Processing math: 100%
Research article Special Issues

An SEIR model for information propagation with a hot search effect in complex networks

  • We formulate an SEIR model for information propagation with the effect of a hot search in complex networks. Mathematical analysis is conducted in both a homogeneous network and heterogenous network. The results reveal that the dynamics are completely determined by the basic propagation number if the effect of a hot search is absent. On the other hand, when the effect of a hot search is taken into account, there exists no information-free equilibrium, and the information-propagating equilibrium is stable if the threshold is greater than 1. Numerical simulations were performed to examine the sensitivity of the parameters to the basic propagation number and the propagable nodes. Furthermore, the proposed model has been applied to fit the collected data for two types of information spreading in Sina Weibo, which confirmed the validity of our model and simulated the dynamical behaviors of information propagation.

    Citation: Xiaonan Chen, Suxia Zhang. An SEIR model for information propagation with a hot search effect in complex networks[J]. Mathematical Biosciences and Engineering, 2023, 20(1): 1251-1273. doi: 10.3934/mbe.2023057

    Related Papers:

    [1] Yuqi Chen, Xianyong Li, Weikai Zhou, Yajun Du, Yongquan Fan, Dong Huang, Xiaoliang Chen . A hot topic diffusion approach based on the independent cascade model and trending search lists in online social networks. Mathematical Biosciences and Engineering, 2023, 20(6): 11260-11280. doi: 10.3934/mbe.2023499
    [2] Fulian Yin, Hongyu Pang, Lingyao Zhu, Peiqi Liu, Xueying Shao, Qingyu Liu, Jianhong Wu . The role of proactive behavior on COVID-19 infordemic in the Chinese Sina-Microblog: a modeling study. Mathematical Biosciences and Engineering, 2021, 18(6): 7389-7401. doi: 10.3934/mbe.2021365
    [3] Minkyoung Kim, Soohwan Kim . Dynamics of macroscopic diffusion across meta-populations with top-down and bottom-up approaches: A review. Mathematical Biosciences and Engineering, 2022, 19(5): 4610-4626. doi: 10.3934/mbe.2022213
    [4] Fulian Yin, Jinxia Wang, Xinyi Jiang, Yanjing Huang, Qianyi Yang, Jianhong Wu . Modeling and analyzing an opinion network dynamics considering the environmental factor. Mathematical Biosciences and Engineering, 2023, 20(9): 16866-16885. doi: 10.3934/mbe.2023752
    [5] Hong Wu, Zhijian Zhang, Yabo Fang, Shaotang Zhang, Zuo Jiang, Jian Huang, Ping Li . Containment of rumor spread by selecting immune nodes in social networks. Mathematical Biosciences and Engineering, 2021, 18(3): 2614-2631. doi: 10.3934/mbe.2021133
    [6] Yi Wang, Shicheng Zhong, Guo Wang . Preventing online disinformation propagation: Cost-effective dynamic budget allocation of refutation, media censorship, and social bot detection. Mathematical Biosciences and Engineering, 2023, 20(7): 13113-13132. doi: 10.3934/mbe.2023584
    [7] Meili Tang, Qian Pan, Yurong Qian, Yuan Tian, Najla Al-Nabhan, Xin Wang . Parallel label propagation algorithm based on weight and random walk. Mathematical Biosciences and Engineering, 2021, 18(2): 1609-1628. doi: 10.3934/mbe.2021083
    [8] A. N. Licciardi Jr., L. H. A. Monteiro . A complex network model for a society with socioeconomic classes. Mathematical Biosciences and Engineering, 2022, 19(7): 6731-6742. doi: 10.3934/mbe.2022317
    [9] Jun Zhai, Bilin Xu . Research on meme transmission based on individual heterogeneity. Mathematical Biosciences and Engineering, 2021, 18(5): 5176-5193. doi: 10.3934/mbe.2021263
    [10] Fulian Yin, Xinyi Tang, Tongyu Liang, Yanjing Huang, Jianhong Wu . External intervention model with direct and indirect propagation behaviors on social media platforms. Mathematical Biosciences and Engineering, 2022, 19(11): 11380-11398. doi: 10.3934/mbe.2022530
  • We formulate an SEIR model for information propagation with the effect of a hot search in complex networks. Mathematical analysis is conducted in both a homogeneous network and heterogenous network. The results reveal that the dynamics are completely determined by the basic propagation number if the effect of a hot search is absent. On the other hand, when the effect of a hot search is taken into account, there exists no information-free equilibrium, and the information-propagating equilibrium is stable if the threshold is greater than 1. Numerical simulations were performed to examine the sensitivity of the parameters to the basic propagation number and the propagable nodes. Furthermore, the proposed model has been applied to fit the collected data for two types of information spreading in Sina Weibo, which confirmed the validity of our model and simulated the dynamical behaviors of information propagation.



    With the rapid development of mobile smart terminals, online social networks (OSNs) have become the main platform for Internet users to communicate with each other. The popular OSNs, such as Facebook, Twitter and Sina Weibo, enable people to share information simultaneously with a large number of peers and thus play an important role in information propagation. Modeling the process of information spreading in OSNs has attracted more and more attention to understanding the way information is disseminated and the network structures [1,2,3,4,5,6]. One of the primary means of information dissemination in OSNs is through connections between users of the network. When original real-time information is released on the web, it is possible that a very large number of users have the interest to read, comment and retweet it; then, one user becomes the follower of another and, eventually, millions of individual users can share the information.

    In the new media era, a hot search that tends to focus on people's livelihood and current affairs has become an important source to understand what Internet citizens are concerned about, and it serves as another way to affect information diffusion significantly. A hot search is generally calculated according to the real behaviors of the users in OSNs, and it a real-time list is formed based on data indicators such as search volume, blog volume and interaction volume. Whenever a piece of information is pushed to the hot search list, it is highlighted from a amout of network information and is more accessible to online users within a certain period of time, which encourages people to read and forward it. From this perspective, a mathematical model with a hot search effect may be more reasonable to reveal the important consequences of it in the process of information propagation. To our knowledge, there are few results related to hot searches to describe their effect on the time evolution of information spreading. However, some other characteristics emerged in OSNs have been incorporated into models to improve the understanding of information propagation mechanisms, such as the homepage effect in [7] and super spreader in [8].

