虚假商品评论识别的研究与进展

受经济利益驱使,大量恶意用户发布包含不实内容的虚假评论以影响用户的购买决策,从而提高自身商品的销售业绩并打压竞争对手,严重扰乱电子商务运营秩序。为此,介绍虚假评论识别的研究成果,包括虚假评论内容、发布者及虚假评论者群组的识别,对识别过程所使用的特征及检测方法进行对比分析,并给出虚假评论识别效果的评价方式和指标。在此基础上,对未来虚假评论识别研究工作进行探讨和展望。     

基于情感特征和用户关系的虚假评论者的识别

随着电子商务的迅速发展,人们越来越亲睐于网上购物。在网上购物之前,消费者往往会参考该产品相关的评价以决定是否购买。因此虚假评论者的识别具有非常重要的意义。基于虚假评论者和真实评论者在情感极性上存在的差异,在特征建模过程中增加了评论文本的情感特征,并结合用户之间对于特定商品之间的关系,创建了一个多边图的模型并提出了一种识别虚假评论者的方法。实验结果验证了该算法的有效性。     

融合时间特征的虚假评论检测研究

对电商的虚假评论检测,需要充分考虑时间突发特性,因此提出了一种融合时间特征的虚假评论检测方法。基于局部异常因子算法构建时间特征指标,将时间特征、评论特征、评论者特征三者相结合,构建一个较为全面的虚假评论识别框架。通过Yelp数据集验证该方法的有效性,结果表明,该方法的性能较好,AUC值较高。     

基于HDXG算法的虚假评论识别方法

为了提高商品虚假评论的识别效果,提出了一种基于习惯偏差与xgboost算法的虚假评论识别方法。首先,通过提出新的算法来计算情感极性,同时在位置因素的基础上加入本地化情感词,从而提高评论文本情感极性计算的精准度。然后,提出新的用户习惯偏差指标和商家异常波动区间值并将其与几维重要特征融合在一起,进而得到一个关于评论-评论者-商户三者特征的新模型。最后,再与xgboost算法进行结合完成虚假评论的检测。实验结果证明,其能更有效识别虚假的评论信息,为消费者提供更加安全有价值的参考信息。     

基于虚假评论识别的微博评论情感分析的研究与应用

微博作为时下热门的社交网络平台,针对其所产生的评论文本进行情感分析已经成为人工智能领域的一个研究热点。考虑到虚假评论会降低情感分析的准确度,从评论用户的状态和行为出发,提出一种基于用户状态与行为的可信度评价体系,用于提取虚假评论特征。结合该特征与PU(Positive and unlabeled)学习算法进行虚假评论识别;运用SVM分类器和随机梯度下降回归模型对去除虚假评论的文本进行主观句分类与情感分析。实验表明,进行虚假评论识别后的情感分析准确率、召回率分别达到0.88和0.89,比传统方法具有更高的分析效能。     

基于梯度提升决策树的微博虚假消息检测

微博是信息共享的重要平台,同时,也成为虚假消息产生和推广的重要平台,虚假消息的传播严重扰乱了社会秩序。为了快速、有效地识别微博虚假消息,提出一种基于梯度提升决策树（GBDT）的虚假消息检测方法。首先,从评论的角度分析微博虚假消息和真实消息之间存在的差异,在此基础上提取评论中的文本内容、用户属性,信息传播和时间特性的分类特征;然后,基于分类特征,采用GBDT算法实现微博虚假消息识别模型;最后,在两个真实的微博数据集上进行验证。实验结果表明,基于GBDT的识别模型能有效提高微博虚假消息检测的准确率。     

考虑结构与行为特征的水军群组检测算法

在线评论对用户的购买决策有重要的影响作用,部分卖方为提高自身信誉或贬低竞争对手的产品,通过雇佣大量水军有组织、有策略地撰写虚假评论来误导潜在消费者。为了检测这种有组织的水军群组,提出了一个综合考虑网络结构与评论者的行为特征水军群组检测算法。首先,根据评分和评论时间相关性得到评论者之间的紧密度,构建评论者关系图;其次,基于构建的评论者关系图,利用标签传播方法检测社区,得到候选群组集合;最后,复原候选群组对应的二部图,以对比可疑度为评估指标,在每个二部图上找到最终的造假者。基于真实数据集的实验结果证明了该算法的有效性。     

虚假评论识别研究综述

Web 2.0时代,消费者在在线购物、学习和娱乐时越来越多地依赖在线评论信息,而虚假的评论会误导消费者的决策,影响商家的真实信用,因此有效识别虚假评论具有重要意义.文中首先对虚假评论的范围进行了界定,并从虚假评论识别、形成动机、对消费者的影响以及治理策略4个方面归纳了虚假评论的研究内容,给出了虚假评论研究框架和一般识别...     

