Social network public opinion rapid perception and efficient tracing method and device

Abstract

The application provides a social network public opinion fast sensing and efficient tracing method and device. The method comprises the following steps: step one: obtaining a social network topology structure G, G=(V, E), wherein V represents a social network user node set, and E represents a social network user relationship set; step two: determining an optimal target observation user node set S according to the network topology structure G; and step three: determining whether an opinion event occurs and an initial propagation user node causing the opinion event according to information obtained by the target observation user node set S. Through the method provided by the application, the fast sensing and efficient tracing of opinion information can be realized under a small observation node set S.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method and apparatus for rapid perception and efficient tracing of public opinion on social networks. Background Technology

[0002] Social networks are the primary medium for the dissemination of public opinion in modern society. However, public opinion may become distorted during its dissemination; therefore, it is necessary to perceive and trace the source of public opinion.

[0003] However, current information tracing methods are not efficient in information perception and tracing. Summary of the Invention

[0004] This invention provides a method and apparatus for rapid perception and efficient source tracing of public opinion on social networks, in order to solve the problem that the perception and location costs are relatively high in the process of public opinion perception and source tracing in the prior art.

[0005] A method for rapid perception and efficient source tracing of public opinion on social networks includes:

[0006] Step 1: Obtain the social network topology G, G = (V, E), where V represents the set of social network user nodes and E represents the set of social network user relationships;

[0007] Step 2: Based on the network topology G, determine the optimal set of target observation user nodes S;

[0008] Step 3: Based on the information obtained from the target observation user node set S, determine whether the public opinion event has occurred and the initial user node that caused the spread of the public opinion event.

[0009] Furthermore, in the aforementioned method for rapid perception and efficient tracing of public opinion on social networks, step two includes:

[0010] Determine the perception time feature value and the positioning cost feature value for each node in the set of user nodes in the social network;

[0011] Based on the perception time feature value and the positioning cost feature value, the target observation user node set S is determined by the network structure-based device NTM.

[0012] Furthermore, in the aforementioned method for rapid perception and efficient tracing of public opinion on social networks, step two includes:

[0013] Based on the network topology G, determine its corresponding dynamic propagation data set ζ; the dynamic propagation data set ζ is composed of: each dynamic propagation is the time when any node v in the social network user node set V is infected, where the first node to be infected is the propagation source node u.s ;

[0014] The perception time cost T for each node in the network is determined based on the dynamic propagation data set ζ. s And positioning cost C s ;

[0015] Based on the perceived time cost T s And positioning cost C s The target observation user node set S is determined based on a data-driven greedy device GS, along with a preset time threshold δ and a propagation source location cost ε, and a preset time threshold δ of the target observation user node set S.

[0016] Furthermore, in the aforementioned method for rapid perception and efficient source tracing of public opinion on social networks, the step of determining the target observation user node set S based on the perception time feature value and the location cost feature value, using a network structure-based device NTM, includes:

[0017] Step 1: Determine the perception time feature value and the positioning cost feature value corresponding to each node in the social network user node set V, let i = n; i is a natural number, and n is the total number of nodes in the social network user node set V;

[0018] Step 2: Based on the perception time feature value and the positioning cost feature value, determine the comprehensive feature value corresponding to each node, and select the node with the smallest comprehensive feature value from all the comprehensive feature values ​​as the i-th candidate observation point S(i), let i = i-1;

[0019] Step 3: Determine the perception time feature value and positioning cost feature value for each of the remaining nodes; the remaining nodes are the nodes in the social network user node set V excluding the candidate observation points;

[0020] Step 4: From the remaining nodes, select the node with the smallest comprehensive feature value as the i-th candidate observation point S(i), and let i = i-1;

[0021] Step 5: Repeat steps 3-4 to traverse all nodes in the social network user node set V, and finally obtain the first candidate target observation user node set;

[0022] Step 6: Select a preset number of observation points from the first set of candidate target observation user nodes in ascending order of their serial numbers i as the target observation user nodes.

[0023] Furthermore, in the aforementioned method for rapid perception and efficient tracing of public opinion on social networks, the dynamic propagation data set ζ is either a simulated propagation data set obtained by inputting the network topology G into the SIR model, or a real propagation data set obtained based on the network topology G.

[0024] Furthermore, in the aforementioned method for rapid perception and efficient source tracing of public opinion on social networks, the method based on the perception time cost T... s And positioning cost C s The target observation user node set S is determined based on a data-driven greedy device GS, along with a preset time threshold δ and a propagation source location cost ε, and includes:

[0025] Step 1: In T s >δ and C s In the case of >ε, the perceived time cost T for each node in the social network user node set V is determined based on the dynamic propagation data set ζ. s And positioning cost C s ;

[0026] Step 2: Based on the perceived time cost T s And positioning cost C s Determine the comprehensive cost corresponding to each node in the social network user node set V, and select the node s with the lowest comprehensive cost from all nodes and add it to the target observation user node set S;

[0027] Step 3: Calculate the perception time cost T for each of the remaining nodes. s And positioning cost C s And from the remaining nodes, select the node with the lowest overall cost and add it to the target observation user node set S;

[0028] Step 4: Repeat step 3 until the target set of observed user nodes S satisfies T. s <δ or C s <ε;

[0029] Step 5: If the set of target observation user nodes obtained in Step 4 satisfies T s If the value is greater than δ, then calculate the perception time cost T for each of the remaining nodes. s Then select the node with the lowest perceived time cost and add it to the target observation user node set S, repeating until T. s <δ; otherwise, if the set of target observation user nodes obtained in step 4 satisfies C s If the value is greater than ε, then calculate the positioning cost C for each of the remaining nodes. sThen select the node with the lowest positioning cost and add it to the target observation user node set S, repeating until C. s <ε;

[0030] Step 6: Finally, the target observation user node set S is obtained.

