基于社交网络特征的在线购物系统中协同作弊检测

36 浏览量更新于2024-08-27 收藏 197KB PDF 举报

在当前的在线购物环境中，防止协同作弊行为对维护平台公正性和用户信任至关重要。本研究论文"Detecting Collusive Cheating in Online Shopping Systems through Characteristics of Social Networks"提出了一种创新的方法来检测亚马逊（Amazon）这样的电子商务平台上评级的操纵行为。作者们认为，传统的单纯基于评级数值的检测方法可能忽视了社交网络特性在识别作弊中的关键作用。首先，该研究构建了一个基于用户评级和评论的虚拟社交网络。这个网络将用户的行为模式、交互和反馈联系起来，通过分析用户之间的连接和互动，捕捉到可能存在的异常行为模式。这种社交网络视角有助于揭示隐藏在表面评级数据背后的潜在关系网络，例如，一群用户可能通过相互点赞、评论或互相购买商品来提升彼此的商品评分，从而实现协同作弊。其次，论文着重于探究社交网络中孤立组件（disconnected components）的性质。孤立组件指的是那些在一段时间内没有与其他部分连接的子群体。作者对各种社交网络的孤立组件进行深入研究，包括它们的存活时间和最终规模，以此来识别那些可能预示着潜在作弊活动的特定模式。例如，如果发现一些孤立组件突然快速增长或在短时间内迅速消失，这可能是作弊行为的信号，因为正常情况下，这些组件的增长应该遵循自然的社交网络动态。此外，该研究还可能探讨了其他社交网络特征，如用户的影响力（hub nodes）、社区结构、以及用户行为的异常模式（如集中或分散的评级行为）。通过机器学习算法和网络分析技术，研究人员能够建立预测模型，以识别出具有高度可能性的作弊行为，并及时采取措施进行干预。这篇论文对于改进在线购物系统的安全性具有重要意义，它不仅关注评级数据本身，更深入挖掘了社交网络中的行为特征，为我们提供了一种更加全面和有效的方式来识别并防止协作作弊现象。这一研究对于电商平台运营商、数据分析人员以及网络安全专家都具有重要的参考价值。

Detecting Collusive Cheating in Online Shopping

Systems through Characteristics of Social Networks

Jianwei Niu

, Lei Wang

, Yixin Chen

, Wenbo He

State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, China

McGill University, Canada

niujianwei@buaa.edu.cn, wangleildmc@gmail.com, chenyixinabel@gmail.com, wenbohe@gmail.com

Abstract—Detecting the collaborative cheating in an online

shopping system is an important but challenging issue. In this

paper, we propose a novel approach to detect the collusive

manipulation on ratings in Amazon, an online shopping system.

Rather than focusing on rating values, we believe the online

shopping and rating activities have nontrivial attributes in terms

of social network connections. Our major contributions include:

(a) We build a virtual social network based on users’ ratings and

comments, and detect the collusive cheating based on the social

network activities. (b) We investigate the properties of disconnect-

ed components in a wide range of social networks, such as the

longevity and ﬁnal size of the disconnected components before

they join the giant connected component or merge with other

disconnected components. (c) We apply our proposed collusion

detection algorithm to detect the possible collusive cheating on

the ratings based on the data we crawl from Amazon, and the

experimental results validate our approach.

I. INTRODUCTION

With the increase of the popularity and scale of the online

shopping systems such as Amazon, eBay and Taobao, reputa-

tion systems have been widely deployed to protect the honest

buyers and the reputable sellers through the customer ratings,

reviews, and the user recommendations and referrals. However,

the reputation systems are in general vulnerable to many types

of attacks [6], where selﬁsh users try to manipulate the rating. A

typical example is that a group of users collaboratively subvert

the rating of a given product or service. Another example is

so-called Sybil attack [1] where a user creates a large number

of identities and uses them to gain a large inﬂuence [9]. With

the desire to promote their own products, users have strong

motivation to cheat by providing unfair ratings.

Detecting the unfair ratings introduced by collaborative ma-

nipulation (from either a single Sybil attacker or a group of

collusive users) is an important but challenging issue. Previous

efforts have been based on the investigation of rating values

[18]. For example, if a rating value is far away from the

majoritys opinion, or if we observe a sudden change of the

rating values in a short time period, there is likely to be an

attack manipulating the rating scores. The rating-value-based

detection schemes usually make several assumptions such as

(1) the number of manipulated ratings is less than the number

of honest ratings; (2) the bias of the manipulated ratings is

sufﬁciently large; (3) there is no bursty rating input.

In this paper, we propose a novel approach to detect the

collusive manipulation on ratings in real world. Rather than

focusing on rating values, we believe that the online shopping

and rating systems have the common attributes as the online

social networks in terms of social connections. Hence, we

ﬁrst model the normal behavior of the honest users in online

shopping and rating systems as a user in a virtual social

network, and we identify characteristics of normal users from

a wide range of online social networks. We then check if a

user’s behavior in the virtual social network deviates from the

observed normal behavior to identify suspicious users for future

scrutiny. The contributions of this paper are summarized as

follows:

(1) We build a virtual social network based on users’

ratings and comments, and detect the collusion based on the

characteristics of the virtual social networks. We observe that

collusive users trying to promote or badmouth a speciﬁc group

of products, usually by using a special account to do so and

the account may not be used frequently. So in the virtual social

network, if we connect two users when they make similar

comments or ratings on the same product or user, the collusive

users are likely to form disconnected components. By modeling

the collusive users in this way, we employ the characteristics of

online social networks to detect suspicious users in the online

shopping and rating systems.

(2) We investigate the characteristics of the disconnected

components in a wide range of social networks, such as the

longevity and size of the disconnected components. We observe

that the disconnected components exhibit similar characteristics

in different social networks, hence we obtain a generic model

for normal social network users. Then, we detect the collusion

by checking if the disconnected components in the virtual social

network behave signiﬁcantly different from the model.

(3) To evaluate the proposed approach, we crawl the Amazon

data including 7,659,294 transactions and 123,384 users. We

applied the proposed collusion detection algorithm to detect the

possible collusive manipulation on the ratings, and found 836

suspicious users. After the closer investigation by people, 704

of detected users are considered cheating. We believe that the

proposed approaches have a great impact on collusion detection

in online shopping systems.

The rest of this paper is organized as follows. Section II

introduces the related work in this ﬁeld. Section III describes

how to model an online shopping and rating system as a virtual

social network. In Section IV, we study a wide range of online

social networks in the real world and show the characteristics of

them. We propose a collusion detection algorithm in Section V,

2014 IEEE INFOCOM Workshop on Dynamic Social Networks

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38507923

粉丝: 3
资源: 952

基于社交网络特征的在线购物系统中协同作弊检测

Detecting Anomalous Emotion through Big Data from Social Networks Based on a Deep Learning Method

Detecting Influential Nodes Incrementally and Evolutionarily in Online Social Networks

Detecting anomalous emotion through big data from social networks based on a deep learning method

Detecting community structure in bipartite networks based on matrix factorisation

1975.Detecting instrumrent malfunctions in control systems.pdf

Detecting Parkinson's diseases via the characteristics of the intrinsic mode functions of filtered electromyograms

Towards social botnet behavior detecting in the end host

Iteration method for detecting disease genes in terms of the integration of the cellular compartment information with the protein-protein interaction data

Detecting Privacy Leaks in iOS Applications

Detecting Regime Change in Computational Finance

最新资源