    Due to the similarity between the way information spreads and the transmission of infectious diseases, classical epidemic models have been adopted and extended to understand information propagation on social network platforms [7,9,10,11]. For instance, the influence of opinions involving opinion leaders on delayed information propagation dynamics was investigated by establishing the OD-SFI (opinion-delay susceptible-forwarding-immunized) model [10]. In recent years, some scholars have applied complex networks to describe the characteristics of information dissemination and refine the establishing models. Assuming that each node has the same probability of contacting with the other nodes, homogeneous networks have been used in some models to investigate the evolution of different states in information dissemination. Huo et al. proposed a dynamic model of I2SR for the rumor spreading in [12], in which the state nodes were divided into the high rate of active spreaders and the low rate of the ones. To examine the effects of interactions between two rumors, the authors in [13] detailed the propagating dynamics of a nine-state model in homogeneous networks under symmetric conditions. The delays in expert intervention and government action were included in the model of malicious news in [14], which were demonstrated to be a cause of unstable situations of the homogeneous social network. To address the heterogeneity in the information spreading, some studies have focused on modeling the propagation in heterogeneous networks, which contribute to revealing the underlying mechanisms in the spreading process [15,16,17] and identifying control strategies to optimize the network performance [18,19].

    Considering the different factors that affect information dissemination, a variety of models have been proposed to study the dynamic behaviors of the social network system. Zhu et al. investigated the influence of a forced silence function, time delay and network topology on rumor propagation in OSNs comprehensively [20], and they also established a susceptible-believed-denied type model for rumor spreading with a nonlinear incidence rate and time delay in complex networks [21]. To help with designing effective communication strategies during a public hot event, Yin et al. proposed the cross-transmission susceptible-forwarding-immune model to describe the dynamics of co-propagation, particularly with focus on the cross-transmission effects [22]. A multi-scale model was formulated in [23], which explicitly modeled both the disease transmission with a saturated recovery rate and information transmission to evaluate the effect of information transmission on dynamic properties. These studies have characterized the information diffusion and improved the understanding of the dynamics in time evolution.

    In order to study the possible effect of a hot search list in OSNs, we developed a mathematical model for information propagation in complex networks to describe the spreading dynamics, which is different from the existing models. In this model, the hot search is regarded as a propagation source that plays a crucial role in accelerating the spreading and enlarging the influenced range. Particularly, the parameters, including the presence rate and the effect of a hot search, have been introduced to examine the potential impacts of it on information diffusion. The stabilities of the information-free equilibrium and the information-propagating equilibrium were analyzed in homogeneous networks and heterogeneous networks respectively. And, a sensitivity analysis was conducted to investigate the correlation between the concerned parameters and the basic propagation number. In addition, two specific datasets of information dissemination in OSNs were collected for a case study, which validates our proposed SEIR model and illustrates the capacity of it to simulate the dynamical behaviors of information propagation.

    The organization of the remaining part is as follows. In Section 2, we introduce the model framework. In Section 3, the model in homogeneous networks is presented and analyzed in two cases, i.e., with or without the hot search effect. Section 4 focuses on the dynamical behaviors of the model in heterogeneous networks in the same two cases as in Section 3. Numerical simulations are presented in Section 5 to describe the sensitive analysis. In Section 6, the model is demonstrated to fit two types of datasets collected from Sina Weibo. Section 7 gives the summary and conclusions.

    Based on the compartmental model of the classical SEIR type, we formulate a susceptible-hesitant-propagable-rejected model that describes the evolution of information dissemination in OSNs, such as Weibo. All users are divided into classes of susceptible (S), hesitant (E), propagable (I) and rejected (R), respectively. According to the characteristics of online information propagation, the susceptible users refer to the population that have not heard of or known about the target information. When being informed of the related propagation through hotspots, tweets or retweets from other users, but not retweeting or forwarding the topic promptly due to hesitation, the susceptible state shifts subsequently to the hesitant one. Otherwise, if the susceptible have been convinced of the information and choose to retweet it actively, then they progress to the propagable class I. In Class R, the population is in the state of never having access to or no longer being interested in the relevant information. Let S(t), E(t), I(t) and R(t) be the population at time t in the corresponding classes, respectively.

    In OSNs, it is common for users to subscribe or unsubscribe to some topic or login or logout of a certain application program. Consequently, we suppose that all of the new users in the network are susceptible and the individuals in each state of the SEIR model can leave the information propagation progress at the same rate. Let Λ be the recruitment rate into the susceptible compartment and μ be the removal rate from the network. Different from the exposed state in disease transmission, it is optional for the hesitant individuals in information dissemination to forward it and become propagable subjectively, or to neglect it and transfer to the rejected compartment, proportions of which are denoted by ε and γ1, respectively.

    Considering the role of the hot search effect in spreading the information, we introduce a constant rate p for the hot search presence for some topic information in a certain period of time, such as the Weibo Hot Topic List, thus, 1p denotes the average leaving time from S due to the information posted on the hot search list. We assume that there is a proportion of q among pS that is interested in the information and then transfers to the hesitant class. This means that the parameter q reflects the effect of a hot search on susceptible users. The other individuals of the proportion 1q show disinterest and shift to the rejected state. Although the hot search list has an obvious impact on promoting the propagation, it is still possible that direct contacts in OSNs induce a higher proportion of the susceptible population to retweet some topics, since a number of individuals are more likely to trust the people they follow on social platforms.

    The mixed group of users in online social media is considered as a complex network, where each user can be seen as a node in the corresponding state, and their connections are represented by the edges along which the information can spread. The topology of the network has a great influence in the overall behavior of information spreading, and the connectivity fluctuations of the network play a major role by enhancing the incidence of the propagation. Many complex social networks can be classified into homogeneous networks and heterogeneous networks according to their degree distribution. Homogeneous networks are a more idealized type of network. More generally, heterogeneous networks with a more complex topology structure are much closer to real social networks. The process of state transitions in time steps of 0, 5, 15 and 20 is shown graphically in Figure 1, which is on a Barabˊasi-Albert (BA) scale-free network containing 100 nodes. At the very beginning, we place a red node that has forwarded a message in propagable state. The susceptible, hesitant and rejected states are denoted by nodes of blue, yellow and green respectively.

    Figure 1.  Schematic diagram of the state transitions on BA network.

    To clarify the transition process for the status of each user in our model, the flow diagram is illustrated in Figure 2 and the definitions of those parameters are listed in Table 1. All parameters are constants and remain nonnegative.

    Figure 2.  Propagation structure diagram of SEIR model.
    Table 1.  Definitions of parameters in SEIR model.
    Parameter Description
    Λ rate of recruitment into the susceptible class
    μ per capital removal rate from the network
    β information transmitting rate
    α proportion of the incidence moving to hesitant class
    ε transition rate from E to I
    γ1 transition rate from E to R
    γ2 transition rate from I to R
    p rate of hot search presence
    q the effect of hot search

     | Show Table
    DownLoad: CSV

    First, according to the proposed framework of information propagation in Figure 2, we consider the SEIR model in a homogeneous network, which is described by the mean-field theory, as follows:

    {dS(t)dt=Λβ¯kS(t)I(t)μS(t)pS(t),dE(t)dt=αβ¯kS(t)I(t)εE(t)γ1E(t)μE(t)+pqS(t),dI(t)dt=(1α)β¯kS(t)I(t)+εE(t)γ2I(t)μI(t),dR(t)dt=p(1q)S(t)+γ1E(t)+γ2I(t)μR(t), (3.1)

    where ¯k denotes the average degree of the network. The initial conditions of System (3.1) are given as

    S(0)0,E(0)0,I(0)0,R(0)0.

