Method and device for controlling false information on social networks

Abstract

The present invention discloses a method and device for controlling false information on a social network, and relates to the technical field of network information analysis. Large networks can be decomposed quickly and efficiently, thereby controlling the spread of false information in a timely manner and reducing the losses caused by it. The method comprises: determining the first sequence with the smallest network resilience among multiple first sequences as the optimal sequence; deleting multiple third nodes with the highest probability from the coarse-grained network and adding them to a second set of removed nodes; placing the newly occupied nodes at the beginning of the removed sequence according to the sequence number, placing the unoccupied nodes included in the unoccupied node set at the middle of the removed sequence according to the node sequence number, and placing the removed nodes at the end of the removed sequence to form a removed sequence, and connecting the optimal sequence and the removed sequence to obtain a final sequence.

Description

Technical Field

[0001] The present invention relates to the technical field of network information analysis, and more specifically to a method and device for controlling false information on social networks. Background Art

[0002] In a social network model, nodes represent accounts, and edges represent interactions between accounts, such as following, forwarding, and commenting. A misinformation campaign often begins when a node posts a piece of misinformation. This information then spreads to connected nodes through edges until it reaches the entire network. To minimize the negative impact, robust measures are needed to control the spread of misinformation, such as blocking accounts and deleting misinformation. These measures, reflected in the social network model, are essentially node removal. Once a node is removed, it loses its connections with other nodes and can no longer interact with them. Therefore, the mechanisms for dissemination no longer apply to it, effectively curbing its spread.

[0003] However, removing nodes will incur certain costs. For example, blocking an account may prevent normal information from being disseminated, affecting the connectivity of the social network. The network decomposition problem aims to control the spread of false information while minimizing the cost. The solution to this problem is to identify a set of nodes in the network. When all the nodes in the set are removed, the structure and function of the network can be destroyed to the greatest extent, thus controlling the spread of false information. Existing methods are prone to poor control of the spread of false information due to the need to remove too many nodes. Furthermore, existing methods cannot balance the computing time and computing effect, and often require a long time of calculation to obtain an approximate optimal solution. In the face of sudden false information dissemination events, if decisions and measures cannot be taken in a short period of time, the best control opportunity is likely to be lost. Summary of the Invention

[0004] Embodiments of the present invention provide a method and apparatus for controlling false information on social networks, which can quickly and efficiently decompose large networks, thereby timely controlling the spread of false information and reducing the losses caused by it.

[0005] An embodiment of the present invention provides a method for controlling false information on a social network, including:

[0006] Constructing an initial propagation network, determining an initial sequence and an initial exchange threshold corresponding to the initial propagation network; traversing each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node; when the connectivity component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the first node in the initial sequence to obtain a first sequence; determining the first sequence with the smallest network resilience among the multiple first sequences as the optimal sequence;

[0007] The initial nodes included in the initial propagation network are divided into a first occupied node set and a first removed node set according to the optimal sequence; the connected components included in the initial propagation network are defined as new nodes, the multiple new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; a coarse-grained network is formed according to the edges between the first occupied node in the first occupied node set and the first removed node in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set;

[0008] The leaf nodes and the central node in the unoccupied node set included in the coarse-grained network are removed, and the central node is added to the second removed node set; the third node included in the coarse-grained network is set to an occupied state, and when the unoccupied node set is a non-empty set, iterate according to the initial removal information of the third node, the initial root information of the third node, and the belief propagation. When the belief propagation reaches a self-consistent steady state, the removal probability of each third node is determined according to the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, and the multiple third nodes with the highest probability are deleted from the coarse-grained network and added to the second removed node set;

[0009] The newly occupied nodes included in the newly occupied set are placed at the beginning of the removal sequence according to the sequence number, the leaf nodes are placed in the unoccupied node set, and the unoccupied nodes included in the unoccupied node set are placed in the middle of the removal sequence according to the node sequence number, and the removed nodes included in the second removed node set are placed at the end of the removal sequence to form a removal sequence; the optimal sequence and the removal sequence are connected to obtain a final sequence.

[0010] Preferably, the traversing each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node, and when the connected component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the position of the first node in the initial sequence to obtain the first sequence, further comprising:

[0011] Traversing each first node in the first segment included in the initial sequence, and if a second node within the candidate range corresponding to the first node is determined to be a candidate node based on a judgment probability, adding the second node to the candidate set to form a candidate node;

[0012] Determine the connected components of each candidate node included in the candidate set and the smallest candidate node with the smallest connected component, and occupy the smallest candidate node;

[0013] When the connected component of the smallest candidate node is less than the initial exchange threshold, swapping the positions of the smallest candidate node and the first node in the initial sequence, and updating the initial exchange threshold;

[0014] After traversing each first node included in the first segment, a first sequence is obtained, and the network resilience of the first sequence is determined.

[0015] Preferably, the first segment is a random segment in the initial sequence, and the index of the starting element of the random segment is determined by the following formula:

[0016]

[0017] The index of the end element of the random segment is determined by the following formula:

[0018] γ e =γ s +y2mod(φ2n-10)+10

[0019] Among them, γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, y2 is a uniformly distributed index in [1, 10 9 ] is a random integer, φ1 and φ2 are two adjustable parameters, q c，0 Represents the initial critical threshold corresponding to the initial sequence.

[0020] Preferably, the first segment is a two-stage segment in the initial sequence, and the index of the start element of the two-stage segment is determined by the following formula:

[0021] γ s =q c，T n

[0022] The index of the ending element of the two-phase segment is determined by the following formula:

[0023]

[0024] Among them, q c，T represents the critical threshold corresponding to the T-th round time series, q c，T =min{q|L(q)≥θ T}; γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, θ T is the maximum connected component ratio threshold in round T, φ3, φ4>1, φ5 are adjustable parameters, y2 is a uniform distribution in [1, 10 9 ], y3 is a random number uniformly distributed in [0, 1).

[0025] Preferably, dividing the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the optimal sequence specifically includes:

[0026] When the initial propagation network forms a maximum connected component, determining the proportion of occupied nodes of the optimal sequence in the initial propagation network as a critical threshold, and dividing the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the critical threshold;

[0027] The first occupied node set is determined by the following formula:

[0028]

[0029] The node set is determined by the following formula:

[0030]

[0031] Among them, V o represents the first occupied node set, V u represents the first set of removed nodes, express The corresponding critical threshold, represents the optimal sequence.

[0032] Preferably, removing the leaf nodes and the central node from the unoccupied node set included in the coarse-grained network and adding the central node to the second removed node set specifically includes:

[0033] Determine the first ψ third nodes with the highest degree centrality among the third nodes included in the coarse-grained network as central nodes; or

[0034] Determine the new node formed by ψ second nodes at the end of the optimal sequence in the coarse-grained network as a central node, where ψ is a positive integer greater than zero;

[0035] The second removed node set is an empty set before the central node is added.

