-
Social network is a system in which a group of people or groups are connected in some way. The interaction of humans in social networks forms a huge and complex human-machine system. With the development of large-scale networks, the network security problem has become a hot issue involving information leakage and hostile attacks of individuals, organizations, society and countries due to the openness and sharing of network. It’s a challenge to provide solid theory that can then be formulated as scalable algorithms and software implementable in practical systems with potential security issues, which has attracted widespread attention of researchers from diverse areas. Over the past decades, many researches on security issues in social networks have emerged, and some works focus on the network performance analysis under typical attacks[1-4] including Sybil attack[5], de-anonymization attacks[6], or inference attacks[7], and so on. Meanwhile, with the increasingly sophisticated defense system, the hostile attacks have evolved into more complicated and diverse formations. More and more types of attacks appear, such as mutual friend attacks[8], attribute couplet attacks[9], composition attacks[10], which have greatly threatened social security and information privacy.
In social networks, positive and secure public opinions play a vital role in establishing a healthy and stable society. A hybrid opinion dynamics comprising averager, copier, and voter agents is studied in Ref.[11]. An adaptive bridge control strategy is proposed to control a special kind of bridge nodes for some other targeted immunization and acquaintance immunization in Ref.[12]. In recent years, the researchers from diverse areas have proposed some models to capture the properties of opinion dynamics and simulate the opinion evolution rule, which can explain the social phenomena such as few opinions surviving[13-14], clustering[15], consensus[16]. The analysis of social opinion in physical models[17] investigate the evolution of public opinion suffered malicious opinion injection. Actually, the evolution of public opinion has been extensively studied by sociologists, which can trace back to the works of Ref.[18] and Ref.[19]. A classical model is proposed by Ref.[20], which studies the process of social group reaching consensus by pooling their opinions. To better explore the internal factors of opinion dynamics, Ref.[21] studied a model (called FJ model) as an improvement of the DeGroot model to describe a phenomenon that the individuals insist on their initial biases, i.e., considering the individuals’ inherent prejudices. In FJ model, disagreements definitely emerge under the influence of prejudices, and an extended work also pointed out the importance of diverging research in Ref.[22]. Recently, Ref.[23] studied a novel multi-dimensional model which describes the opinion dynamics in social networks on several interdependent topics. In addition, Hegselmann-Krause(HK) model[24] and Deffuant-Weisbuch(DW) model[25] were proposed to consider prejudice assimilation and bounded confidence, where each individual only communicates with those individuals whose opinions are close enough.
With the development of social networking tools such as Twitter, Facebook, WeChat, et.al, people are increasingly inclined to express their opinions freely. However, the convenience of free expression also brings hidden danger[26-27]. For example, the intruders could spread malicious opinions among the individuals through the social networking tool, which leads to serious consequences. Specifically in recent years, many network rumor events have been reported frequently. Hence, it is necessary to study the opinion dynamic evolution in social networks with hostile opinion injection. Some similar works can refer to the information diffusion[28]. Based on the traditional DeGroot model, we develop a modified model by considering the malicious opinion injection, where parts of ‘target’ individuals are injected into malicious opinions from the intruder unconsciously. We also investigate the ‘target’ individual selection problem that the intruder can select those ‘key’ individuals as target individuals such that the convergence speed of reaching consensus on the malicious opinion is maximized. We further design a defense mechanism based on statistical learning to resist the malicious opinion injection.
Notation.
${\rm{Tr}}({{\varGamma}} )$ is the trace of matrix${{\varGamma}} $ . For a finite set${{V}}$ , we denote the cardinality of${{V}}$ by$\left| {{V}} \right|$ .
Security Analysis of Opinion Dynamics in Social Networks
doi: 10.12178/1001-0548.2019234
- Received Date: 2019-10-22
- Rev Recd Date: 2020-08-10
- Available Online: 2020-11-25
- Publish Date: 2020-11-23
-
Key words:
- defense mechanism /
- network security /
- node selection strategy /
- opinion dynamics
Abstract: This paper concerns the opinion evolution dynamics with malicious opinion injection in social networks. In the study, a modified DeGroot model is proposed by considering a real-time opinion injection from an intruder, and it is proved that the opinions of all the individuals converge to the opinion of the intruder. A set of ‘key’ individuals influenced by the intruder is then found such that the convergence speed on the malicious opinion is maximized. Further, a defense mechanism for each individual is proposed and the steady-state opinion gap of the individuals is obtained. Mumerical examples show the relation of the node role and the opinion convergence speed, and verify the effectiveness of the defense mechanism.
Citation: | SU Shuang-ping, YANG Wen, ZHAO Zhi-yun. Security Analysis of Opinion Dynamics in Social Networks[J]. Journal of University of Electronic Science and Technology of China, 2020, 49(6): 924-933. doi: 10.12178/1001-0548.2019234 |