    For the initial conditions, there exists a unique solution for Model (3.1) by the existence and uniqueness theory of ordinary differential equations. In addition, the solution with nonnegative initial conditions is nonnegative. Indeed, dS(t)dt|S=0=Λ>0, dE(t)dt|E=0=αβˉkSI+pqS>0 and the similar inequalities hold for I and R.

    Denote the total number of users as N, i.e., N(t)=S(t)+E(t)+I(t)+R(t). Adding up all equations in (3.1) gives

    dN(t)dt=ΛμN(t),

    which implies limt+N(t)=Λμ. Obviously,

    D={(S,E,I,R)|S,E,I,R0,S+E+I+RΛμ}

    is the positively invariant domain of System (3.1).

    To reveal the hot search effect on the process of information spreading, we conduct dynamic analysis in two cases, i.e., q=0 and q0.

    For the case of q=0, System (3.1) becomes

    {dS(t)dt=Λβ¯kS(t)I(t)μS(t)pS(t),dE(t)dt=αβ¯kS(t)I(t)εE(t)γ1E(t)μE(t),dI(t)dt=(1α)β¯kS(t)I(t)+εE(t)γ2I(t)μI(t),dR(t)dt=pS(t)+γ1E(t)+γ2I(t)μR(t). (3.2)

    For an epidemic model, there is a typical threshold parameter that determines whether the infectious disease will spread, i.e., the basic reproductive number, which gives the expected number of secondary cases which are produced in a completely susceptible population by an infective individual in his infectious period. Due to the similarity, such a threshold, i.e., the basic propagation number, exists in the information propagation model, and the information spread in the network is determined by the magnitude of this parameter.

    By the next generation matrix approach [24], we can calculate the basic propagation number as

    R10=ρ(FV1)=εαβΛ¯k+(1α)βΛ¯k(ε+γ1+μ)(ε+γ1+μ)(γ2+μ)(μ+p),

    where ρ denotes the spectral radius of FV1 and

    F=(0αβΛ¯kμ+p0(1α)βΛ¯kμ+p),V=(ε+γ1+μ0εγ2+μ).

    The value of R10 can be seen as the sum of two separate terms representing secondary infections, which is produced by the hesitant and propagable individuals, respectively.

    It is obvious that System (3.2) has an information-free equilibrium:

    Q0=(Λμ+p,0,0,pΛμ(μ+p)).

    As for the information-propagating equilibrium Q1=(S,E,I,R), it exists if R10>1, where

    S=Λβ¯kI+μ+p,E=αβ¯kSIε+γ1+μ,I=ba,R=pS+γ1E+γ2Iμ,

    with

    a=(ε+γ1+μ)(γ2+μ)β¯k>0,b=(ε+γ1+μ)[(μ+p)(γ2+μ)(1α)βΛ¯k]Λεαβ¯k<0.

    Theorem 3.1. For System (3.2), if R10<1, then the information-free equilibrium point Q0 is globally asymptotically stable, and it is unstable if R10>1.

    Proof. Note that the users in the state R do not appear in other states, thus, the R equation can be neglected in the remaining analysis. We first discuss the local stability of Q0 when R10<1 and its instability when R10>1.

    The characteristic equation of System (3.2) at Q0 is

    f0(λ0):=(λ0+μ+p)(a0λ20+b0λ0+c0)=0,

    with

    a0=μ+p,b0=(μ+p)(ε+γ1+μ)+(α1)βΛ¯k+(γ2+μ)(μ+p),c0=[(α1)βΛ¯k+(γ2+μ)(μ+p)](ε+γ1+μ)εαβΛ¯k.

    It is easy to see that b0>0 and c0>0 if R10<1, then, f0(λ0)=0 has only roots with the negative real part, which certifies the local stability of Q0 by the Hurwitz criterion [25]. Otherwise, if R10>1, then c0<0, so Q0 is unstable.

    Then, consider the Lyapunov function of the following form:

    V(t)=εE(t)+(ε+γ1+μ)I(t).

    Calculating the time derivative of V(t), along with System (3.2), gives

    dV(t)dt|(3.2)=(γ2+μ)(ε+γ1+μ)(R101)I(t).

    If R10<1, it follows that dVdt0 and the equality dVdt=0 holds only at Q0. According to LaSalle's invariance principle [26], the information-free equilibrium Q0 is globally asymptotically stable.

    Theorem 3.2. If R10>1, the information-propagating equilibrium Q1 is locally stable.

    Proof. The Jacobi matrix of System (3.2) at Q1 is

    J1(Q1)=(β¯kIμp0β¯kSαβ¯kIεγ1μαβ¯kS(1α)β¯kIε(1α)β¯kSγ2μ),

    which leads to the characteristic equation

    f1(λ1):=A3λ31+A2λ21+A1λ1+A0=0, (3.3)

    where

    A3=β¯kI+μ+p,A2=(α1)βΛ¯k+(β¯kI+3μ+p+ε+γ1+γ2)(β¯kI+μ+p),A1=(β¯kI+2μ+p+ε+γ1)[(α1)βΛ¯k+(γ2+μ)(β¯kI+μ+p)]+(1α)β¯kIβΛ¯kεαβΛ¯k+(β¯kI+μ+p)2(ε+γ1+μ),A0=[(α1)βΛ¯k+(γ2+μ)(β¯kI+μ+p)](β¯kI+μ+p)(ε+γ1+μ)+(1α)β¯kIβΛ¯k(ε+γ1+μ)εαβΛ¯k(μ+p).

    It is easy to obtain that Ai>0 (i=0,1,2,3) and A1A2A0A3>0 when R10>1. The Hurwitz criteria implies that the characteristic equation at Q1 only has roots with the negative real part, which ensures the local stability of Q1.

    Note that Model (3.1) with q0 does not contain an information-free equilibrium and consequently has no basic propagation number. It is observed that when the threshold R10>1, there exists an information-propagating equilibrium for System (3.1), denoted by Q2=(S,E,I,R), where

    S=Λβ¯kI+μ+p,E=αβ¯kSI+pqSε+γ1+μ,I=b+b24ac2a,R=p(1q)S+γ1E+γ2Iμ,

    with the same a and b as in Q1 and c=Λεpq<0. Assume α>q, which implies that direct contacts in the network attract a higher proportion of the susceptible population to follow some information; then, we obtain the following local stability of Q2.

    Theorem 3.3. For System (3.1), if the threshold R10>1, then the information-propagating equilibrium Q2=(S,E,I,R) is locally asymptotically stable.

    Proof. When neglecting the R equation, the Jacobi matrix of System (3.1) at Q2 is

    J2=(β¯kIμp0β¯kSαβ¯kI+pqεγ1μαβ¯kS(1α)β¯kIε(1α)β¯kSγ2μ).