[0036] An embodiment of the present invention further provides a device for controlling false information on a social network, comprising:

[0037] a determination unit configured to construct an initial propagation network, determine an initial sequence and an initial exchange threshold corresponding to the initial propagation network; traverse each first node included in the initial sequence, obtain and occupy a minimum candidate node corresponding to each first node, and when a connected component of the minimum candidate node is less than the initial exchange threshold, replace the first node in the initial sequence with the minimum candidate node to obtain a first sequence; and determine the first sequence with the minimum network resilience among the multiple first sequences as the optimal sequence;

[0038] A composition unit is configured to divide initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to an optimal sequence; define connected components included in the initial propagation network as new nodes, wherein the plurality of new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; and form a coarse-grained network based on edges between a first occupied node in the first occupied node set and a first removed node in the first removed node set, edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set;

[0039] a removal unit, configured to remove leaf nodes and central nodes from an unoccupied node set included in the coarse-grained network, and add the central node to a second removed node set; set a third node included in the coarse-grained network to an occupied state, and when the unoccupied node set is a non-empty set, iterate according to initial removal information of the third node, initial root information of the third node, and belief propagation; when the belief propagation reaches a self-consistent steady state, determine the removal probability of each third node according to the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, delete multiple third nodes with the highest probability from the coarse-grained network, and add them to the second removed node set;

[0040] The obtaining unit is configured to place the newly occupied nodes included in the newly occupied set at the beginning of the removal sequence according to the sequence number, place the leaf nodes into the unoccupied node set, place the unoccupied nodes included in the unoccupied node set at the middle of the removal sequence according to the node sequence number, and place the removed nodes included in the second removed node set at the end of the removal sequence to form a removal sequence; and connect the optimal sequence and the removal sequence to obtain a final sequence.

[0041] An embodiment of the present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes any one of the above-mentioned methods for controlling false information on social networks.

[0042] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes any one of the above-mentioned methods for controlling false information on social networks.

[0043] An embodiment of the present invention provides a method and device for controlling false information on a social network, the method comprising: constructing an initial propagation network, determining an initial sequence and an initial exchange threshold corresponding to the initial propagation network; traversing each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node, and when the connected component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the position of the first node in the initial sequence to obtain a first sequence; determining the first sequence with the smallest network resilience among multiple first sequences as the optimal sequence; dividing the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the optimal sequence; defining the connected components included in the initial propagation network as new nodes, a plurality of new nodes forming a newly occupied node set, and the first removed node set forming an unoccupied node set corresponding to the newly occupied node set; and determining the unoccupied node set according to the edges between the first occupied node in the first occupied node set and the first removed node in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set and the unoccupied node set. The invention relates to a method for forming a coarse-grained network by using a set of nodes; removing leaf nodes and central nodes from an unoccupied node set included in the coarse-grained network, and adding the central node to a second removed node set; setting the third node included in the coarse-grained network to an occupied state; when the unoccupied node set is a non-empty set, iterating according to the initial removal information of the third node, the initial root information of the third node, and belief propagation; when the belief propagation reaches a self-consistent steady state, determining the removal probability of each third node according to the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, deleting multiple third nodes with the highest probability from the coarse-grained network, and adding them to the second removed node set; placing the newly occupied nodes included in the newly occupied set at the beginning of the removed sequence according to the sequence number, placing the leaf nodes into the unoccupied node set, placing the unoccupied nodes included in the unoccupied node set in the middle of the removed sequence according to the node sequence number, and placing the removed nodes included in the second removed node set at the end of the removed sequence to form a removed sequence; and connecting the optimal sequence with the removed sequence to obtain a final sequence. Due to the adoption of a network coarse-graining strategy and a special node exchange strategy, the method provided by the embodiment of the present invention can be used to solve the problem of rapid network decomposition and has higher network decomposition efficiency, thereby timely and effectively controlling the spread of false information and reducing the losses caused by it. Furthermore, this method achieves network decomposition, removes fewer nodes, and spreads false information on a smaller scale, thereby reducing the cost of controlling the spread of false information. In theory, it can better protect social networks and achieve the purpose of controlling the spread of false information without excessively damaging network connectivity. Furthermore, this method has low time complexity and space complexity, high computational efficiency, and can quickly respond to sudden false information dissemination events. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 A schematic flow chart of a method for controlling false information on social networks provided by an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a coarse-grained operation provided by an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of the optimal sequence and removal sequence merging operation provided in an embodiment of the present invention;

[0048] Figure 4 A schematic flow chart of a method for controlling false information on social networks based on coarse-grained belief propagation provided in the first embodiment of the present invention;

[0049] Figure 5 A schematic diagram of the optimal sequence generation process provided in Example 1 of the present invention;

[0050] Figure 6 A schematic diagram of the results of obtaining the change in the proportion of removed nodes of the maximum connected component using different methods in the powergrid network, the p2p-Gnutella08 network, the ca-CondMat network, the com-Amazon network, the web-Google network, and the LiveJournal network provided by an embodiment of the present invention;

[0051] Figure 7 A schematic diagram of the calculation time using different methods in different networks provided by an embodiment of the present invention;

[0052] Figure 8 A schematic diagram of the structure of a device for controlling false information on a social network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] The network decomposition problem has been proven to be NP-hard (Non-deterministic Polynomial-time Hardness), attracting many researchers to design various methods to obtain optimal solutions. Among them, there are three common methods: centrality-based methods, percolation-based methods, and de-looping-based methods.

[0055] 1) Centrality-based methods: These include degree centrality, betweenness centrality, eigenvector centrality, PageRank, and closeness centrality. The idea is to score all nodes and remove those with the highest scores. This approach fails to account for the impact of node removal on the network structure and the changes in node scores. To address this shortcoming, researchers have proposed adaptive centrality-based methods, which re-score the remaining nodes in the network after each node removal. Due to their high time complexity, this approach struggles to decompose large networks. 2) Percolation-based methods: Percolation refers to the phenomenon in which, when a node joins a network, it connects with other nodes, resulting in the emergence of a maximum connected component. This method can be used to describe the super-propagation of misinformation, such as when an account in one community spreads misinformation to an account in another community that is unaware of the misinformation. Percolation-based methods embody a contrarian approach, such as the explosion immunity method, which identifies nodes that should not be blocked during the spread of misinformation and designs a candidate node sampling mechanism to reduce the search space. However, the parameters in these methods are difficult to adjust for networks with diverse characteristics, limiting their generalizability. 3) De-looping Methods: A loop is a network topology structure in which all nodes on a path from a node back to itself form a closed path. De-looping methods include the minimum sum method and belief propagation extraction method, which disconnect all loop structures, prune large tree structures, and re-add some unimportant nodes to the network. However, the application of message passing methods in this type of method relies on a key assumption: each node's neighbors are independent of each other and the network cannot have a loop structure. Therefore, these methods show mediocre performance on social networks with a large number of loop structures that are common in practice.

[0056] The aforementioned methods for controlling the spread of misinformation through network decomposition are ineffective because they require the removal of too many nodes, a critical issue in practical applications. In the early stages of misinformation, people lack sufficient understanding of the information and cannot verify its authenticity, so excessive account blocking is not recommended. However, during peak misinformation periods, when too many accounts participate in discussions, targeted blocking of influential accounts is necessary to minimize the impact on social network connectivity, rather than blocking accounts that participate disproportionately. Therefore, these existing methods struggle to effectively control the spread of misinformation.

[0057] To address the above issues, an embodiment of the present invention provides a method for controlling false information on social networks. First, target network information is extracted and a sequence is initialized based on the network information. Then, the node with the smallest connected component is continuously moved forward in the sequence to achieve preliminary optimization of the sequence. By occupying nodes according to the optimized sequence, the size of the largest connected component can be limited, thereby limiting the scale of false information propagation. A coarse-grained network is generated using the optimized sequence. Nodes are iteratively removed from the coarse-grained network using belief propagation, and a removal sequence is generated based on the order in which the nodes are removed. Finally, the optimized sequence and the removal sequence are connected to generate a final sequence, achieving further sequence optimization. The final sequence has the characteristics of both occupation optimization and removal optimization, which can further reduce the scale of false information propagation. The present invention can quickly and efficiently decompose large networks, thereby timely controlling the spread of false information and reducing the losses caused by it.