[0031] Furthermore, in the aforementioned method for rapid perception and efficient source tracing of public opinion on social networks, the method based on the perception time cost T... s And positioning cost C s The target observation user node set S is determined based on a data-driven greedy device GS, along with a preset time threshold δ and a propagation source location cost ε, and includes:

[0032] Step 1: In T s In the case of >δ, the sensing time cost T for each node is determined based on the dynamic propagation data set ζ. s ;

[0033] Step 2: From all perceived time costs T s In the process, the node s with the lowest perception time cost is selected as the observation point;

[0034] Step 3: Calculate the perception time cost T for each of the remaining nodes. s And from the remaining nodes, select the node with the lowest perception time cost as the observation point;

[0035] Step 4: Repeat step 3 until T. s <δ, merge the observation points obtained in step 2 with the observation points obtained in step 3 to obtain the target observation user node set S1;

[0036] Step 5: Based on the target observation user node set S1 and node set V, determine the first node set V'; V' = V - S1, let i = n;

[0037] Step 6: Determine the location cost feature value α(v) corresponding to each node in the first node set V', and select the node with the smallest location cost feature value from all the location cost feature values ​​α(v) as the i-th candidate observation point, and update i = i-1 at the same time;

[0038] Step 7: For the remaining nodes in the first node set V', determine the location cost feature value α(v) corresponding to each remaining node, and select the node with the smallest location cost feature value as the i-th candidate observation point S(i), where i = i-1;

[0039] Step 8: Repeat step 7 to traverse all nodes in the first node set V', and finally obtain a second alternative target observation user node set;

[0040] Step 9: Select nodes from the second candidate target observation user node set according to the sequence number i in ascending order, until a node corresponding to C in the second candidate target observation user node set is selected. s If ε < ε, then the selected node will be used as the target observation user node set S2;

[0041] Step 10: Merge the target observation user node set S1 and the target observation user node set S2 to obtain the target observation user node set S.

[0042] Furthermore, in the aforementioned method for rapid perception and efficient source tracing of public opinion on social networks, the method based on the perception time cost T... s And positioning cost C s The target observation user node set S is determined based on a data-driven greedy device GS, along with a preset time threshold δ and a propagation source location cost ε, and includes:

[0043] Step 1: Determine the location cost feature value corresponding to each node in the set of user nodes V of the social network, and let i = n;

[0044] Step 2: Select the node with the smallest positioning cost feature value from all positioning cost feature values ​​as the i-th candidate observation point S(i), and let i = i-1;

[0045] Step 3: Determine the location cost feature value for each of the remaining nodes, and select the node with the smallest location cost feature value as the candidate observation point S(i) and i = i-1;

[0046] Step 4: Repeat step 3 to traverse all nodes in the social network user node set V, and finally obtain a third alternative target observation user node set;

[0047] Step 5: Select nodes from the third candidate target observation user node set in ascending order of their serial numbers (i) and add them to the target observation user node set S, until the positioning cost C of set S is determined based on the dynamic propagation data set ζ. s <ε;

[0048] Step 6: If the set S is determined based on the dynamically propagated data set ζ, then T s If the value is greater than δ, then calculate the perceived time cost T of each remaining node in the social network user node set V after removing the nodes from the current set S. s ;

[0049] Step 7: Select the perceived time cost T s The node corresponding to the minimum value is added to the target observation user node set S;

[0050] Step 8: Repeat steps 6-7 until the target set of observed user nodes S satisfies T. s <δ, thus obtaining the target observation user node set S.

[0051] A device for rapid perception and efficient tracing of public opinion on social networks includes:

[0052] The acquisition unit is used to acquire the social network topology G, G = (V, E), where V represents the set of user nodes in the social network and E represents the set of user relationships in the social network.

[0053] The determining unit is used to determine the optimal set of target observation user nodes S based on the network topology G;

[0054] The tracing unit is used to determine whether a public opinion event has occurred and the initial user node that caused the spread of the public opinion event based on the information obtained from the target observation user node set S.

[0055] The method provided by this invention optimizes the arrangement of observation user nodes on social networks, enabling rapid perception and efficient tracing of public opinion information with a relatively small set of observation nodes S. This allows for early detection and effective control of the further spread of negative public opinion, reducing its economic losses and stability threats to society. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the public opinion dissemination rapid perception and efficient source tracing device of the present invention;

[0058] Figure 2 This is a schematic diagram illustrating the network decomposition strategy used in this invention;

[0059] Figure 3The figure shows a comparison of the proportion of target observation user node sets with the ES method under the same perception time and positioning cost conditions obtained by four observation point deployment devices in the RCV36 network group. In the figure, (a), (b), (c), and (d) represent the comparison of the proportion of target observation user node sets with the same perception time and positioning cost conditions obtained by the GS, NTM, HM-I, HM-II devices and the ES strategy under different configurations, respectively.