    Then, the characteristic equation of J2 follows

    f2(λ2):=B3λ32+B2λ22+B1λ2+B0=0,

    where Bi can be obtained by substituting I for I in the expression of Ai(i=3,2,1) in Eq (3.3), and B0=A0+εβΛpq¯k. Notice that I>I leads to Bi>Ai for i=0,1,2,3. If α>q, it can be verified that B1B2B0B3>0. According to the Hurwitz criteria [24], we have the local stability of Q2 when R10>1.

    The results of numerical simulation (see Figure 3) indicate that the hot search in a social network of information propagation has an obvious effect on the evolution of state density in Model (3.1), which is illustrated when q=0 and q=0.3 for the homogenous network with size N=1000. From the comparison, we can see that it is necessary to take the effect of a hot search into consideration in network propagation modeling.

    Figure 3.  Plots of state density with and without the hot search effect. (a): q=0. (b): q=0.3.

    Due to the characteristics in real world networks, specifically in OSNs, heterogeneous networks are used to describe the pattern of information propagation more accurately. In this subsection, in order to take into account the heterogeneity induced by the presence of nodes with different connectivity, all users in the network are partitioned into n groups according to their connections from 1 to n, with n being the maximum number of connectivity. Assuming that the users in the group i have the same degree ki(i=1,2,,n), the densities of the susceptible, hesitant, propagable and rejected nodes of connectivity ki at time t are denoted as Ski(t),Eki(t),Iki(t) and Rki(t), respectively. In addition, we assume that the recruitment rate equals the removal rate in the heterogeneous network, i.e., Λ=μ, and that these state variables of time are connected by means of the normalization condition, Ski(t)+Eki(t)+Iki(t)+Rki(t)=1. Based on the mean field theory, we formulate the SEIR model of information dissemination in a heterogeneous network as follows:

    {dSki(t)dt=ΛβkiSki(t)Θ(t)μSki(t)pSki(t),dEki(t)dt=αβkiSki(t)Θ(t)εEki(t)γ1Eki(t)μEki(t)+pqSki(t),dIki(t)dt=(1α)βkiSki(t)Θ(t)+εEki(t)γ2Iki(t)μIki(t),dRki(t)dt=p(1q)Ski(t)+γ1Eki(t)+γ2Iki(t)μRki(t), (4.1)

    with the initial conditions

    Ski(0)>0,Eki(0)0,Iki(0)0,Rki(0)0

    and Ski(0)+Eki(0)+Iki(0)+Rki(0)=1. Define Θ(t)=1kni=1kip(ki)Iki(t), which denotes the probability that any given node is connected to a node in the propagable state. p(ki) denotes the connectivity distribution of a node with the degree ki and k=ikip(ki) denotes the finite average degree of all nodes in the network, which address the effects of contact heterogeneity in information spreading.

    For System (4.1), it is claimed that the solution with the above initial condition is nonnegative for all t0. Actually, if there exists t0>0 such that Ski(t0)=0 and Ski(t)>0 for t(0,t0), then Ski(t0)=Λ>0, which implies that Ski(t)0 for all t0. By the expression of Θ(t), it is known that if there is ki satisfying Iki(0)>0, then Θ(0)>0. By the continuity of Θ(t), there exists t1 such that Θ(t)>0 for any t(0,t1). Then, for Eki(t), we have

    Eki(t)Eki(0)e(ε+γ1+μ)t0

    for t(0,t2). If there exists t2>t1 such that Θ(t2)=0 and Θ(t)>0 for t(0,t2), then

    dΘ(t)dt|t=t2=1kni=1kip(ki)[(1α)βkiSki(t)Θ(t)+εEki(γ2+μ)Iki(t)]|t=t20,

    leading to a contradiction. Thus, Θ(t)>0 is valid for all t>0.

    In the following, we will discuss the model in two cases of q=0 and q0, respectively, as in Section 3.2. The dynamic analysis is conducted theoretically and the effect of a hot search on the time evolution of state variables is demonstrated numerically.

    If the hot search effect is not considered (q=0), then System (4.1) is reduced as

    {dSki(t)dt=ΛβkiSki(t)Θ(t)μSki(t)pSki(t),dEki(t)dt=αβkiSki(t)Θ(t)εEki(t)γ1Eki(t)μEki(t),dIki(t)dt=(1α)βkiSki(t)Θ(t)+εEki(t)γ2Iki(t)μIki(t),dRki(t)dt=pSki(t)+γ1Eki(t)+γ2Iki(t)μRki(t). (4.2)

    By the next generation matrix approach, similar to R10, the basic propagation number for System (4.2) is defined as

    R20=εαβΛ+(1α)βΛ(ε+γ1+μ)(ε+γ1+μ)(γ2+μ)(μ+p)k2k,

    in which the term k2=ik2ip(ki) presents the connectivity fluctuations.

    It is easy to obtain that System (4.2) always has the information-free equilibrium

    G0={(S0k1,E0k1,I0k1,R0k1),,(S0kn,E0kn,I0kn,R0kn)},

    where S0ki=Λμ+p,E0ki=0,I0ki=0,R0ki=pΛμ(μ+p) for i=1,2,,n. If R20>1, there exists an information-propagating equilibrium

    G1={(Sk1,Ek1,Ik1,Rk1)(Skn,Ekn,Ikn,Rkn)},

    where

    Ski=ΛβkiΘ+μ+p,Eki=αβkiΛΘ(βkiΘ+μ+p)(ε+γ1+μ),Iki=(1α)βkiΛΘ(βkiΘ+μ+p)(γ2+μ)+εαβkiΛΘ(βkiΘ+μ+p)(ε+γ1+μ)(γ2+μ),Rki=pSki+γ1Eki+γ2Ikiμ

    for i=1,2,,n, with Θ=1kni=1kip(ki)Iki(t). In fact, we have

    Θ=1kni=1kip(ki)[(1α)βkiγ2+μ+εαβki(ε+γ1+μ)(γ2+μ)]ΛΘβkiΘ+μ+p;

    then, denote

    F1(Θ)=11kni=1kip(ki)[(1α)βkiγ2+μ+εαβki(ε+γ1+μ)(γ2+μ)]ΛβkiΘ+μ+p.

    It follows that

    F1(1)=11kni=1kip(ki)Λ[(ε+γ1+μ)α(γ1+μ)](ε+γ1+μ)(γ2+μ)βkiβkiΘ+μ+p>0,F1(0)=11kni=1kip(ki)[1αγ2+μ+εα(ε+γ1+μ)(γ2+μ)]Λβkiμ+p=1R20,

    and

    F1(Θ)=1kni=1kip(ki)[(1α)βkiγ2+μ+εαβki(ε+γ1+μ)(γ2+μ)]Λβki(βkiΘ+μ+p)2>0.