[0058] Figure 1 A flowchart of a method for controlling false information on a social network provided by an embodiment of the present invention is shown as follows: Figure 1 As shown, the method provided by the embodiment of the present invention mainly includes the following steps:

[0059] Step 101: construct an initial propagation network, determine an initial sequence and an initial exchange threshold corresponding to the initial propagation network; traverse each first node included in the initial sequence, obtain and occupy the smallest candidate node corresponding to each first node; when the connected component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the first node in the initial sequence to obtain a first sequence; determine the first sequence with the smallest network resilience among the multiple first sequences as the optimal sequence;

[0060] Step 102: Divide the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the optimal sequence; define the connected components included in the initial propagation network as new nodes, wherein the multiple new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; form a coarse-grained network based on the edges between the first occupied node in the first occupied node set and the first removed node in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set;

[0061] Step 103: remove leaf nodes and central nodes from the unoccupied node set included in the coarse-grained network, and add the central node to the second removed node set; set the third nodes included in the coarse-grained network to an occupied state, and when the unoccupied node set is a non-empty set, iterate based on the initial removal information of the third node, the initial root information of the third node, and belief propagation. When the belief propagation reaches a self-consistent steady state, determine the removal probability of each third node based on the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, delete multiple third nodes with the highest probability from the coarse-grained network, and add them to the second removed node set;

[0062] Step 104: Place the newly occupied nodes included in the newly occupied set at the beginning of the removal sequence according to the sequence number; put the leaf nodes into the unoccupied node set, and place the unoccupied nodes included in the unoccupied node set in the middle of the removal sequence according to the node sequence number; place the removed nodes included in the second removed node set at the end of the removal sequence to form a removal sequence; and connect the optimal sequence and the removal sequence to obtain a final sequence.

[0063] It should be noted that the method for controlling false information on social networks provided by the embodiments of the present invention is aimed at the nodes included in the initial propagation network, the nodes included in the initial sequence, the nodes included in the candidate range, the nodes included in the candidate set, the nodes included in the coarse-grained network, the nodes included in the first occupied node set, the nodes included in the first removed node set, etc. When explaining the method for controlling false information on social networks provided by the embodiments of the present invention, many nodes are involved. In order to avoid confusion between nodes appearing in different steps in the method, the nodes included in the initial propagation network are referred to as initial nodes; the nodes included in the initial sequence are referred to as first nodes; the nodes in the candidate range are referred to as second nodes; the nodes included in the candidate set are referred to as candidate nodes; the nodes included in the coarse-grained network are referred to as third nodes; the nodes corresponding to the first occupied node set are referred to as first occupied nodes, the nodes corresponding to the first removed node set are referred to as first removed nodes, the nodes corresponding to the newly occupied node set are referred to as newly occupied nodes, the nodes corresponding to the unoccupied node set are referred to as unoccupied nodes, the nodes corresponding to the second removed node set are referred to as second removed nodes, and the nodes corresponding to the removed sequence are referred to as fourth nodes.

[0064] In step 101, data related to the spread of disinformation is first input. This data includes the account information of the disseminating accounts, the number of accounts, and the interactions between accounts. Furthermore, using individuals involved in the disinformation spread as nodes and the interactions between individuals as edges, a disinformation spread network G(V, E) corresponding to the data is constructed. This network can also be called the initial dissemination network, with the node set V and the edge set E as the initial dissemination network.

[0065] Use degree centrality to sort all nodes in the network and construct an initial sequence S0. Determine the initial exchange threshold, where the initial exchange threshold is determined by the following formula (1):

[0066] θ λ =L(q c，0 )n (1)

[0067] Among them, θ λ represents the initial propagation threshold, n represents the size of the initial propagation network, that is, n=|V|, || represents the size of the set, L(q c，0 ) represents the initial maximum connected component ratio, which is the node ratio q is q c，0 The maximum proportion of connected components in the network.

[0068] Furthermore, the initial maximum connected component proportion L(q c，0 ) is determined by the following formula (2):

[0069]

[0070] Among them, C max is the set of all nodes in the largest connected component of the network, q c，0 The proportion of occupied nodes q when percolation occurs in the process of occupying nodes according to the initial sequence S0 is called the initial critical threshold q c，0 , which is determined by the following formula (3):

[0071] q c，0 =min{q|L(q)≥θ} (3)

[0072] Among them, θ is the maximum connected component ratio threshold for determining whether percolation occurs.

[0073] Furthermore, after the initial sequence and initial exchange threshold corresponding to the initial propagation network are determined, multiple optimizations may be performed on the initial propagation sequence.

[0074] Specifically, for each first node included in the initial sequence, if the second node within the candidate range of the first node is judged to be a candidate node according to the judgment probability, the second node is added to the candidate set as a candidate node; then for each candidate node in the candidate set, the connected component of each candidate node is determined; and the smallest candidate node with the smallest connected component is selected and occupied from each candidate set.

[0075] If the minimum connected component of the smallest candidate node is less than or equal to the initial exchange threshold, the smallest candidate node can be used to replace the position of the first node in the initial sequence, and then the initial exchange threshold is updated by the following formula (4).

[0076] θ λ =θ λ (y1+η) (4)

[0077] Among them, θ λ represents the initial exchange threshold, which is used to determine whether positions need to be exchanged. y1 is a random number uniformly distributed in (0, 1]. η is the update amplitude of the initial exchange threshold, which is an adjustable parameter greater than 1.

[0078] It should be noted that in the above steps, the judgment probability is determined by the following formula (5), and the candidate range is determined by the following formula (6-1):

[0079]

[0080] V(i)={S0[v]|v∈[γ i ,min(γ i +m,n-1)]} (6-1)

[0081] Among them, γ i represents the index of the first node i in the initial sequence, S0[v] represents the (v+1)th first node in the initial sequence, and m is an adjustable parameter.

[0082] In the embodiment of the present invention, after all first nodes included in the initial sequence are traversed in order, a first sequence can be obtained, and the network resilience is calculated based on the first sequence.

[0083] Network resilience represents the effectiveness of controlling the spread of false information. A smaller value for network resilience indicates a better control of the spread of false information. The formula for determining network resilience is as follows:

[0084]

[0085] In formula (7-1), F represents network resilience and n represents the size of the initial propagation network.

[0086] Furthermore, based on the relationship between the network resilience determined by the first sequence and the network resilience determined by the initial sequence, the first sequence or the initial sequence is determined to be the optimal sequence. For example, after the initial sequence S0 completes the first traversal, the first sequence S is obtained. The network resilience determined by the first sequence S is F, and the network resilience determined by the initial sequence S0 is F0. When F≤F0, the optimal sequence is the first sequence, that is, S I =S; when F>F0, the optimal sequence is the initial sequence, that is, S I =S0, where S I Represents the optimal sequence obtained after the I-th optimization.

[0087] In the embodiment of the present invention, when Imax After parallel optimization, the sequence S with the smallest network resilience F is selected. I , which can be written as

[0088] It should be noted that, in the embodiment of the present invention, when the initial sequence is I max During secondary optimization, the initial sequence may be a complete sequence, or it may include a partial fragment of the initial sequence.

[0089] Specifically, a segment is cut from the initial sequence. For each first node included in the segment, if the second node within the candidate range of the first node is judged to be a candidate node according to the judgment probability, the second node is added to the candidate set as a candidate node; then for each candidate node in the candidate set, the connected components of each candidate node are determined, and the smallest candidate node with the smallest connected component is selected and occupied from each candidate set.

[0090] If the minimum connected component of the smallest candidate node is less than or equal to the initial exchange threshold, the smallest candidate node can be used to replace the position of the first node in the segment. After traversing all the first nodes included in the segment in order, a first sequence can be obtained.