[0060] Figure 4 This diagram illustrates the expected distribution of the target observed user node set proportions on three networks, obtained by NTM devices and other methods regarding the upper limit of the positioning cost threshold. In the diagram, (ad), (eh), and (il) represent the upper limit of the positioning cost threshold obtained using different methods in the AirTrans, Power, and Gowalla networks, respectively. The diagram illustrates the expected result of the proportion of the target observed user node set. Figures (a,e,i), (b,f,j), (c,g,k), and (d,h,l) represent the perception time thresholds, respectively. The diagram illustrates the expected proportion of the target observed user node set with different upper limits of positioning cost thresholds of 0.2, 0.4, 0.6, and 0.8. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0062] The goal of rapid perception of public opinion dissemination is to find an optimal set of observation points to detect dissemination of public opinion in a timely manner by monitoring the state of nodes in the target observation user node set. Current methods primarily construct the target observation user node set by greedily searching for the node with the shortest detection time on the available dissemination dataset. However, the performance of the greedy approach is heavily dependent on the quantity and quality of the collected dissemination data, which significantly limits its practical application. Methods developed based on network structure, such as degree centrality and acquaintance strategies, generally perform worse than the greedy approach when a dissemination dataset is available.

[0063] Similar to the problem of rapid perception of public opinion dissemination, the problem of effective source tracing can also be viewed as optimizing a set of observed user nodes, but with the goal of minimizing the cost of searching for the source of the dissemination. Traditional methods first assume that the states of some or all nodes are known, aiming to design an efficient estimator to assess the probability of each node being a source. Then, in this case, there are typically two strategies to locate the true source of the dissemination. The first strategy centers on the node with the highest likelihood value and gradually expands the search area by considering the node's first-level neighbors, second-level neighbors, third-level neighbors, etc., until the true source of the dissemination is found. Within this range, the effectiveness of each method is evaluated using error distance; if the corresponding inferred source is closer to the true source in terms of error distance, the estimator is considered better. This optimization objective is generally suitable for homogeneous networks, but unstable for heterogeneous networks because short error distances may be associated with a large number of nodes. As an alternative, the second strategy first provides nodes ranked according to the estimator's probability, and then checks each node sequentially until the true source is found. In this scenario, all nodes with a likelihood greater than 0 could be propagation sources. Therefore, the fewer the number of associations among nodes with a likelihood greater than 0—that is, the optimization objective is to minimize the number of nodes with a likelihood greater than 0—the better the estimator will be. The cost of locating the true propagation source is difficult to estimate and measure for both strategies. Currently, Liu et al. have proposed a candidate set framework that includes the true propagation source. By optimizing the user node set, they reduce the size of the candidate set, thereby minimizing the cost of searching for the propagation source.

[0064] To date, there has been a great deal of research on the source location of public opinion dissemination, including rapid perception and efficient source tracing. However, there is a lack of methods that can optimize and more efficiently trace the source while considering rapid perception and early monitoring of public opinion, so as to achieve a more systematic and comprehensive solution to the problem of public opinion dissemination.

[0065] To overcome the shortcomings of existing social network public opinion control technologies, this invention provides a device for rapid perception and efficient source tracing of public opinion dissemination, the specific implementation of which is as follows:

[0066] First, input the social network topology information and the public opinion propagation dataset; then, calculate the data-driven perception time cost and location cost, as well as the perception time feature value and location feature value based on the network structure, for user nodes that are not selected as observation points. Next, iteratively select nodes that meet the conditions according to different devices and add them to the target observation user node set. Repeat the above iterative operation until the threshold conditions of perception time and location are met. Then, the target observation user node set is the target set of different devices.

[0067] This invention can facilitate rapid perception and efficient source tracing of public opinion dissemination on social networks by optimizing the arrangement of user nodes for observation. This allows for early detection and effective control of the further spread of negative public opinion, reducing its economic losses and stability threats to society.

[0068] like Figure 1 As shown, this invention provides a device for rapid perception and efficient source tracing of public opinion propagation on social networks. It mainly solves the problem of monitoring and controlling public opinion propagation through four methods: a data-driven greedy scheme (GS), a network topology-based method (NTM), and a hybrid method (HM). Specifically, it includes the following steps: Step 1: Input the potential social network topology G = (V, E), where V represents the set of social network user nodes, containing n = |V| nodes, and E represents the set of social network user relationships, containing m = |E| edges; Input the dynamic propagation data set ζ of network G based on the propagation model SIR (Susceptible-Infected-Recovered), where each propagation data in the set contains the propagation source node u. s Let , and the time t(v) at which any node v is infected. Under the SIR propagation model, nodes in the network have three states: susceptible (S), infected (I), and recovered (R). In the initial propagation phase, all nodes in the network except the propagation source node u... s The node is in the infected state (I), while all other nodes are in the infected state (S).