基于DCA算法的微博虚假信息检测

通过对微博虚假信息的分析,基于DCA算法的思想,提出一种检测微博虚假信息的方法。以新浪微博为例,从虚假信息发布者的用户属性和虚假信息评论的文本内容两个方面进行分析。从用户方面选取用户的特征属性,如是否认证、有无简介、地址信息是否详细、关注数、粉丝数等,从评论内容方面选取评论与微博内容的相关性、评论的支持性及其置信度等特征属性,将以上属性的分析结果作为区别虚假信息与真实信息的特征信号,并基于树突状细胞算法（Dendritic Cells Algorithm, DCA）实现新浪微博虚假信息的识别。使用新浪微博真实数据对算法有效性进行了验证和对比实验,结果表明该方法能够有效检测出新浪微博中的虚假信息,具有较高的检测准确率。     

融合情感极性与信任函数的虚假评论检测方法

在线评论是用户判断商品质量的一个依据。虚假评论严重影响了消费者的购买行为,现有的虚假评论检测方法从文本出发,忽略了评分的虚假性,评分通常是不精确和不确定的,对虚假评论检测效果不佳。提出融合情感极性与信任函数的虚假评论检测方法(EP-BFRD),利用信任函数处理给定评论者评分中的不确定性和不准确性,考虑与其他评分者提供的评分的相似性,以检测误导性,并判断评论文本情感极性与评分一致性。综合考虑信任函数处理的结果以及评分与文本情感一致性的结果来判断评论的虚假性。在一个真实的数据库上进行实验,实验表明该方法可有效解决虚假评论检测问题。     

垃圾商品评论检测研究综述

商品评论对消费者的购买意愿有明显导向作用,欺诈者可杜撰评论来过度褒奖或恶意贬低商品,以此来促进己方或是打击对方的商品销售,垃圾商品评论检测成为了一项迫切需要的技术。首先将相关研究分为以评论内部（文本特征）为中心和以评论外部（文本特征）为中心的两大类,然后分别综述它们在特征选择、学习方法上的研究进展,并介绍了垃圾商品评论检测领域的常用评论数据集,在此基础上,展望了该领域的热点研究方向。     

A trustworthy online recommendation system based on social connections in a privacy-preserving manner

Certain consumer websites provide reviews from previous buyers to help new customers make purchasing decisions. However, fake reviews can have an adverse impact on user trust. Most previous suggestions for addressing this problem are still subject to various security concerns in terms of privacy, reliability, and authenticity. To ensure the security of online review systems, this paper proposes the development of a secure online-evaluation method based on social connections to establish evaluation authenticity and provide protection against evaluation forgery while preserving the reviewer’s identity. The proposed method enables users to recognize evaluations from their friends to identify reviews from more trustworthy sources, and authenticates online reviews to prevent possible forgery. In addition, it preserves the privacy of friendship relationships from application server and other users and identifier relations between the personal identifier and online identifier. The proposed approach can be applied to Internet auctions and online games, and is shown to be secure and efficient, with sufficient matching probability to be practical.     