[0091] Furthermore, after sequentially traversing all the first nodes included in the fragment, a first sequence can be obtained. The network resilience is calculated based on the first sequence. Here, in order to distinguish the network resilience determined based on the fragment from the network resilience determined based on the entire sequence, the network resilience determined based on the fragment is called marginal network resilience. The marginal network resilience is determined by the following formula (7-2):

[0092]

[0093] Among them, the marginal network resilience ΔF represents the marginal false information propagation control effect when all the first nodes in the segment are removed. The smaller its value, the better the false information propagation control effect.

[0094] In an embodiment of the present invention, the fragment cutting operation reduces the number of nodes that need to be processed in a single exchange operation, which can effectively reduce the complexity of the method and speed up the calculation speed; and after simulation experiments, it has been verified that after adding the fragment cutting operation, the method of the present invention can control the spread of false information in a large network to a smaller scale in a shorter time.

[0095] It should be noted that in this step, the fragment can be determined by the following formula (8):

[0096] κ=S[γ s :min(γ e +1,n)] (8)

[0097] Here, S[a:b] represents a segment of S, whose starting element is indexed as a and whose ending element is indexed as b-1.

[0098] Furthermore, the candidate range can be determined by the following formula (6-2):

[0099] V(i)={S[v]|v∈[γ i ,min(γ i +m,γ e )]} (6-2)

[0100] In the embodiment of the present invention, the selection of the fragment may include random slicing as one fragment, or two-stage slicing as one fragment. Specifically:

[0101] (1) When a segment is a random slice, the index of the starting element of the segment is determined by the following formula (9):

[0102]

[0103] The index of the end element of the segment is determined by the following formula (10):

[0104] γ e =γ s +y2mod(φ2n-10)+10 (10)

[0105] Among them, γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, y2 is a uniformly distributed index in [1, 10 9 ] is a random integer. φ1 and φ2 are two adjustable parameters.

[0106] (2) When a segment is a two-stage slice, the index of the starting element of the segment is determined by the following formula (11):

[0107] γ s =q c，T n (11)

[0108] The index of the end element of the segment is determined by the following formula (12):

[0109]

[0110] Among them, γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, φ3 is an adjustable parameter, q c，T It is the ratio q of occupied nodes when percolation occurs during the process of occupying nodes in the initial sequence, which is determined by the following formula (13):

[0111] q c，T =min{q|L(q)≥θ T} (13)

[0112] L(q) is the maximum connected component proportion when the occupied node proportion is q, which is determined by the following formula (14):

[0113]

[0114] θ T is the maximum connected component ratio threshold in round T, and its calculation formula is shown in the following (15):

[0115]

[0116] Among them, φ4>1 and φ5 are two adjustable parameters. y3 is a random number uniformly distributed in [0, 1).

[0117] In step 102, according to the optimal sequence obtained in step 101, the initial nodes included in the initial propagation network are divided into first occupied nodes and first removed nodes, multiple first occupied nodes form a first occupied node set, and multiple first removed nodes form a first removed node set.

[0118] Specifically, when the initial propagation network forms the maximum connected component, the proportion of the first occupied node of the optimal sequence in the initial propagation network is determined as the critical threshold. The initial nodes included in the initial propagation network are divided into a first occupied node set and a first removed node set according to the critical threshold. The first occupied node set is determined by the following formula (16):

[0119]

[0120] The first set of removed nodes is determined by the following formula (17):

[0121]

[0122] Among them, V o represents the first occupied node set, V u represents the first set of removed nodes, express The corresponding critical threshold, represents the optimal sequence.

[0123] For example, to get the optimal sequence Then, according to the optimal sequence Occupy nodes in an empty network. When the maximum connected component is formed in the empty network, the proportion of occupied nodes at this time is recorded as the critical threshold. Can be used to convert the optimal sequence The nodes in the are divided into two categories: one is called the first occupied nodes, which are located in the optimal sequence The first half of the equation is shown in formula (16). Once all the first occupied nodes are occupied, the maximum connected component is generated. One type is called the first removed nodes, which are located in the optimal sequence. The second half of , as shown in formula (17). Once all the first removed nodes are removed, there is no maximum connected component.

[0124] Furthermore, each connected component in the initial propagation network is regarded as a new node, and multiple new nodes form a new occupied node set, while the first removed node set remains unchanged, directly forming an unoccupied node set. Figure 2 As shown in the figure, the initial propagation network is G, and the initial nodes included in the initial propagation network G are divided into the first occupied nodes o∈V o and remove node u∈V u , the first occupied node can form multiple connected components, therefore, each connected component is defined as a new node, and multiple new nodes form a new occupied node set, such as Figure 2 o′∈V in o ′ is the newly occupied node; in the coarse-grained operation, if the first removed node included in the first removed node set does not change, the unchanged first removed node can be called the unoccupied node set corresponding to the newly occupied node set, where the unoccupied node set can be expressed as u′∈V u ′.

[0125] After the new nodes and initial nodes included in the initial propagation network are divided into a newly occupied node set and an unoccupied node set based on the connected components, a coarse-grained network can be constructed. Specifically, the edges between the first occupied nodes in the first occupied node set are no longer considered. The coarse-grained network is composed of the edges between the first occupied nodes in the first occupied node set and the first removed nodes in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set.

[0126] For example, the first occupied node set V o The first occupied node and the first removed node set V in u The edges of the first removed node, the first removed node set V u The edges between the first removed nodes are extracted to form a new edge combination E′, and then V′=V o ′∪V u ' and E' can be used to form a coarse-grained network G', whose network size is n' = |V'|. Figure 2 As shown, the first occupied node set V in the initial propagation network G oThe first occupied node included in forms multiple connected components, and the above connected components are positioned as a new node to form a new occupied node set V o ′, in the coarse-grained network G′, the edges between the first removed nodes in the original first removed node set, the edges between the first removed node and the first occupied node in the original first occupied node set, the first removed node in the first removed node set, and the new nodes in the new occupied node set are retained.

[0127] In step 103, the leaf nodes and central nodes in the unoccupied node set included in the coarse-grained network are removed, and then a second removed node set is constructed, and the central node deleted from the unoccupied node set is added to the second removed node set.

[0128] It should be noted that the central node in the unoccupied node set can be the first ψ third nodes with the highest degree centrality among the third nodes included in the coarse-grained network, which are determined as the central node; or the new node formed by the ψ second nodes at the end of the optimal sequence in the coarse-grained network can be determined as the central node. In the embodiment of the present invention, ψ is a positive integer greater than zero.

[0129] Furthermore, after the leaf nodes and the central node included in the coarse-grained network are removed, the coarse-grained network is traversed to obtain a plurality of third nodes with the highest removal probability in the coarse-grained network.

[0130] Specifically, all third nodes included in the coarse-grained network are set to an occupied state. When the set of unoccupied nodes is a non-empty set, iterations are performed based on the initial removal information of the third node, the initial root information of the third node, and belief propagation to obtain the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations.

[0131] Furthermore, when belief propagation reaches a self-consistent steady state, the removal probability of each third node is determined based on the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, and the multiple third nodes with the highest probability are deleted from the coarse-grained network, added to the second removed node set, and deleted from the unoccupied node set.

[0132] In this embodiment of the present invention, the initial removal information of the third node can be expressed as in, It is expressed as the probability that node i should be removed when node j has been removed. The initial root information of the third node can be expressed as in, It is expressed as the probability that node i is the root node of its connected component when node j has been removed.

[0133] The above belief propagation formula is as follows:

[0134]

[0135]

[0136]

[0137] Among them, e x The x in the example is an adjustable parameter, which is usually set to 12. is the set of occupied neighbor nodes of node i when the node is removed in round T′. for Eliminate the node set from which node j is removed.