[0069] Step 2: Let the set of target observation user nodes be S, initialized to empty; let the monitoring time of observation point u be τ(u), and the upper limit of node monitoring time be τ. o If the infection time of a node t(u) > τ o Then τ(u)=τ o Otherwise, τ(u) = t(u); let the perception time threshold be δ and the propagation source inference cost threshold be ε.

[0070] Step 3: For each node v in the node set V that does not belong to S, i.e., node v∈V\S, calculate the monitoring time cost T of set U=S∪{v} after it is added to set S, based on the propagation dataset ζ. s (U,ζ),

[0071]

[0072] where ξ(u s ) indicates that node u s Dissemination data as a source of transmission.

[0073] Step 4: For each node v in the node set V\S, calculate the positioning cost C of the set U = S ∪ {v} on the propagation data set s (U, ζ),

[0074]

[0075] where V c represents the candidate set of propagation sources generated by the network decomposition method from the observation points in the propagation data, where the propagation source u s must belong to V c , |V c | represents the size of the candidate set, that is, the number of nodes included.

[0076] Specifically, the specific implementation schemes of the network decomposition strategy and the candidate set construction method in Step 4 are as follows: Take the node set U as the target observation user node set V o , remove the edges connected to the nodes in the set V o in the graph g and the set V o , and obtain the subgraph G'(V', E') constructed by different connected components, where V' = V\V o , E' = (V' × V') ∩ E, then the connected coverage α(u) of the observation point u is

[0077] α(u) = ∪ v∈Γ'(u) c i (v) (3)

[0078] where, Γ'(u) represents the neighbors of the observation point u in the subgraph G', and c i (v) represents the connected component where its neighbor node is located in the subgraph G'. Then the candidate set can be defined as

[0079]

[0080] where represents the set of target observation user nodes that are first detected to be infected. The candidate set constructed according to this method must contain the propagation source u s .

[0081] As Figure 2 shown, if V o = {u, v}, then the original network is decomposed into a subgraph G' = {c i , i = [1, 5]} composed of five connected components, and the examples of the composition of its connected coverage and candidate set are: In the public opinion propagation, for example, when t(u) < t(v), V' o = {u}, the connected coverage is α(u) = {c1, c2, c3}, and the candidate set is v c = α(u) ∪ {u}; when t(u) = t(v), V'o = {u, v}, meaning the intersection of the connected cover of nodes is the candidate set, V c =c1∪c3∪{u,v}.

[0082] Step 5: For each node v in the node set V\S, calculate the perceptual time feature value and localization cost feature value of the set U = S∪{v} based only on the network structure. The localization cost feature value is the connectivity cover α(v) of node v obtained through network decomposition when the set U is the target set of observed user nodes. This value can effectively suppress the cost function C. s The larger α(v) is, the higher the positioning cost; the perception time feature value is the key threshold κ(α(v)) of the degree of the connected coverage α(v).

[0083] Specifically, research shows that the key threshold κ = for network structure <k 2 > / <k>It plays an important role in determining whether propagation has occurred. Therefore, κ(α(v)) is used to represent the critical threshold of the connected component α(v) to minimize the perception time. The larger κ(α(v)) is, the earlier the perception time.

[0084] Same as step 5, V o =U, then

[0085]

[0086] Where k' u This indicates that node u is in network G'(V o The degree in \{v}), network G'(V) o \{v}) refers to the node set V o {v} is used as the observation point to obtain a subgraph through network decomposition in step 5.

[0087] Step 6: Based on the calculation results of Steps 3-5, the data-driven greedy device GS, the network structure-based device NTM, and the two hybrid devices HM-I and HM-II select different nodes under different conditions to construct the target observation user node set S, and repeat Steps 3-5 until the termination condition of the corresponding method is met.

[0088] Specifically, the specific implementation schemes for the four different devices in step 6 are as follows:

[0089] 1) Data-driven greedy device GS: when T is satisfied s >δ and C s When the condition is greater than ε, select node s = argmin v∈V\S [T s (U,ζ)×C s [U,ζ] is added to the target observation user node set S, where T s and C s Both refer to the perception time cost and positioning cost of S, with the set U = S∪{v}. Otherwise, when T s When the value is greater than δ, the node to be added to the target observation user node set S is s = argmin. v∈V\S T s (U,ζ); when C s When ε > 0, select node s = argmin v∈V\S C S (U,ζ) is added to the target observation user node set S.

[0090] (2) Network-based device NTM: Initialize i = n, V' = V, then set node S(i) = argmin v∈V' α(v)×[κ(α(v))] -φ , V'=V'\{S(i)}, i=i-1.

[0091] (3) HM-I device based on hybrid strategy: when T s When >δ, select node s = argmin v∈V\S' T s (U,ζ) are added to the target observation user node set S'. Otherwise, let i = n, V' = V'\S', when i > |S'|, S'(i) = argmin v∈V' α(v), V'=V'\{S'(i)}, i=i-1. Then, let S=S', i=|S'|+1, when C s >ε, S=S∪S'(i), i=i+1.