    When the basic propagation number R20>1, from above, it can be seen that there is only one positive Θ such that F1(Θ)=0. Consequently, there exists a unique information-propagating equilibrium G1.

    In the following, we discuss the stability of the equilibria by the Hurwitz criteria; it is necessary to utilize the Jacobi matrix at the equilibria, which is a 3n×3n order matrix when the recovery state is not considered.

    Theorem 4.1. If R20<1, then the information-free equilibrium G0 of System (4.2) is globally asymptotically stable, and it is unstable when R20>1.

    Proof. We will first establish the local stability of G0 by using the Hurwitz criterion and then the global stability by using the Lyapunov function.

    Let A11=(μ+p), A22=(ε+γ1+μ), A33=(γ2+μ) and Bij=βkiS0kikkjp(kj); then, we consider the following matrix ˜J0=(mij)3n×3n at the component

    (S0k1,E0k1,I0k1,,S0kn,E0kn,I0kn)

    of G0 for ki=1,2,,n:

    ˜J0=(A11000B11B1n0000B21B2n0A1100Bn1Bnn00A220αB11αB1n0000αB21αB2n000A22αBn1αBnn00ε0A33(1α)B11(1α)B1n0000(1α)B21(1α)B2n000ε(1α)Bn1A33(1α)Bnn).

    Then the characteristic equation of ˜J0 has the following form:

    g(λ):=(λ+μ+p)nP2n(λ)=0,

    with P2n(λ) being a characteristic polynomial of degree 2n that corresponds to the following matrix ˜J1=(˜mij)2n×2n:

    ˜J1=(A2200αB11αB1n0A220αB21αB2n00A22αBn1αBnnε00A33(1α)B11(1α)B1n0ε0(1α)B21(1α)B2n00ε(1α)Bn1A33(1α)Bnn).

    Noting that S0k1=S0k2==S0kn, and then making a similar transformation on ˜J1 leads to the following upper triangular matrix ˜J2, which has the same eigenvalues as ˜J1:

    ˜J2=(˜m1100αB11αB12αB1n0A22000000A22000000A33000000A33000000A33),

    with ˜m11=A22(1α)(ε+γ1+μ)+αεA33ni=1Bii. Therefore, the characteristic equation of ˜J2 has the eigenvalue A22 with multiplicity n1, the eigenvalue A33 with multiplicity n and a single eigenvalue ˜m11. In order to show the sign of ˜m11, it can be further expressed as

    ˜m11=(ε+γ1+μ)+βS0kiγ2+μk2k((1α)(ε+γ1+μ)+αε)=(ε+γ1+μ)(R201).

    If R20<1, then all eigenvalues of ˜J0 are negative, which ensures the local stability of G0. Otherwise, if R20>1, there exists an eigenvalue with a positive real part, which implies that G0 is unstable.

    Then, we consider a Lyapunov function of the following form:

    V(t)=ni=1[εEki(t)+(ε+γ1+μ)Iki(t)].

    Calculating the derivative of V(t) along System (4.2) gives

    dV(t)dt=ni=1ε[(αβSki(t)Θ(t)(ε+γ1+μ)Eki(t))+(ε+γ1+μ)((1α)βkiSki(t)Θ(t)+εEki(t)(γ2+μ)Iki(t))](εα+(ε+γ1+μ)(1α))βΛμ+pk2kni=1Iki(ε+γ1+μ)(γ2+μ)ni=1Iki(γ2+μ)(ε+γ1+μ)(R201)ni=1Iki.

    If R20<1, then we can see that dV(t)dt0, and it can be verified that the largest invariant set where dV(t)dt=0 is the singleton {G0}. By the LaSalle invariance principle, the theorem is proved.

    Remark. In Model (4.1), the parameter p is the presence rate of the hot search. If there is no such source emerging in the process of information spreading, i.e., p=0, then by the method in [27,28], the information-propagating equilibrium G1 is globally asymptotically stable when R20>1.

    Similarly, suppose α>q in the case of q0. For System (4.1), due to the hot search effect, there exists no information-free equilibrium. However, there exists the information-propagating equilibrium

    G2={(Sk1,Ek1,Ik1,Rk1)(Skn,Ekn,Ikn,Rkn)},

    where

    Ski=ΛβkiΘ+μ+p,Eki=Λ(αβkiΘ+pq)(ε+γ1+μ)(βkiΘ+μ+p),Iki=(1α)ΛβkiΘ(γ2+μ)(βkiΘ+μ+p)+εΛ(αβkiΘ+pq)(ε+γ1+μ)(γ2+μ)(βkiΘ+μ+p),Rki=p(1q)Ski+γ1Eki+γ2Ikiμ,

    for i=1,2,,n, and the unique Θ satisfying

    Θ=1kni=1kip(ki)[(1α)ΛβkiΘ(γ2+μ)(βkiΘ+μ+p)+εΛ(αβkiΘ+pq)(ε+γ1+μ)(γ2+μ)(βkiΘ+μ+p)].

    In fact, let

    F2(Θ)=Θ1kni=1kip(ki)[(1α)ΛβkiΘ(γ2+μ)(βkiΘ+μ+p)+εΛ(αβkiΘ+pq)(ε+γ1+μ)(γ2+μ)(βkiΘ+μ+p)];

    then, the positive equilibrium Θ is given by F2(Θ)=0. We have that

    limΘ0+F2(Θ)=εΛpq(ε+γ1+μ)(γ2+μ)(μ+p)<0,limΘ1F2(Θ)=11kni=1kip(ki)εΛ(βki+pq)+βkiΛ(1α)(γ1+μ+ε)(ε+γ1+μ)(γ2+μ)(βki+μ+p)>0.

    Now, we examine the derivative of F2(Θ); then,

    F2(Θ)=1+1γ2+μ1kni=1kip(ki)[(1α)βkiΛ(μ+p)(βkiΘ+μ+p)2+εΛε+γ1+μαβki(μ+p)βkipq(βkiΘ+μ+p)2]>1,

    implying that the function F2(Θ) is strictly increasing. Therefore, there is only one positive Θ such that F2(Θ)=0, and the existence and uniqueness of the positive equilibrium G2 follows. We have the following result.

    Theorem 4.2. For System (4.1), there exists a unique information-propagating equilibrium G2.

    The numerical investigations in Figures 4 and 5, performed on the four networks (regular network, Watts-Strogatz (WS) network, Erd¨os-Rˊenyi (ER) network and BA network) with size N=1000, confirm the theoretical analysis about the influence of q on the evolution of state density in Model (4.1). In the figures, with the other parameters taken as the same values, the state variables are depicted when q is fixed to 0 and 0.65, respectively, From Figure 4, it can be seen that the curves of the propagable density I(t) is always lower than that of the rejected density R(t) when q=0. However, in the case of q=0.65, the curves of I(t) are obviously higher than those of R(t) in the initial time steps, and then they decline gradually to be lower than the density of R(t), as shown in Figure 5. In addition, it can be observed that the density of I(t) in Figure 4 is lower than that in Figure 5 at the same time steps, although it is not particularly obvious. Therefore, the comparison results in the fact that the hot search effect is not ignorable in the progress of information propagation.