[0138] The self-consistent steady state,removal probability is determined by the following formula:

[0139]

[0140]

[0141]

[0142] Among them, ∈1 and ∈2 are error values for judging whether the iteration can be terminated. represents the removal information of node i after t′ iterations, represents the root information of node i after t′ iterations, Represents the third node i∈V u The probability that ′ should be removed.

[0143] In step 104, the construction of the removal sequence includes two parts: preliminary construction of the removal sequence and optimization of the removal sequence. The preliminary construction of the removal sequence includes: first, placing the second removal node included in the second removal node set at the end of the removal sequence as the fourth node in the reverse order of the removal order of the third node, for example, the first second removal node removed is placed at S′[n′-1]; then, placing the leaf nodes into the unoccupied node set, and placing the unoccupied nodes included in the unoccupied node set in the middle of the removal sequence according to the node sequence number; finally, placing the newly occupied nodes in the newly occupied node set at the beginning of the removal sequence according to the sequence number, thereby obtaining the preliminary construction of the removal sequence.

[0144] The optimization of the initially constructed removal sequence includes: temporarily occupying the second removal nodes included in the second removal node set in turn, checking whether each second removal node is in the ring structure, and if it is determined that the second removal node is not in the ring structure, occupying the second removal node and replacing the fourth node at the front of the removal sequence with the second removal node, and then removing the second removal node from the second removal node set and placing it in the unoccupied node set.

[0145] For example, temporarily occupy the second removed node set V r ′, check whether the second removed node r is in the ring structure. If not, occupy the second removed node r and swap the position of the second removed node r with the fourth node at the front of the removal sequence in the removal sequence S′. The fourth node at the front (the second removed node after position replacement) is S′[n′-|V r ′|], and remove the second node r from the second removed node set V r ′ and put it into the unoccupied node set V u ′.

[0146] Remove all nodes j∈V included in the second removed node set and the unoccupied node set in the network u ′∪V r ', let l = 0. Temporarily occupy the unoccupied node j in turn. If the size of the connected component where the unoccupied node j is located is λ j ≤θ λ ′, then occupy the unoccupied node j, and exchange the position of the unoccupied node j with the first unoccupied node in the removal sequence in S′. The first unoccupied node in the removal sequence is s′[|V′ o +l|]. After each exchange of positions, make θ λ ′=θ λ '+eta' and l=l+1. where θ λ ′ is the exchange threshold used in belief propagation, η′ is θ λ ′ is the update amplitude, and l is the number of records exchanged.

[0147] Furthermore, the optimal sequence obtained in step 101 and the removed sequence obtained above are connected to obtain a final sequence.

[0148] For the optimal sequence obtained in step 101 and the removed sequence obtained above, according to the number of original network nodes n, the number of coarse-grained network nodes n′, the critical threshold corresponding to the optimal sequence and occupy node size n′, respectively taking the optimal sequence and removing the sequence fragments and S′[|V o ′|:n′], and connect the two fragments to obtain the final sequence Sf For example, Figure 3 As shown in the figure, the dotted line represents the node boundary corresponding to the critical threshold in the sequence. The gray segments in the optimal sequence and the removed sequence are ignored, and the segment on the left of the dotted line of the optimal sequence and the segment on the right of the dotted line of the removed sequence are connected to form the final sequence.

[0149] It should be noted that after obtaining the final sequence through the above steps, an empty network identical to the initial propagation network can be established. This network, with the same structure as the initial propagation network, contains no nodes. Based on the final sequence, nodes can be continuously occupied in the empty network, and these occupied nodes will form many connected components. When the number of occupied nodes reaches a certain limit, a maximum connected component emerges among the connected components, and its proportion exceeds the maximum connected component proportion threshold, indicating percolation. At this point, it can be considered that misinformation has spread widely within the social network.

[0150] In practical applications, if nodes are occupied according to different sequences, percolation will occur when different numbers of nodes are occupied. However, according to the final sequence provided by the embodiment of the present invention, percolation will only occur when more nodes are occupied. This means that the method provided by the embodiment of the present invention can greatly block the spread of false information by blocking only a small number of accounts in the social network, and can control the spread of false information to a smaller range while ensuring strong network connectivity.

[0151] In summary, the method for controlling false information on social networks provided by the embodiments of the present invention, due to the adoption of a network coarse-graining strategy and a special node exchange strategy, can be used to solve the problem of rapid network decomposition and has higher network decomposition efficiency, thereby timely and effectively controlling the spread of false information and reducing the losses caused by it; furthermore, the method realizes network decomposition, removes fewer nodes, and the scale of false information propagation is smaller, so the cost of controlling the spread of false information is lower. In theory, it can play a better protective role for social networks and achieve the purpose of controlling the spread of false information without excessively damaging network connectivity; furthermore, the method has low time complexity and space complexity, high computational efficiency, and can quickly respond to sudden false information propagation events.

[0152] In order to more clearly introduce the method for controlling false information on social networks provided by the embodiment of the present invention, the following Figure 4 and Figure 5 Taking the example, the method for controlling false information in social networks based on coarse-grained belief propagation provided by the first embodiment of the present invention is introduced in detail.

[0153] like Figure 4 As shown, the method mainly includes the following steps:

[0154] Step 201: Input data related to the spread of false information, including the account information of the spreading accounts, the number of accounts, and the interactions between accounts. Using individuals involved in the spread of false information as nodes and the interactions between individuals as edges, a false information dissemination network G(V, E) corresponding to the data is constructed, with the network's node set V and edge set E. All nodes in the network are sorted using degree centrality, an initial sequence S0 is constructed, and an initial exchange threshold is determined.

[0155] The initial exchange threshold is determined by formula (1), formula (2) and formula (3).

[0156] Step 202: for the initial sequence S0, execute I max Parallel optimization, in I max In the optimization sequence S I Select the sequence with the smallest network resilience F The optimal sequence

[0157] Specifically, generate the optimal sequence The detailed process is as follows Figure 5 As shown, the following steps are included:

[0158] Step 202-1: Set all initial nodes included in the initial propagation network G to the removed state. For each first node i in the initial sequence S0, determine whether each second node in the candidate range V(i) of the first node i is a candidate node with probability p. If so, place the second node in the candidate set V(i). h For the candidate set V h For each candidate node h in the network, calculate the size of the connected component, denoted as λ h , occupying V h Medium h The smallest candidate node h min .

[0159] Step 202-2, if min(λ h )≤θ λ , then execute step 202-3; otherwise, execute step 202-1.

[0160] Step 202-3, exchange h min and the position of the first node i in the initial sequence, obtain the first sequence S, and update the exchange threshold θ λ =θ λ (y1+η). Where, θ λ is the exchange threshold for determining whether positions need to be exchanged, y1 is a random number uniformly distributed in (0, 1], and η is the update amplitude of the exchange threshold, which is an adjustable parameter greater than 1.

[0161] Step 202-4, traverse all first nodes i in sequence order, if yes, execute step 202-5, otherwise execute step 202-1;

[0162] Step 202-5: Calculate the network resilience F based on the obtained first sequence S.

[0163] Step 202-6: Network resilience F represents the effect of controlling the spread of false information. The smaller the value, the better the effect of controlling the spread of false information. If F≤F0, execute step 202-7 and set S I =S; otherwise, execute step 202-8, let S I =S0.

[0164] Completion I max After parallel optimization, the sequence S with the smallest network resilience F is selected. I , recorded as Among them, F0 is the network resilience calculated according to the initial sequence S0, S I This is the sequence obtained from the first parallel optimization.