[0092] (4) HM-II device based on hybrid strategy: Initialize i = n, V' = V, when i > 0, S'(i) = argmin v∈V' α(v), V'=V'\{S'(i)}, i=i-1. Then let i=1, when C s >ε, S=S∪S'(i), i=i+1; when T s When >δ, S=S∪argmin v∈V\S T s (U,ζ), the set U=S∪{v}.

[0093] Step 7: Calculate the effect of different strategies on T s >δ and C s The proportion of the target observed user node set under the ε condition This indicator is used as an indicator for evaluating the device, where This represents the average value obtained by repeating the same device multiple times in the same test network.

[0094] Table 1 Experimental Network Dataset

[0095] RCV36 36 10-20 21-66 AirTrans 1 3146 18164 Power 1 4941 6594 Gowalla 1 196591 950327

[0096] To verify the effectiveness of this invention, an experiment was conducted on a real social network, and the experimental network data is shown in Table 1. The relevant experimental parameters were set as follows: (1) The dynamic propagation dataset was generated on a real network by the SIR propagation model, and the node infection probability and recovery probability were random values ​​in the interval (0,1]; (2) Each propagation dataset contained 1000 valid datasets, where valid refers to the propagation range accounting for 5% of the total number of nodes, and the remaining τ o =10; (3) The parameter φ in the device is 1.

[0097] The comparative methods used in the experiment included: Exhaustive Search (ES), which uses an exhaustive approach to find observation points in the network that enable fast perception and efficient source tracing; High Degree (HD), which selects nodes with higher degree in the network as observation points; and Collective Influence (CI), Min-sum and Reverse-greedy (MSRG), and Finding key players in Networks through Deep Reinforcement Learning (FINDER) methods from relevant research fields.

[0098] Figure 3 This paper presents a comparative distribution of the target observation user node set ratios obtained by four different observation point layout devices in an RCV36 network group, compared with the ES method, under the condition of the same sensing time and positioning cost. The figures include 16 configurations with δ values ​​of 0.1, 0.2, 0.3, and 0.4, and ε values ​​of 0.1, 0.2, 0.3, and 0.4, representing the target observation user node set ratio distribution in 36 networks. Each figure contains 576 scattered points. The average ratio of the target observation user node set ratios of different methods to the ES method is marked in the lower right corner of each figure.

[0099] Depend on Figure 3 As can be seen, compared with ES (exhaustive search), the solutions obtained by the four observation point placement devices proposed in this invention have an average size 1.06 times that of ES, indicating that this invention can effectively handle the problem of rapid perception and efficient source tracing of public opinion. In particular, Figures (a) and (b) show that the NTM device outperforms GS, indicating that the correlation-based optimization device based on network structure is better than the greedy device. Furthermore, the hybrid device HM-II has the best performance among the four proposed devices.

[0100] Figure 4 This diagram illustrates the expected distribution of the target observer user node set proportions on three networks, obtained by NTM devices and other methods, regarding the upper limit of the positioning cost threshold. The horizontal axis represents the upper limit of the positioning cost threshold. That is, random selection Similarly, testing different configurations... This represents the upper limit of the perceived time cost threshold, randomly selected. Different test sets are formed; the vertical axis represents the expected proportion of the target observed user node set. Where A t Indicates the number of random occurrences.

[0101] Depend on Figure 4 It can be seen that, under different perception time cost thresholds ε and positioning cost thresholds δ, a comparison of the expected proportion R(·) of the target observation user node set obtained by the NTM device and other methods reveals that the proposed network-based NTM device can clearly obtain the optimal target observation user node set with the smallest R(·) in all cases. This indicates that the present invention has significant advantages in solving the problem of observation point deployment for rapid perception and efficient source tracing of public opinion, and its performance is relatively stable in different configurations, demonstrating robustness and broad application prospects.

[0102] Based on network propagation data and network structure, this invention develops a data-driven greedy device, a network structure-based device, and a hybrid device. By optimizing the arrangement of observation points in the network, this invention can simultaneously solve the problems of early perception and efficient location of public opinion propagation in social networks, achieving low-cost and high-efficiency public opinion propagation monitoring and source location, thereby realizing effective control of the public opinion propagation process.

[0103] In summary, the device of this invention achieves rapid perception and efficient source tracing of online public opinion. The proposed data-driven greedy device, network-structure-based device, and hybrid device can all achieve effective public opinion monitoring and control. In particular, the data-driven greedy device suffers from high complexity in achieving efficient source tracing; therefore, optimization based on network structure can effectively improve efficiency, with the second hybrid device exhibiting the best performance. Thus, this invention combines rapid perception and efficient source tracing of public opinion, demonstrating effectiveness, efficiency, and robustness through data-driven approaches and network structure optimization, making it suitable for public opinion control on social networks.

[0104] The method for rapid perception and efficient source tracing of public opinion on social networks provided by this invention includes the following steps:

[0105] Step 1: Obtain the social network topology G, G = (V, E), where V represents the set of social network user nodes and E represents the set of social network user relationships;

[0106] Step 2: Based on the network topology G, determine the optimal set of target observation user nodes S;

[0107] Step 3: Based on the information obtained from the target observation user node set S, determine whether the public opinion event has occurred and the initial user node that caused the spread of the public opinion event.

[0108] The method provided by this invention enables rapid perception and efficient tracing of public opinion information with a relatively small set of observation nodes S.