基于情感极性与SMOTE过采样的虚假评论识别方法*

为了有效识别商品虚假评论,提出一种基于情感极性与SMOTE过采样的虚假评论识别方法。首先,根据在线虚假评论的特点,构建一个多维虚假评论特征模型。其次,在情感极性算法中增加了情感极性均值和情感极性标准差等统计指标来全面刻画虚假评论。最后,针对虚假评论中的类不平衡问题,使用SMOTE算法优化随机森林分类模型,从而提高虚假评论识别效果。基于大众点评网的真实评论数据进行了多组实验,实验结果表明该方法在正负样本不平衡的虚假评论数据集中具有更高的准确率、召回率及F值。综合考虑情感极性和正负样本不平衡等因素可帮助电商平台有效过滤虚假评论,为消费者提供更加真实可靠的评论数据。     

Does certainty tone matter? Effects of review certainty,reviewer characteristics,and organizational niche width on review usefulness

Given the proliferation of online review websites—e.g., Yelp.com—that prominently display a large number of online custome0r reviews, scholars have made efforts to investigate what makes a review “useful.” However, there is little research that offers insight into how review content, reviewer characteristics, and review contexts jointly influence review usefulness. We specially examine the role of review certainty on review usefulness. Drawing on dual-process and social influence theories, we examine the interaction effects of review certainty, reviewer popularity, reviewer expertise, and the niche width of a restaurant on review usefulness. In particular, this study focuses on how review certainty interacts with other contextual factors to influence users’ evaluation of the usefulness of online reviews. Utilizing a zero-inflated negative binomial Poisson regression, we empirically tested our hypotheses based on 10,097 reviews on 2,383 restaurants from Yelp.com. Our results indicated that (1) the impact of review certainty on review usefulness decreases with reviewer popularity but does not vary with reviewer expertise and (2) the niche width of a restaurant—as a contextual feature—interacts with review certainty and reviewer characteristics in influencing review usefulness. Theoretically, these findings contribute to online customer review literature and certainty literature, as well as social media research, provide new guidelines for predicting review usefulness, and add new insights into understanding the role of organizational positioning for customer evaluations. In practice, our findings can help online review platforms better understand how to screen and select useful reviews for visitors.     

Leveraging Readability and Sentiment in Spam Review Filtering Using Transformer Models

Online reviews significantly influence decision-making in many aspects of society. The integrity of internet evaluations is crucial for both consumers and vendors. This concern necessitates the development of effective fake review detection techniques. The goal of this study is to identify fraudulent text reviews. A comparison is made on shill reviews vs. genuine reviews over sentiment and readability features using semi-supervised language processing methods with a labeled and balanced Deceptive Opinion dataset. We analyze textual features accessible in internet reviews by merging sentiment mining approaches with readability. Overall, the research improves fake review screening by using various transformer models such as Bidirectional Encoder Representation from Transformers (BERT), Robustly Optimized BERT (Roberta), XLNET (Transformer-XL) and XLM-Roberta (Cross-lingual Language model–Roberta). This proposed research extracts and classifies features from product reviews to increase the effectiveness of review filtering. As evidenced by the investigation, the application of transformer models improves the performance of spam review filtering when related to existing machine learning and deep learning models.     

The determinants of helpfulness of online reviews

More and more people are gravitating to reading online product reviews prior to making purchasing decisions. Because a number of reviews that vary in usefulness are posted every day, much attention is being paid to measuring their helpfulness. The goal of this paper is to investigate the various determinants of the helpfulness of reviews, and it also intends to examine the moderating effect of product type, that is, the experience or search goods in relation to the helpfulness of online reviews. The study results show that reviewer reputation, the disclosure of reviewer identity, and review depth positively affect the helpfulness of an online review. The moderating effects of product type exist for these determinants on helpfulness. That is, the number of reviews for a product and the disclosure of reviewer identity have a greater influence on the helpfulness for experience goods, while reviewer reputation, review extremity, and review depth are more important for helpfulness in relation to search goods. The interaction effects exist for average review rating and average review depth for a product with review helpfulness on product sales. The results of the study will identify helpful online reviews and assist in designing review sites effectively.     

在线评论质量有用特征识别: 基于GBDT特征贡献度方法

面对海量的在线评论,有用特征识别有助于消费者选择高质量的评论,为合理决策提供支持。该文基于信息采纳模型理论,在数码相机和手机数据集上提取了四类影响评论质量的有用特征集合,以logistic岭回归和基本decision tree模型作为基准模型,并结合递归特征消除(RFE)降维方法,比较检验了GBDT模型对评论质量分类和特征降维上的表现,揭示了各特征项对评论质量分类结果的“贡献度”,进而识别关键特征。实验结果表明,基于GBDT模型对评论质量分类效果较好,评论发表时间、评论者排名、关键特征数量、评论字数是影响评论质量的关键特征。