[0165] In the embodiment of the present invention, the optimal sequence may also be generated by the following method. That is, step 202 may also obtain the optimal sequence by the following alternative method. Specifically:

[0166] When T=0, the initial sequence S0 obtained in step 1 is set to S0; 0<T≤T max When the best sequence obtained in the previous round T is Set it as S0. Cut out a segment from S0, and the segment considered in the Tth round is determined by formula (8):

[0167] κ=S[γ s :min(γ e +1,n)] (8)

[0168] For fragment κ, execute I in parallel max Suboptimization.

[0169] The optimization content is as follows: Let t = 0, ΔF * =1, all nodes in the network are set to removed state. When t≤t max When, for each node i∈κ in the segment, in the candidate range V(i)={S[v]|v∈[γ i ,min(γ i +m,γ e )]} Select candidate nodes for probability and form a candidate set V h , where m=m0t+1, m0 is an adjustable parameter.

[0170] For Vh For each candidate node h in the network, calculate the size of the connected component λ h . Occupy V h Medium h The smallest candidate node h min If min(λ h )≤θ λ , then exchange h min and the position of node i in segment κ, and let θ λ =θ λ (y1+η). After traversing all nodes i, the marginal network resilience ΔF is calculated based on the current sequence S, as shown in formula (7-2):

[0171]

[0172] Marginal network resilience ΔF represents the marginal false information propagation control effect when all nodes in the segment are removed. The smaller its value, the better the false information propagation control effect. * , then let ΔF * =ΔF,S * =S.

[0173] Then, let t = t + 1 and return to select candidate nodes. max After the exchange on the fragment, if ΔF * ≤ΔF0, then let S I =S * Otherwise, let S I =S0. Among them, ΔF0 is the marginal network resilience calculated according to S0, S I This is the sequence obtained from the first parallel optimization.

[0174] Completion I max After parallel optimization, the sequence S with the smallest ΔF is selected. I , recorded as

[0175] In this embodiment, there are two ways to determine the above-mentioned fragments. Please refer to the above-mentioned embodiments for details, and no further details will be given here.

[0176] Step 203: The optimal sequence obtained in step 202 Divide the initial nodes included in the initial propagation network G into the first occupied node set V o and the first removed node set V u .

[0177] Furthermore, in order to coarse-grain the initial propagation network G, that is, to obtain the coarse-grained network G′ based on the initial propagation network G, each connected component in the initial propagation network G is regarded as a new node, and multiple new nodes are combined into a new occupied node set V o ′. The original first removed node set V u Remain unchanged and directly form the unoccupied node set V u ′.

[0178] No longer consider the first occupied node set V o The edge between the first occupied nodes in the o The first occupied node and the first removed node set V u The edges of the first removed node and the first removed node set V u The edges between the first removed nodes are extracted to form a new edge combination E′. At this point, V′=V o ′∪V u ′ and E′ can be used to form a coarse-grained network G′, whose network size is n′=|V′|.

[0179] Occupy all third nodes in the coarse-grained network G'. Remove the set of unoccupied nodes V from the coarse-grained network G'. u ′, and then construct the second removed node set V r ', add the central node to the second removed node set V r 'Inside.

[0180] Step 204, set T' = 0. When T' ≤ T' max or V u When there are still nodes in ′, the T′th round of node removal begins.

[0181] Let t′=0, randomly generate a decimal between [0, 1], and assign it to the initial removal information and initial root information in, is the probability that node i should be removed when node j has been removed; is the probability that node i is the root node of its connected component when node j has been removed.

[0182] Step 205, enter the belief propagation iteration, according to Iterate with the belief propagation formula to calculate and

[0183] Step 206, let t'=t'+1, according to and belief propagation formula and If it is determined that belief propagation has reached a self-consistent steady state, step 207 is executed; otherwise, step 205 is executed.

[0184] Step 207, let And calculate each node i∈V u The probability that ′ should be removed Remove from network The highest first θ nodes i∈V u ′, and move these nodes from V u ' and put it into the second removed node set V r ′, let T′=T′+1.

[0185] Step 208, T'<T' max Or unoccupied node set V u ', execute step 204; otherwise, execute step 209.

[0186] Step 209: Construct a removal order S′, and remove the second node set V in reverse order of the node removal order. r The second removed node in ' is placed at the end of S'. Put the leaf node into the unoccupied node set V u ′, and the unoccupied node set V u The unoccupied nodes in ′ are placed in the middle of S′ according to the node sequence number. o The newly occupied nodes in ′ are placed at the beginning of S′ according to the node sequence number.

[0187] Step 210: Temporarily occupy the second removed node set V in sequence r ', check whether the second removed node is in the ring structure, if not, execute step 211.

[0188] Step 211: For all second removed nodes that are not in the ring structure, occupy the second removed node and exchange the second removed node with S'[n'-|V r ′|], remove the second node from the second removed node set V r ′ and put it into the unoccupied node set V u ′.

[0189] Step 212: Remove all nodes j∈V in the network. u ′∪V r ', let l = 0. Temporarily occupy node j in turn, if the size of the connected component where it is located is λ j ≤θ λ ’, then execute step 213.

[0190] Step 213, occupy j and exchange j with S′[|Vo ′+l|], so that θ λ ′=θ λ '+n' and l=l+1. where θ λ ′ is the exchange threshold used in belief propagation, η′ is θ λ ′ is the update amplitude, and l is the number of records exchanged.

[0191] Step 214: Connect the optimal sequence obtained in the above steps with the removed sequence S′ to obtain the final sequence S f .

[0192] To verify the effectiveness of the method for controlling false information on social networks provided by an embodiment of the present invention, an experiment was conducted on a false information propagation network, and its network parameters are shown in Table 1.

[0193] Table 1 Network datasets used in the experiment

[0194] Dataset Number of nodes Number of sides Euroroads 1174 1417 Yeast 2375 11693 powergrid 4941 6594 p2p-Gnutella08 6301 20777 ca-CondMat 23133 93497 email-Enron 36692 183831 loc-Gowalla 196591 950327 com-Amazon 334863 925872 web-Google 875713 4322051 roadNet-PA 1379917 1921660 as-Skitter 1696415 11095298 LiveJournal 3997962 34681189

[0195] Six commonly used network decomposition methods, including DC (Degree Centrality), ADC (Adaptive Degree Centrality), CI (Collective Influence), BPD (Belief Propagation-guided Decimation), GND (Generalized Network Dismantling), FINDER (FInding key players in Networks through DEEP Reinforcement learning), and ECA-I (Effective Combined Attack-I) and ECA-II (Effective Combined Attack-II) provided in an embodiment of the present invention, were selected to conduct a network decomposition performance comparison experiment. The experimental content is: the network in Table 1 is decomposed using the above methods, and a node removal sequence S is generated according to the order in which the nodes are removed during the decomposition process. r , records the proportion of removed nodes when seepage occurs during the decomposition process Recorded as removal threshold Determined by the following formula:

[0196]

[0197] in, Indicates that when the ratio of removed nodes is In the embodiment of the present invention, the maximum connected component ratio threshold θ for determining whether the seepage phenomenon occurs is set to 0.01.