[0109] Specifically, the propagation network topology is G = (V, E), where V represents the set of user nodes in the social network, containing n = |V| nodes, and E represents the set of user relationships in the social network, containing m = |E| edges. This invention determines the target set of observed user nodes S using two methods: one is to input the network topology G into the propagation model SIR (Susceptible-Infected-Recovered), simulate the dynamic propagation data set ζ of the network topology G based on this model, and then determine the target set of observed user nodes S through the dynamic propagation data set ζ; the other is to directly determine the target set of observed user nodes S from the network topology G.

[0110] The method for directly determining the set of target observation user nodes S from the network topology G is as follows:

[0111] Determine the perception time feature value k(α(v)) and the location cost feature value α(v) for each node in the set of user nodes of the social network.

[0112] Based on the perception time feature value κ(α(v)) and the positioning cost feature value α(v), the target observation user node set S is determined by the network structure-based device NTM.

[0113] The method for determining the target set of user observation nodes S by dynamically propagating the data set ζ is as follows:

[0114] Based on the network topology G, determine its corresponding dynamic propagation data set ζ; the dynamic propagation data set ζ is composed of: each dynamic propagation is the time t(v) at which any node v in the social network user node set V is infected, where the first node to be infected is the propagation source node u. s ;

[0115] The perception time cost T for each node in the network is determined based on the dynamic propagation data set ζ. s And positioning cost C s ;

[0116] Based on the perceived time cost T s And positioning cost C s The target observation user node set S is determined based on a data-driven greedy device GS, along with a preset time threshold δ and a propagation source location cost ε, and a preset time threshold δ of the target observation user node set S.

[0117] The following section provides a detailed explanation of how to determine the target set of user nodes S, including the following four methods.

[0118] The first method is:

[0119] Step 1: Determine the perception time feature value κ(α(v)) and the positioning cost feature value α(v) for each node in the social network user node set V, and let i = n; i is a natural number, and n is the total number of nodes in the social network user node set V;

[0120] Step 2: Based on the perception time feature value κ(α(v)) and the positioning cost feature value α(v), determine the comprehensive feature value α(v)×[κ(α(v))] for each node. -φ From all the comprehensive eigenvalues, select the node corresponding to the smallest comprehensive eigenvalue as the i-th candidate observation point S(i), that is, S(i) = argmin v∈V' α(v)×[κ(α(v))] -φ ;

[0121] Step 3: Determine the perception time feature value κ(α(v)) and the localization cost feature value α(v) for each of the remaining nodes; the remaining nodes are the nodes in the social network user node set V excluding the candidate observation points;

[0122] Step 4: From the remaining nodes, select the node with the smallest comprehensive feature value as the i-th candidate observation point S(i), and let i = i-1;

[0123] Step 5: Repeat steps 3-4 to traverse all nodes in the social network user node set V, and finally obtain the candidate target observation user node set;

[0124] Step 6: Select a preset number of observation points from the set of candidate target observation user nodes in ascending order of their serial numbers i as the target observation user nodes.

[0125] The second method is:

[0126] Step 1: In T s >δ and C s In the case of >ε, the perceived time cost T for each node in the social network user node set V is determined based on the dynamic propagation data set ζ. s And positioning cost C s ;

[0127] Step 2: Based on the perceived time cost T s And positioning cost C s Determine the comprehensive cost corresponding to each node in the set of user nodes V of the social network, and select the node s with the lowest comprehensive cost from all nodes, and add it to the target observation user node set S, i.e., s = argmin v∈V\S [T s (U,ζ)×C s (U,ζ)];

[0128] Step 3: Calculate the perception time cost T for each of the remaining nodes. s And positioning cost C s And from the remaining nodes, select the node with the lowest overall cost as the observation point;

[0129] Step 4: Repeat step 3 until the target set of observed user nodes S satisfies T. s <δ or C s <ε;

[0130] Step 5: If the set of target observation user nodes obtained in Step 4 satisfies T s If the value is greater than δ, then calculate the perception time cost T for each of the remaining nodes. s Then select the node with the lowest perceived time cost and add it to the target observation user node set S, repeating until T. s <δ; otherwise, if the set of target observation user nodes obtained in step 4 satisfies C s If the value is greater than ε, then calculate the positioning cost C for each of the remaining nodes. s Then select the node with the lowest positioning cost and add it to the target observation user node set S, repeating until C. s <ε;

[0131] Step 6: Finally, the target observation user node set S is obtained.