    Figure 4.  Plots of state density with q=0 in four networks.
    Figure 5.  Plots of state density with q=0.65 in four networks.

    In this section, we describe the numerical simulations of Models (3.1) and (4.1) to examine the sensitivity analysis, which demonstrate the effects of some parameters on the basic propagation number, as well as the density of propagable nodes, allowing us to improve the understanding of information spreading.

    For convenience, in the following simulations, let K be the average degree ˉk in the homogeneous network, as well as k2k in the heterogeneous network. First, we consider the SEIR model on the regular network and the ER network in the absence of the hot search effect (q=0). In this scenario, there exists a basic propagation number R0 (denoting R10 and R20). To explore the sensitivity of parameter variations on R0 due to the uncertainty in estimating the input parameter, the partial rank correlation coefficients (PRCCs) was calculated based on the Latin hypercube sampling scheme.

    For our performance, the sample size was taken as n=2000 and seven parameters were chosen as the input variables. The values of the PRCCs were calculated for each of the seven parameters, as shown in Figure 6, which can be used to observe the contribution made by each input parameter to the variability of R0. The positive or negative value corresponding to each parameter reveals the correlation that is generally evaluated as being important, moderate or not significantly different from zero. In Table 2, the specific values of the PRCCs and p-values are presented, noting that p-values are regarded as zero when smaller than 0.0001. If we consider that absolute values of PRCCs that are greater than 0.4 indicate an important correlation between the input parameters and R0, the values in (0.2,0.4) are moderate correlations and the values in (0,0.2) are not significantly different from zero, it can be seen that K (PRCC = 0.6898) has an important impact on R0, and β, γ2 and p have a moderate impact, with the PRCC values being 0.3757, -0.3300 and -0.3716 respectively. Compared with them, the remaining parameters affect R0 slightly.

    Figure 6.  Sensitivity analysis of PRCCs for input parameters and R0 (outcome variable).
    Table 2.  PRCC values for input parameters and R0.
    Input parameters The basic propagation number R0
    PRCC p-value
    β 0.3757 0
    α -0.1253 0
    ε 0.0615 0.0059
    K 0.6898 0
    γ1 -0.1087 0
    γ2 -0.3300 0
    p -0.3716 0

     | Show Table
    DownLoad: CSV

    We focus on the sensitive parameters to examine the variation in R0 due to the change of the concerned parameters. In Figure 7, the contour plots on the regular network and ER network are presented in terms of parameter pairs (β, p) and (β, γ2) respectively. For comparison, the degree in the regular network is taken as k=4 and k4 in the ER network, so that k2k7.03 when k228.1327. The effects of β and p are shown in the first row in Figure 7. It can be seen that the basic propagation number R0 is less than 1 only when β is approximately in the interval of [0,0.4] in the regular network, compared with the interval of [0,0.2] in the ER network. In particular, if β=0.2 and p=0.6, then R0 is less than 1 in the regular network; however, it is greater than 1 in the ER network. In addition, for a fixed p and varying β, the basic propagation number R0 experiences more variation in its value in the ER network than in the regular network. For the pair (β, γ2), the contour plot in the second row presents the numerical result similar to that for the parameter pair (β, p).

    Figure 7.  Contour plots of the basic propagation number on the regular network (Left) and ER network (Right), revealing the estimated effects of the parameter pairs (β, p) and (β, γ2).

    If the hot search effect is taken into account, it is previously referred that there exists no information-free equilibrium in this case and R0 is only a critical parameter related to the stability of the information-propagating equilibrium. To further analyze the impact of different connectivity K on the information propagation process, we consider the model on four networks with size N=1000 and connectivity ranging from 4 to 16. The profile for the propagable population is depicted in Figure 8. From the curves, we can see that, starting on nodes with different connectivity K, the higher the connectivity, the higher the peak of the density I(t), resulting in a higher number of propagable individuals in information spreading. On the contrary, the classes of nodes with few connections have a relatively small density at its peak and a slight shift of the peak time, which suggests that a lower prevalence in information propagation can be achieved by managing the connected nodes to some extent.

    Figure 8.  Effect of the parameter K on the density of I(t) on four networks.

    For OSNs, the hot search list can serve as a source of hot information that is more easily accessible and attracts more attention. Once becoming interested, the internet users have the potential to retweet and spread the information actively. In order to provide further insight into the hot search effect, the density of propagable nodes as a function of time was investigated for the four types of networks in Figure 9. When the hot search effect is absent, i.e., q=0, it is shown that there is a very small variation in the density of I(t). With the increase of q from 0 to 0.6, there is a much higher number of propagable individuals in each type of the network, which provides striking evidence of the peculiar influence of the hot search on the process of information propagation.

    Figure 9.  Hot search effect on the density of I(t) on four networks.

    To validate the proposed SEIR model, real data from Sina Weibo (the largest online social media in China) were collected through Python software. Then, the model was employed to fit the data, with the aim of minimizing the difference between model simulation and the collected data. In the case study, two datasets, named Information A and Information B, were used to illustrate the capability of our model to simulate the dynamical behaviors of information propagation.

    Information A is a promotional tweet from an actress for her starring TV series. Table 3 lists the number of retweets in the first 13 hours after the microblog was posted. From the initial time, the data were recorded in half-hour time intervals; then, a total of 27 data points was obtained. Information B is a tweet about the performance of Chinese women players in a World Table Tennis Championship. The collecting time lasted 15 hours, and the data of the forwarding after information release is listed in Table 4, in which the data were also collected every half an hour, the same as with Information A. In these two tables, the first data point denotes the publisher's release, counted as 1. In data collecting, when the retweet of the concerned information decreased to 1 and experienced no fluctuations in the next hour, it was regarded as the end time, and the last data point was 1.