[0198] Table 2 Removal thresholds obtained by different methods on different networks

[0199]

[0200] As shown in Table 2, the bold numbers represent the minimum values in each row. The ECA-I improvement degree is the removal threshold of the ECA-I method relative to each method. The degree of reduction is calculated as follows:

[0201]

[0202] Where σ(Λ) is the removal threshold of the ECA-I method relative to the method Λ The degree of reduction, is the removal threshold of method Λ is the removal threshold of the ECA-I method

[0203] From the experimental results, it can be seen that the methods ECA-I and ECA-II provided by the embodiment of the present invention are relatively good. On most networks with less than 105 nodes, the removal threshold obtained by the ECA-I method is The minimum removal threshold is obtained by the ECA-II method often on networks with more than 105 nodes. Compared with the DC and GND methods, the removal threshold of the ECA-I method The average reduction is more than 60%; compared with ADC, CI, and FINDER methods, the removal threshold of the ECA-I method is The average reduction is more than 40%; compared with BPD and ECA-II methods, the removal threshold of ECA-I method is Slightly reduced. It can be seen that compared to other methods, the ECA-I and ECA-II methods provided by embodiments of the present invention require fewer nodes to be removed to control the spread of false information. The average proportion of nodes required to be removed for the ECA-I method is slightly lower than that for the ECA-II method. In real-world situations, when the spread of false information is controlled, the method provided by embodiments of the present invention results in a lower percentage of blocked accounts.

[0204] As shown in Table 2, a "-" indicates that the computation took too long or exceeded memory limits. For large-scale networks such as as-Skitter and LiveJournal, the CI, GND, and FINDER methods take too long to complete network decomposition, making it difficult to complete the task. The ECA-I and ECA-II methods provided by the embodiments of the present invention can achieve network decomposition in large-scale networks such as roadNet-PA, as-Skitter, and LiveJournal. This shows that the methods provided by the embodiments of the present invention perform well on large-scale networks.

[0205] Figure 6 The results of the change of the maximum connected component ratio with respect to the ratio of removed nodes obtained by different methods on powergrid (a), p2p-Gnutella08 (b), ca-CondMat (c), com-Amazon (d), web-Google (e), and LiveJournal (f) networks are given. In the figure, the horizontal axis is is the ratio of nodes removed from the network, the vertical axis is the proportion of the maximum connected component; the methods ECA-I and ECA-II provided in the embodiments of the present invention often have an obvious and early falling inflection point, and the proportion of removed nodes is When the value is smaller, the maximum connected component ratio is lower. For example, on the ca-CondMat network, when When ECA-I and ECA-II methods obtain is about 0.15, and the CI method obtains is about 0.5, and the BPD method obtains is about 0.65, and the method provided in the embodiment of the present invention obtains Much smaller than other methods Corresponding to the actual situation, under the condition of blocking the same proportion of accounts, using the method provided by the embodiment of the present invention to control the spread of false information can reduce the scale of the spread of false information and control the spread of false information to a greater extent.

[0206] Figure 7A comparison chart of the computation time of different methods on different networks is provided. In the figure, the horizontal axis represents different networks, namely p2p-Gnutella08, ca-CondMat, email-Enron, loc-Gowalla, web-Google, and as-Skitter networks, and the vertical axis represents the computation time. Compared with the CI, BPD, and FINDER methods, the ECA-I and ECA-II methods provided by the embodiments of the present invention significantly reduce computation time and significantly improve efficiency. For example, the computation speed of the ECA-I method is approximately 373 times faster than that of the CI method on the email-Enron network and approximately 7 times faster than that of the BPD method on the p2p-Gnutella08 network. On some networks, such as email-Enron and as-Skitter networks, the ECA-II method is slightly faster than the ECA-I method. It can be seen that the methods provided by the embodiments of the present invention have high computational efficiency and can quickly respond to sudden false information dissemination events.

[0207] The spread of false information is very similar to the spread of infectious diseases, so researchers often use infectious disease models to study the spread of false information. The method provided in the embodiment of the present invention uses the Susceptible-Infected-Recovered (SIR) model to conduct a controlled false information spread scale measurement experiment. The experimental process is as follows: after removing the node sequence S calculated by each method, r Take out a fragment S r [n-ζn-1:n-1], removing all nodes from the segment to simulate account blocking, where ζ is an adjustable ratio of removed nodes. Disinformation begins to spread. The model contains three different states of nodes: susceptible, infected, and recovered. Susceptible individuals will be infected with a certain probability, becoming infected. Infected individuals will become recovered after a period of illness, and recovered individuals will not change their status. At a specific moment, the scale of disinformation spread Γ after removing the same ratio of nodes for each method is observed, reflecting the control effectiveness of each method.

[0208] Table 3. Scale of false information propagation Γ after removing the same proportion of nodes using different methods

[0209]

[0210] As shown in Table 3, the bold numbers represent the minimum values within each row. The ECA-I improvement degree is the degree to which the false information propagation scale Γ of the ECA-I method is reduced relative to that of each method, and is calculated as follows:

[0211]

[0212] Among them, σ′(Λ) is the degree of reduction of the false information propagation scale Γ of the ECA-I method relative to method Λ, Γ(Λ) is the false information propagation scale Γ obtained by method Λ, and Γ(ECA-I) is the false information propagation scale Γ obtained by the ECA-I method.

[0213] The experimental results show that, on most networks, the disinformation control methods ECA-I and ECA-II provided by the embodiments of the present invention achieve the smallest dissemination scale. Compared with the DC, ADC, GND, and FINDER methods, the disinformation spread scale Γ of the ECA-I method is reduced by an average of over 80%; compared with the CI method, the disinformation spread scale Γ of the ECA-I method is reduced by an average of approximately 29%; and compared with the BPD and ECA-II methods, the disinformation spread scale Γ of the ECA-I method is reduced by an average of over 10%. It can be seen that compared with other methods, the disinformation spread of the ECA-I and ECA-II methods provided by the embodiments of the present invention is often minimized, with the disinformation spread of the ECA-I method being smaller than that of the ECA-II method. In real-world scenarios, when blocking the same proportion of accounts, the disinformation control methods provided by the embodiments of the present invention can achieve a smaller spread of disinformation.

[0214] Based on the same inventive concept, an embodiment of the present invention provides a device for controlling false information on a social network. Since the principle of solving the technical problem of the device is similar to that of a method for controlling false information on a social network, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0215] Figure 8 A schematic diagram of the structure of a device for controlling false information on a social network provided by an embodiment of the present invention is shown in FIG. Figure 8 As shown, the apparatus includes a determining unit 801 , a composing unit 802 , a removing unit 803 and an obtaining unit 804 .

[0216] Determination unit 801 is configured to construct an initial propagation network and determine an initial sequence and an initial exchange threshold corresponding to the initial propagation network; traverse each first node included in the initial sequence, obtain and occupy the smallest candidate node corresponding to each first node, and when the connectivity component of the smallest candidate node is less than the initial exchange threshold, replace the first node in the initial sequence with the smallest candidate node to obtain a first sequence; and determine the first sequence with the smallest network resilience among the multiple first sequences as the optimal sequence;

[0217] The composition unit 802 is configured to divide the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the optimal sequence; define the connected components included in the initial propagation network as new nodes, wherein the multiple new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; and form a coarse-grained network based on the edges between the first occupied node in the first occupied node set and the first removed node in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set.

[0218] a removal unit 803 for removing leaf nodes and central nodes from an unoccupied node set included in the coarse-grained network, and adding the central node to a second removed node set; setting a third node included in the coarse-grained network to an occupied state, and when the unoccupied node set is a non-empty set, iterating based on initial removal information of the third node, initial root information of the third node, and belief propagation; when the belief propagation reaches a self-consistent steady state, determining a removal probability of each third node based on removal information of the third node after multiple iterations and root information of the third node after multiple iterations, deleting multiple third nodes with the highest probability from the coarse-grained network, and adding them to the second removed node set;

[0219] The obtaining unit 804 is configured to place the newly occupied nodes included in the newly occupied set at the beginning of the removal sequence according to the sequence number, place the leaf nodes into the unoccupied node set, place the unoccupied nodes included in the unoccupied node set in the middle of the removal sequence according to the node sequence number, and place the removed nodes included in the second removed node set at the end of the removal sequence to form a removal sequence; and concatenate the optimal sequence and the removal sequence to obtain a final sequence.

