪特征对文本特征的增益效果.本研究结论为电商平台改进虚假好评过滤机制与消费者识别两类虚假好评提供参考借鉴.
Identification of Advertising and Rebate Fake Positive Reviews:A Model Framework with Both Emotional and Text Features
Online product reviews have become an important reference for online purchases.However,this value also triggers businesses to intentionally create fake positive reviews to mislead consumers'purchasing decisions.As social e-commerce extends to all aspects of life,fake positive reviews have also shown character-istics of scale and industrialization,seriously restricting the healthy development of the digital economy.There are currently two types of fake positive reviews:"advertising fake reviews"written by spe-cialized swiping companies hired by merchants for advertising pro-motion,and the other is"rebate fake reviews"written by consumers according to merchant requirements to obtain rebates.The extant research on identifying fake positive reviews mainly focuses on advertising fake positive reviews.However,due to the concealment expression of rebate fake positive reviews,existing methods have limitations in identifying the two types of fake positive reviews.Therefore,it is important to study the identification methods of the two types of fake positive reviews to establish a good online busi-ness environment.This paper constructed a model framework that combines emotional and text features to identify two types of fake positive reviews and authentic ones.First,labeling rules were extracted based on the practice of review writers,and Chinese datasets were established for two types of fake positive reviews and authentic ones.Afterward,the PAD(Pleasant-Arousal-Dominance)emotion theory was intro-duced to deconstruct the emotions contained in reviews,combined with the emotional intensification modifier to construct the emo-tional features of reviews.Meanwhile,n-gram segmentation and TF-IDF were utilized to vectorize reviews to construct text features.Then,eighteen mainstream classification algorithms were used to construct a recognition model for multi-class positive reviews.The research results found significant differences in the distribution of the emotional intensification modifier effects and the three dimen-sions of PAD between two types of fake positive reviews and au-thentic ones.Introducing multidimensional emotional features can significantly improve classification algorithms'ability to identify three types of positive reviews,especially the highly concealed re-bate fake positive reviews.The influence of the emotional intensifi-cation modifier helps to improve the classification algorithm's abili-ty to distinguish positive reviews with extreme emotional intensity,while the three dimensions of the PAD theory makes classification algorithms capable of distinguishing three types of positive reviews with various degrees of emotional intensity.This paper extends previous literature from the following aspects.First,it introduces rebate fake positive reviews,expands the types of objects for research on fake positive reviews,and proposes a la-beling method close to the real transaction environment,providing a data basis for in-depth research on identifying fake reviews.Second,it introduces the PAD emotion theory to expand the emotional eval-uation of reviews from the positive-negative dimension to include arousal-non-arousal and active-passive dimensions and examines the impact of the emotional intensification modifier,improving the three-dimensional level of emotional modeling in reviews.Third,it constructs a fake positive review evaluation model that includes both text and emotional features,which exhibits the characteristic of not relying on specific classification algorithms in identifying three types of positive reviews,helping to establish a filtering mechanism for identifying different fake positive reviews.Fourth,it reveals the unique value of emotional and text features of reviews in identifying two types of fake positive reviews and authentic ones.The emotional evaluation and recognition model for multiple types of positive reviews presented in this paper has outstanding practical significance.First,e-commerce platform operators can enhance the identification ability of existing review filtering mechanisms for different types of fake positive reviews based on the fake positive review identification framework proposed in this paper.Second,consumers can preliminarily identify the type of positive reviews and consequently filter out products and merchants based on the multidimensional emotional characteristics reflected in the reviews.Third,the framework of the fake review recognition method pro-posed in this paper can also be extended to other scenarios of fake information recognition using text as a carrier.

Advertising Fake Positive ReviewRebate Fake Positive ReviewAuthentic Positive ReviewEmotion theoryReview Text

李岩、林树海、牟博佼

展开 >

中国矿业大学(北京)管理学院

中国地质大学(北京)经济管理学院

广告虚假好评 返利虚假好评 真实好评 情绪理论 评论文本

中国矿业大学(北京)越崎青年学者项目中央高校基本科研业务费项目国家自然科学基金项目

800015Z11A2159012104771972177

2024

南开管理评论
南开大学国际商学院

南开管理评论

CSTPCDCSSCICHSSCD北大核心
影响因子:3.438
ISSN:1008-3448
年,卷(期):2024.27(5)