    Table 3.  Data for Information A.
    Time 1 2 3 4 5 6 7 8 9 10 11 12 13 14
    Number 1 76 97 144 89 93 55 29 22 12 14 17 16 6
    Time 15 16 17 18 19 20 21 22 23 24 25 26 27
    Number 4 5 4 4 3 1 1 1 4 3 2 1 1

     | Show Table
    DownLoad: CSV
    Table 4.  Data for Information B.
    Time 1 2 3 4 5 6 7 8 9 10 11 12
    Number 1 332 660 704 352 251 213 127 69 66 40 30
    Time 13 14 15 16 17 18 19 20 21 22 23 24
    Number 25 26 30 30 17 20 19 13 13 18 9 16
    Time 25 26 27 28 29 30 31
    Number 20 18 12 18 7 1 1

     | Show Table
    DownLoad: CSV

    In our numerical simulations, we use the BA network to fit the collected data for Information A and Information B, according to the experimental results in [4], revealing the phenomenon that some real networks, including social networks, have a power-law exponent that also exists in microblog networks. In fitting the data, we determine the values of parameters by using the least squares method to minimize their root mean square error (RMSE) and mean absolute error (MAE) for better accuracy; then, the model was applied on the BA network with an average degree of 12. The simulation results and the collected data are plotted in Figures 10 and 11. The experimental illustrations show that the whole process for the two cases can be divided into three stages. The retweet of the Weibo information increased dramatically at the initial stage, implying that most users interested in it were involved in forwarding the information shortly after its release. Then, the popularity of it decayed and the forwarders of the microblog decreased in the following second stage. At the last stage, the number of retweets approached a very low level eventually.

    Figure 10.  Data fitting for Information A.
    Figure 11.  Data fitting for Information B.

    The parameters and initial values used in Figures 10 and 11 are listed in Table 5. In these two cases, we set E(0)=1, I(0)=1 and R(0)=1. From the comparison of the model simulation and the data of Information A, it can be observed from Figure 10 that the proposed model captured the trend of the data accurately, although there exists some error in the fitting. In Figure 11, there is an obvious deviation of the simulation results from the real data of forwarding Information B in the decay period, and the possible reason may be that the sharp decline in the data occurred at midnight when there are relatively few users active on Weibo, which is not considered in the simulation. The values of the RMSE and MAE are given in Table 6.

    Table 5.  List of parameters and variables values used in case study.
    Parameter & variable Λ β α ε μ γ1 γ2 p q S(0)
    Information A 4 0.89 0.68 0.78 0.004 0.12 0.22 0.65 0.66 277
    Information B 4 0.85 0.68 0.76 0.004 0.20 0.22 0.58 0.65 1997

     | Show Table
    DownLoad: CSV
    Table 6.  RMSE and MAE.
    Case RMSE MAE
    Information A 7.29 4.85
    Information B 92.53 59.68

     | Show Table
    DownLoad: CSV

    Regarding the case study, it is worth noting that the best-fit parameter q came out to be almost the same, i.e., q=0.66 for Information A and q=0.65 for Information B. The possible reason is that these two kinds of data had similar structural characteristics and three stages, as mentioned before. Particularly, from Figures 10 and 11, it can be seen that a single wave of forwarding occurred in the propagating process. Although the fitting results were not accurate enough to match the collected data, they may give us some hint that the hot search exerts a similar effect on information propagation with these types of characteristics.

    In this paper, an extended SEIR model was investigated to capture the dynamic property of information propagation in complex networks. In particular, the effect of a hot search in OSNs was incorporated into our model, with aim of demonstrating its performance in information diffusion. Based on the proposed model framework, we conducted dynamical analysis of the model in homogeneous networks and heterogeneous networks, including the stability of the information-free equilibrium and the information-propagating equilibrium. In addition, numerical simulations were carried out to validate the theoretical results and analyze the sensitivity of the concerned parameters. Furthermore, we focused on two types of specific information spreading in Weibo and performed data fitting on the BA network, which confirmed the validity of our model in application. The presented results allow us to have a better knowledge of the process of information propagation in OSNs.

    In fact, real-world networks in which information is disseminated are more complex and dynamic; it is reasonable to expect that the proposed model can characterize the behaviors of information diffusion more accurately, and consequently have better performance in practical realization. In this perspective, the incorporation of many elements of realism, such as a time delay in forwarding, followers' awareness and competition of different online information, is important in modeling to acquire a better understanding of information propagation. Other than the case with one single wave of forwarding, numerical examples with more characteristics should be considered to improve the level of agreement between the model and real data, which will be left for us to explore in future.

    This research was supported by the National Natural Science Foundation of China (#12071268) and Natural Science Basic Research Plan in Shaanxi Province of China (#2022JM-038, #2022JM-042).

    The authors declare that there is no conflict of interest regarding the publication of this paper.