[0220] It should be understood that the units included in the aforementioned device for controlling false information on social networks are merely logical divisions based on the functions implemented by the device. In actual applications, these units can be combined or separated. Furthermore, the functions implemented by the device for controlling false information on social networks provided in this embodiment correspond one-to-one with the method for controlling false information on social networks provided in the aforementioned embodiment. A more detailed description of the processing flow implemented by this device has been provided in the aforementioned method embodiment 1 and will not be repeated here.

[0221] Another embodiment of the present invention also provides a computer device, which includes: a processor and a memory; the memory is used to store computer program code, and the computer program code includes computer instructions; when the processor executes the computer instructions, the electronic device executes each step of a method for controlling false information on a social network in the method flow shown in the above method embodiment.

[0222] Another embodiment of the present invention further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer device, the computer device executes each step of a method for controlling false information on a social network in the method flow shown in the above method embodiment.

[0223] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0224] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for controlling false information on social networks, characterized in that: include: Construct an initial propagation network and determine the initial sequence and initial exchange threshold corresponding to the initial propagation network; Traversing each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node, and when the connectivity component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the first node in the initial sequence to obtain a first sequence; determining the first sequence with the smallest network resilience among the multiple first sequences as the optimal sequence; The initial nodes included in the initial propagation network are divided into a first occupied node set and a first removed node set according to the optimal sequence; the connected components included in the initial propagation network are defined as new nodes, the multiple new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; a coarse-grained network is formed according to the edges between the first occupied node in the first occupied node set and the first removed node in the first removed node set, the edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set; The leaf nodes and the central node in the unoccupied node set included in the coarse-grained network are removed, and the central node is added to the second removed node set; the third node included in the coarse-grained network is set to an occupied state, and when the unoccupied node set is a non-empty set, iterate according to the initial removal information of the third node, the initial root information of the third node, and the belief propagation. When the belief propagation reaches a self-consistent steady state, the removal probability of each third node is determined according to the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, and the multiple third nodes with the highest probability are deleted from the coarse-grained network and added to the second removed node set; The newly occupied nodes included in the newly occupied set are placed at the beginning of the removal sequence according to the sequence number, the leaf nodes are placed in the unoccupied node set, and the unoccupied nodes included in the unoccupied node set are placed in the middle of the removal sequence according to the node sequence number, and the removed nodes included in the second removed node set are placed at the end of the removal sequence to form a removal sequence; the optimal sequence and the removal sequence are connected to obtain a final sequence.

2. The method according to claim 1, wherein The traversal of each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node, when the connected component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the position of the first node in the initial sequence to obtain the first sequence, further comprising: Traversing each first node in the first segment included in the initial sequence, and if a second node within the candidate range corresponding to the first node is determined to be a candidate node based on a judgment probability, adding the second node to the candidate set to form a candidate node; Determine the connected components of each candidate node included in the candidate set and the smallest candidate node with the smallest connected component, and occupy the smallest candidate node; When the connected component of the smallest candidate node is less than the initial exchange threshold, swapping the positions of the smallest candidate node and the first node in the initial sequence, and updating the initial exchange threshold; After traversing each first node included in the first segment, a first sequence is obtained, and the network resilience of the first sequence is determined.

3. The method according to claim 2, wherein The first segment is a random segment in the initial sequence, and the index of the starting element of the random segment is determined by the following formula: The index of the end element of the random segment is determined by the following formula: c e =c s +y2mod(φ2n-10)+10 Among them, γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, y2 is a uniformly distributed index in [1,10 9 ] is a random integer, φ1 and φ2 are two adjustable parameters, q c,0 represents the initial critical threshold corresponding to the initial sequence, n represents the size of the initial propagation network, T represents the current number of iterations in the iterative node occupation process, T max The maximum number of iterations to occupy the node process for a given iteration.

4. The method according to claim 2, wherein The first segment is a two-stage segment in the initial sequence, and the index of the starting element of the two-stage segment is determined by the following formula: c s =q c,T n The index of the ending element of the two-phase segment is determined by the following formula: Among them, q c,T represents the critical threshold corresponding to the T-th round time series, q c,T =min{q|L(q)≥θ T }; γ s Indicates the index of the starting element of the fragment, γ e Indicates the index of the end element of the fragment, θ T is the maximum connected component ratio threshold in round T, φ3, φ4>1, Φ5 are adjustable parameters, y2 is a uniform distribution in [1,10 9 ], y3 is a random number uniformly distributed in [0,1), n represents the size of the initial propagation network, and L(q) is the maximum connected component ratio when the occupied node ratio is q.

5. The method according to claim 1, wherein The step of dividing the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the optimal sequence specifically includes: When the initial propagation network forms a maximum connected component, determining the proportion of occupied nodes of the optimal sequence in the initial propagation network as a critical threshold, and dividing the initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to the critical threshold; The first occupied node set is determined by the following formula: The first removed node set is determined by the following formula: Among them, V o represents the first occupied node set, V u represents the first set of removed nodes, express The corresponding critical threshold, represents the optimal sequence, and n represents the size of the initial propagation network.

6. The method according to claim 1, wherein The removing of leaf nodes and central nodes from the unoccupied node set included in the coarse-grained network and adding the central node to the second removed node set specifically includes: Determine the first ψ third nodes with the highest degree centrality among the third nodes included in the coarse-grained network as central nodes; or Determine the new node formed by ψ second nodes at the end of the optimal sequence in the coarse-grained network as a central node, where ψ is a positive integer greater than zero; The second removed node set is an empty set before the central node is added.

7. A device for controlling false information on social networks, characterized in that: include: A determination unit, used to construct an initial propagation network and determine an initial sequence and an initial exchange threshold corresponding to the initial propagation network; Traversing each first node included in the initial sequence, obtaining and occupying the smallest candidate node corresponding to each first node, and when the connectivity component of the smallest candidate node is less than the initial exchange threshold, the smallest candidate node replaces the first node in the initial sequence to obtain a first sequence; determining the first sequence with the smallest network resilience among the multiple first sequences as the optimal sequence; A composition unit is configured to divide initial nodes included in the initial propagation network into a first occupied node set and a first removed node set according to an optimal sequence; define connected components included in the initial propagation network as new nodes, wherein the plurality of new nodes form a new occupied node set, and the first removed node set forms an unoccupied node set corresponding to the new occupied node set; and form a coarse-grained network based on edges between a first occupied node in the first occupied node set and a first removed node in the first removed node set, edges between the first removed nodes in the first removed node set, the newly occupied node set, and the unoccupied node set; a removal unit, configured to remove leaf nodes and central nodes from an unoccupied node set included in the coarse-grained network, and add the central node to a second removed node set; set a third node included in the coarse-grained network to an occupied state, and when the unoccupied node set is a non-empty set, iterate according to initial removal information of the third node, initial root information of the third node, and belief propagation; when the belief propagation reaches a self-consistent steady state, determine the removal probability of each third node according to the removal information of the third node after multiple iterations and the root information of the third node after multiple iterations, delete multiple third nodes with the highest probability from the coarse-grained network, and add them to the second removed node set; The obtaining unit is configured to place the newly occupied nodes included in the newly occupied set at the beginning of the removal sequence according to the sequence number, place the leaf nodes into the unoccupied node set, place the unoccupied nodes included in the unoccupied node set at the middle of the removal sequence according to the node sequence number, and place the removed nodes included in the second removed node set at the end of the removal sequence to form a removal sequence; and connect the optimal sequence and the removal sequence to obtain a final sequence.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the social network false information control method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to execute the method for controlling false information on a social network according to any one of claims 1 to 6.

Images