[0132] The third method is:

[0133] Step 1: In T s In the case of >δ, the sensing time cost T for each node is determined based on the dynamic propagation data set ζ. s ;

[0134] Step 2: From all perceived time costs T s In the process, the node s with the lowest perception time cost is selected as the observation point;

[0135] Step 3: Calculate the perception time cost T for each of the remaining nodes. s And from the remaining nodes, select the node with the lowest perception time cost as the observation point;

[0136] Step 4: Repeat step 3 until T. s <δ, merge the observation points obtained in step 2 with the observation points obtained in step 3 to obtain the target observation user node set S1;

[0137] Step 5: Based on the target observation user node set S1 and node set V, determine the first node set V'; V' = V - S1, let i = n;

[0138] Step 6: Determine the location cost feature value α(v) corresponding to each node in the first node set V', and select the node with the smallest location cost feature value from all the location cost feature values ​​α(v) as the i-th candidate observation point, and update i = i-1 at the same time;

[0139] Step 7: For the remaining nodes in the first node set V', determine the location cost feature value α(v) corresponding to each remaining node, and select the node with the smallest location cost feature value as the i-th candidate observation point S(i), where i = i-1;

[0140] Step 8: Repeat step 7 to traverse all nodes in the first node set V', and finally obtain a second alternative target observation user node set;

[0141] Step 9: Select nodes from the second candidate target observation user node set according to the sequence number i in ascending order, until a node corresponding to C in the second candidate target observation user node set is selected. s If ε < ε, then the selected node will be used as the target observation user node set S2;

[0142] Step 10: Merge the target observation user node set S1 and the target observation user node set S2 to obtain the target observation user node set S.

[0143] The fourth method is:

[0144] Step 1: Determine the location cost feature value α(v) for each node in the set of user nodes V of the social network, and let i = n;

[0145] Step 2: Select the node with the smallest positioning cost feature value from all positioning cost feature values ​​as the i-th candidate observation point S(i), and let i = i-1;

[0146] Step 3: Determine the location cost feature value α(v) for each of the remaining nodes, and select the node with the smallest location cost feature value as the candidate observation point S(i) and i = i-1;

[0147] Step 4: Repeat step 3 to traverse all nodes in the social network user node set V, and finally obtain a third alternative target observation user node set;

[0148] Step 5: Select nodes from the third candidate target observation user node set in ascending order of their serial numbers (i) and add them to the target observation user node set S, until the positioning cost C of set S is determined based on the dynamic propagation data set ζ. s <ε;

[0149] Step 6: If the observation point set S is determined to satisfy T based on the dynamic propagation data set ζ s If the value is greater than δ, then calculate the perceived time cost T of each remaining node in the social network user node set V after removing the nodes from the current set S. s ;

[0150] Step 7: Select the perceived time cost T s The node corresponding to the minimum value is added to the target observation user node set S4;

[0151] Step 8: Repeat steps 6-7 until the target set of observed user nodes S satisfies T. s <δ, thus obtaining the target observation user node set S.

[0152] This invention also provides a device for rapid perception and efficient tracing of public opinion on social networks, comprising:

[0153] The acquisition unit is used to acquire the social network topology G, G = (V, E), where V represents the set of user nodes in the social network and E represents the set of user relationships in the social network.

[0154] The determining unit is used to determine the optimal set of target observation user nodes S based on the network topology G;

[0155] The tracing unit is used to determine whether a public opinion event has occurred and the initial user node that caused the spread of the public opinion event based on the information obtained from the target observation user node set S.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.< / k>

Claims

1. A method for rapid perception and efficient source tracing of public opinion on social networks, characterized in that, include: Step 1: Obtain the social network topology , ,in Represents a set of user nodes in a social network. Represents a set of user relationships on a social network; Step 2: Based on the network topology Determine the optimal set of target user nodes. ; Step 3: Observe the set of user nodes according to the target. The information obtained is used to determine whether a public opinion event has occurred and the user nodes that initially led to its spread. Step two includes: According to the network topology Determine its corresponding dynamic propagation data set The dynamic propagation data set The composition is as follows: each dynamic propagation consists of a set of user nodes in the social network. any node The time of infection, with the first node infected being the source node of transmission. ; According to the dynamic propagation data set Determine the sensing time cost for each node in the network. and positioning costs ; Based on the perceived time cost and positioning costs and the set of target user nodes. preset time threshold cost threshold for inferring the source of transmission The target set of user observation nodes is determined based on a data-driven greedy device GS. ; Wherein, the perceived time cost and positioning costs The calculation steps are as follows: Let the set of target observed user nodes be Initialize to empty; set the observation point to... Monitoring time is The maximum node monitoring time is If the node's infection time ,but ,otherwise Let the perception time threshold be... The cost threshold for inferring the source of propagation is ; For node sets China does not belong to Each node in , i.e., node According to the propagation dataset Calculate its addition to the set Post-set Perceived time cost , (1) in Indicates that by node Data on the spread of information as a source of transmission; For node sets Each node in Calculate its addition to the set Post-set Location cost on propagation dataset , (2) in This represents the candidate set of propagation sources inferred from the observation points in the propagation data using the network decomposition method. This indicates the size of the candidate set, i.e., the number of nodes it contains; n = |V| nodes; Among them, the data-driven greedy device GS is based on the perceived time cost. and positioning costs Determine the set of user nodes in the social network. The comprehensive cost is calculated for each node, and the node with the lowest comprehensive cost is selected from all nodes. Add to the target user node set .

2. The method for rapid perception and efficient source tracing of public opinion on social networks according to claim 1, characterized in that, The dynamic propagation data set To make the network topology Input the SIR model to obtain the simulated propagation data set; or according to the network topology. The obtained set of real propagation data.

3. The method for rapid perception and efficient source tracing of public opinion on social networks according to claim 1, characterized in that, The based on the perceived time cost and positioning costs And the preset time threshold of the target observation user node set S. cost threshold for inferring the source of transmission The target set of user observation nodes is determined based on a data-driven greedy device GS. include: Step 1: In and In the case of the aforementioned dynamic propagation data set Determine the set of user nodes in the social network The perception time cost for each node in the process and positioning costs ; Step 2: Based on the perceived time cost and positioning costs Determine the set of user nodes in the social network. The comprehensive cost is calculated for each node, and the node with the lowest comprehensive cost is selected from all nodes. Add to the target user node set ; Step 3: Calculate the perception time cost for each of the remaining nodes. and positioning costs Then, from the remaining nodes, select the node with the lowest overall cost and add it to the target observation user node set. ; Step 4: Repeat step 3 until the target set of observed user nodes is reached. satisfy ; Step 5: If the target observation user node set obtained in Step 4 satisfies Then calculate the perception time cost for each of the remaining nodes. The node with the lowest perceived time cost is then selected and added to the target observation user node set. Repeat until Otherwise, if the target observation user node set obtained in step 4 satisfies Then calculate the positioning cost for each of the remaining nodes. The node with the lowest positioning cost is then selected and added to the target observation user node set. Repeat until ; Step 6: Finally, obtain the target set of observed user nodes. .

4. The method for rapid perception and efficient source tracing of public opinion on social networks according to claim 1, characterized in that, The based on the perceived time cost and positioning costs and the set of target user nodes. preset time threshold cost threshold for inferring the source of transmission The target set of user observation nodes is determined based on a data-driven greedy device GS. include: Step 1: In In the case of the aforementioned dynamic propagation data set Determine the perception time cost for each node. ; Step 2: From all perceived time costs In the process, select the node with the lowest perceived time cost. As an observation point; Step 3: Calculate the perception time cost for each of the remaining nodes. And from the remaining nodes, select the node with the lowest perception time cost as the observation point; Step 4: Repeat step 3 until... The observation points obtained in step 2 are merged with those obtained in step 3 to obtain the target observation user node set. ; Step 5: Observe the set of user nodes according to the target. Given a set of nodes V, determine the first set of nodes. ; ,make ; Step 6: Determine the first set of nodes The location cost feature value corresponding to each node in the process and from all positioning cost features In the process, the node corresponding to the minimum positioning cost feature value is selected as the first node. One alternative observation point, updated simultaneously. Location cost characteristic value is a set When observing a set of user nodes as a target, the network decomposition method is used to obtain information about the nodes. connectivity coverage size; Step 7: For the first set of nodes Among the remaining nodes, determine the positioning cost feature value corresponding to each remaining node. And select the node corresponding to the minimum location cost feature value as the first node. One alternative observation point ,and ; Step 8: Repeat step 7 to traverse the first set of nodes. All nodes in the dataset are used to obtain a second set of candidate target observation user nodes. Step 9: Observe the user nodes from the second candidate target set according to their sequence numbers. Select nodes from smallest to largest until the second candidate target is observed, and observe a node in the user node set. Then the selected node will be used as the target observation user node set. ; Step 10: Set the target observation user nodes and the target observation user node set By merging, the target set of observed user nodes is obtained. .

5. The method for rapid perception and efficient source tracing of public opinion on social networks according to claim 1, characterized in that, The based on the perceived time cost and positioning costs and the set of target user nodes. preset time threshold cost threshold for inferring the source of transmission The target set of user observation nodes is determined based on a data-driven greedy device GS. include: Step 1: Determine the set of user nodes in the social network The location cost feature value corresponding to each node in the process ,make Location cost characteristic value is a set When observing a set of user nodes as a target, the network decomposition method is used to obtain information about the nodes. connectivity coverage size; Step 2: Select the node with the smallest positioning cost feature value from all positioning cost feature values, and use it as the first node. One alternative observation point ,make ; Step 3: Determine the positioning cost feature value for each of the remaining nodes, and select the node with the smallest positioning cost feature value as a candidate observation point. and ; Step 4: Repeat step 3 to traverse the set of user nodes in the social network. All nodes in the dataset are used to ultimately obtain a third alternative target observation user node set; Step 5: From the third candidate target observation user node set, based on the sequence number and Select nodes in ascending order and add them to the target observation user node set. Until according to the dynamically propagated data set Determine set Positioning costs ; Step 6: If based on the aforementioned dynamic propagation data set Determine set satisfy Then calculate the set of user nodes in the social network. Remove the current set The perception time cost of each remaining node in the middle node ; Step 7: Select perceived time cost The node corresponding to the minimum value is added to the target observation user node set. ; Step 8: Repeat steps 6-7 until the target set of observed user nodes is reached. satisfy The target observation user node set is obtained. .

6. A device for rapid perception and efficient tracing of social network public opinion based on the method for rapid perception and efficient tracing of social network public opinion as described in claim 1, characterized in that, include: Acquisition unit, used to acquire social network topology. , ,in Represents a set of user nodes in a social network. Represents the set of user relationships in a social network; Determining unit, configured to determine based on the network topology Determine the optimal set of target user nodes. ; The tracing unit is used to observe the set of user nodes based on the target. The information obtained determines whether a public opinion event has occurred and the user nodes that initially spread the event.

Images