    [1] A. DeLegge, H. Wangler, Is this the end for Facebook? A mathematical analysis, Appl. Math. Comput., 305 (2017), 364–380. https://doi.org/10.1016/j.amc.2017.02.014 doi: 10.1016/j.amc.2017.02.014
    [2] E. Stai, E. Milaiou, V. Karyotis, S. Papavassilion, Temporal dynamics of information diffusion in twitter: Modeling and experimentation, IEEE Trans. Comput. Soc. Syst., 5 (2018), 256–264. https://doi.org/10.1109/TCSS.2017.2784184 doi: 10.1109/TCSS.2017.2784184
    [3] Q. Gao, F. Abel, G. J. Houben, Y. Yu, A comparative study of users' microblogging behavior on Sina Weibo and Twitter, in User Modeling, Adaptation, and Personalization, Springer, Berlin Heidelberg, 7379 (2012), 88–101. https://doi.org/10.1007/978-3-642-31454-4-8
    [4] Y. Li, M. Qian, D. Jin, P. Hui, A. Vasilakos, Revealing the effificiency of information diffusion in online social networks of microblog, Inf. Sci., 293 (2015), 383–389. https://doi.org/10.1016/j.ins.2014.09.019 doi: 10.1016/j.ins.2014.09.019
    [5] C. Lei, Z. Lin, H. Wang, The free boundary problem describing information diffusion in online social networks, J. Differ. Equations, 254 (2013), 1326–1341. https://doi.org/10.1016/j.jde.2012.10.021 doi: 10.1016/j.jde.2012.10.021
    [6] F. Nian, Y. Shi, J. Cao, Modeling information propagation in high-order networks based on explicit-implicit relationship, J. Comput. Sci., 55 (2021), 101438. https://doi.org/10.1016/j.jocs.2021.101438 doi: 10.1016/j.jocs.2021.101438
    [7] D. Zhao, J. Sun, Y. Tan, J. Wu, Y. Dou, An extended SEIR model considering homepage effect for the information propagation of online social networks, Phys. A., 512 (2018), 1019–1031. https://doi.org/10.1016/j.physa.2018.08.006 doi: 10.1016/j.physa.2018.08.006
    [8] M. Fu, J. Feng, D. Lande, O. Dmytrenko, D. Manko, R. Prakapovich, Dynamic model with super spreaders and lurker users for preferential information propagation analysis, Phys. A., 561 (2021), 125266. https://doi.org/10.1016/j.physa.2020.125266 doi: 10.1016/j.physa.2020.125266
    [9] S. Abdullah, X. Wu, An epidemic model for news spreading on twitter, in 2011 IEEE 23rd International Conferenc on Tools with Artificial Intelligence, (2011), 163–169. https://doi.org/10.1109/ICTAI.2011.33
    [10] F. Yin, X. Zhu, X. Shao, X. Xia, Y. Pan, J. Wu, Modeling and quantifying the influence of opinion involving opinion leaders on delayed information propagation dynamics, Appl. Math. Lett., 121 (2021), 107356. https://doi.org/10.1016/J.AML.2021.107356 doi: 10.1016/J.AML.2021.107356
    [11] G. Guan, Z. Guo, Bifurcation and stability of a delayed SIS epidemic model with saturated incidence and treatment rates in heterogeneous networks, Appl. Math. Model., 101 (2022), 55–75. https://doi.org/10.1016/J.APM.2021.08.024 doi: 10.1016/J.APM.2021.08.024
    [12] L. Huo, L. Wang, N. Song, C. Ma, B. He, Rumor spreading model considering the activity of spreaders in the homogeneous network, Phys. A., 468 (2016), 855–865. https://doi.org/10.1016/j.physa.2016.11.039 doi: 10.1016/j.physa.2016.11.039
    [13] R. Jie, J. Qiao, G. Xu, Y. Meng, A study on the interaction between two rumors in homogeneous complex networks under symmetric conditions, Phys. A., 454 (2016), 129–142. https://doi.org/10.1016/j.physa.2016.02.048 doi: 10.1016/j.physa.2016.02.048
    [14] A. Jain, J. Dhar, V. Gupta, Rumor model on homogeneous social network incorporating delay in expert intervention and government action, Commun. Nonlinear Sci. Numer. Simul., 84 (2020), 105189. https://doi.org/10.1016/j.cnsns.2020.105189 doi: 10.1016/j.cnsns.2020.105189
    [15] Y. Zan, DSIR double-rumors spreading model in complex networks, Chaos Solitons Fractals, 110 (2018), 191–202. https://doi.org/10.1016/j.chaos.2018.03.021 doi: 10.1016/j.chaos.2018.03.021
    [16] Q. Liu, T. Li, M. Sun, The analysis of an SEIR rumor propagation model on heterogeneous network, Phys. A., 469 (2017), 372–380. https://doi.org/10.1016/j.physa.2016.11.067 doi: 10.1016/j.physa.2016.11.067
    [17] C. Wan, T. Li, Z. Sun, Global stability of a SEIR rumor spreading model with demographics on scale-free networks, Adv. Differ. Equations, 2017 (2017), 253. https://doi.org/10.1186/s13662-017-1315-y doi: 10.1186/s13662-017-1315-y
    [18] S. Yu, Z. Yu, H. Jiang, J. Li, Dynamical study and event-triggered impulsive control of rumor propagation model on heterogeneous social network incorporating delay, Chaos Solitons Fractals, 145 (2021), 110806. https://doi.org/10.1016/J.CHAOS.2021.110806 doi: 10.1016/J.CHAOS.2021.110806
    [19] S. Chen, H. Jiang, L. Li, J. Li, Dynamical behaviors and optimal control of rumor propagation model with saturation incidence on heterogeneous networks, Chaos Solitons Fractals, 140 (2020), 110206. https://doi.org/10.1016/j.chaos.2020.110206 doi: 10.1016/j.chaos.2020.110206
    [20] L. Zhu, W. Liu, Z. Zhang, Delay differential equations modeling of rumor propagation in both homogeneous and heterogeneous networks with a forced silence function, Appl. Math. Comput., 370 (2020), 124925. https://doi.org/10.1016/j.amc.2019.124925 doi: 10.1016/j.amc.2019.124925
    [21] L. Zhu, G. Guan, Dynamical analysis of a rumor spreading model with self-discrimination and time delay in complex networks, Phys. A., 533 (2019), 121953. https://doi.org/10.1016/j.physa.2019.121953 doi: 10.1016/j.physa.2019.121953
    [22] F. Yin, X. Shao, B. Tang, X. Xia, J. Wu, Modeling and analyzing cross-transmission dynamics of related information co-propagation, Sci. Rep., 11 (2021), 1–20. https://doi.org/10.1038/S41598-020-79503-8 doi: 10.1038/S41598-020-79503-8
    [23] T. Li, Y. Xiao, Linking the disease transmission to information dissemination dynamics: An insight from a multi-scals model study, J. Theor. Biol., 526 (2021), 110796. https://doi.org/10.1016/j.jtbi.2021.110796 doi: 10.1016/j.jtbi.2021.110796
    [24] P. van den Driessche, J. Watmough, Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission, Math. Biosci., 180 (2002), 29–48. https://doi.org/10.1016/S0025-5564(02)00108-6 doi: 10.1016/S0025-5564(02)00108-6
    [25] R. C. Robinson, An Introduction to Dynamical Systems: Continuous and Discrete, 2nd edition, New York, 2012.
    [26] J. Lasalle, The stability of dynamical systems, Philadelphia, SIAM, 1976.
    [27] G. Liu, Z. Liu, Z. Jin, Dynamics analysis of epidemic and information spreading in overlay networks, J. Theor. Biol., 444 (2018), 28–37. https://doi.org/10.1016/j.jtbi.2018.02.010 doi: 10.1016/j.jtbi.2018.02.010
    [28] X. Zhao, Z. Jing, Global asymptotic behavior in some cooperative systems of functional differential equations, Canad. Appl. Math. Quart., 4 (1996), 421–444.
  • This article has been cited by:

    1. Jinlong Fu, Yan Song, Yike Feng, Rumor Spreading Model Considering the Roles of Online Social Networks and Information Overload, 2023, 11, 2169-3536, 123947, 10.1109/ACCESS.2023.3328396
    2. Jianlin Jia, Yuwen Huang, Wanting Zhang, Yanyan Chen, Information propagation characteristic by individual hesitant-common trend on weighted network, 2024, 12, 2296-424X, 10.3389/fphy.2024.1410089
    3. Guocui Liu, Fuzhong Nian, Yang Liu, Information consistency and inhibition impact on diffusion in networks, 2024, 0129-1831, 10.1142/S0129183124502413
    4. F. Mortaji, A. Abta, H. Laarabi, M. Rachik, Global stability analysis, stabilization and synchronization of an SEIR epidemic model, 2024, 0020-7179, 1, 10.1080/00207179.2024.2391923
    5. Hongmiao Zhu, Zhen Jin, A dynamics model of knowledge dissemination in a WeChat Group from perspective of duplex networks, 2023, 454, 00963003, 128083, 10.1016/j.amc.2023.128083
    6. Ahmed H. Msmali, Fazal Dayan, Muhammad Rafiq, Nauman Ahmed, Abdullah Ali H. Ahmadini, Hassan A. Hamali, A Nonstandard Computational Investigation of SEIR Model with Fuzzy Transmission, Recovery and Death Rates, 2023, 77, 1546-2226, 2251, 10.32604/cmc.2023.040266
  • Reader Comments
  • © 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2485) PDF downloads(140) Cited by(6)

Figures and Tables

Figures(11)  /  Tables(6)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog