Information propagation snapshot-oriented rumor source positioning method, device and equipment and medium
By collecting network topology snapshots in social networks and extracting features, building neural network models for pre-training and training, the problem of difficult to locate false information dissemination sources in large-scale networks is solved, and the effectiveness of rumor source positioning is improved.
Patent Information
- Application Number
- CN202411630189.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-15
- Publication Date
- 2025-02-14
AI Technical Summary
It is difficult for the existing technology to effectively locate the source of dissemination of false information or rumors in large-scale networks, affecting the containment of their spread.
By collecting network topology snapshots of social networks, extracting snapshot subgraphs and adjacency matrices, performing singular value decomposition to obtain graph topology features and information propagation features, and fusing them into unified features, building neural network models for pre-training and training, realizing Positioning the source of rumors spread.
This reduces the computational complexity of the search space of the rumor source, improves the effectiveness of embedded features, and thus improves the effectiveness of the rumor source positioning.
Smart Images
Figure CN119444257A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of information dissemination technology, and in particular to a rumor source locating method, device, equipment and medium for information dissemination snapshots. Background Art
[0002] With the widespread popularity of mobile Internet and social applications, users can freely post and repost content of their interest on social networking platforms. However, this has also led to a problem that cannot be ignored: the rapid spread of false information and rumors on social networks. Such untrue content may not only cause harm to individuals, but may also cause serious consequences such as social panic. Therefore, tracing the source of false information or rumors on social networks is crucial to curbing their spread.
[0003] Information source location refers to finding the source of false information, biological viruses, etc. in graph-structured data such as social networks and personal relationship networks. In a dynamic and complex network environment, false information source location uses network topology, node status and other features as variables to find the source node of information dissemination among a large number of user nodes.
[0004] In static network propagation source positioning, there are mainly three methods: complete observation-based methods, snapshot observation-based methods, and sensor observation-based methods. For the positioning of real large-scale network information propagation sources, the existing methods are difficult to obtain the number of false information propagation source nodes, which affects the effectiveness of rumor source positioning. Summary of the invention
[0005] The purpose of this application is to provide a rumor source location method, device, equipment and medium for information dissemination snapshots to improve the effectiveness of rumor source location.
[0006] To achieve the above objectives, this application provides the following solutions:
[0007] In a first aspect, the present application provides a rumor source location method for information propagation snapshots, including:
[0008] The network topology snapshots generated by multiple social networks at different times are collected respectively; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state;
[0009] Generate a snapshot subgraph set including multiple snapshot subgraphs from each network topology snapshot, and determine the matrix of each snapshot subgraph; the matrix of each snapshot subgraph is a matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network;
[0010] Perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology features of each snapshot subgraph;
[0011] After embedding the node features of each snapshot subgraph, perform singular value decomposition to obtain the information propagation features of each snapshot subgraph;
[0012] The graph topology features and information propagation features of each snapshot subgraph are projected into the same dimension to obtain the fusion features of each snapshot subgraph;
[0013] The matrix and fusion features of each snapshot subgraph constitute a first training set, and the encoding part of the neural network is pre-trained using the first training set to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot subgraph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source positioning prediction result;
[0014] The pre-trained model is trained using a second training set to obtain a rumor propagation source positioning model; the second training set includes a plurality of samples, each sample includes a matrix of a snapshot subgraph and a rumor propagation source positioning label corresponding to the matrix of the snapshot subgraph;
[0015] The matrix of the snapshot subgraph for rumor source location is input into the rumor source location model to obtain the rumor source location prediction result.
[0016] Optionally, the matrix representation of each snapshot subgraph is:
[0017] in, is the matrix of the i-th snapshot subgraph, A is the adjacency matrix of the social network, I is the identity matrix, U i，n is a matrix indicating whether each node in the i-th snapshot subgraph is in an infected state. If the node is in a rumor infected state, the value is 1, and if the node is not infected, the value is 0.
[0018] Optionally, singular value decomposition is performed on the matrix of each snapshot subgraph to obtain graph topology features of each snapshot subgraph, specifically including:
[0019] The matrix of the snapshot subgraph is subjected to singular value decomposition. If the dimension of the matrix of the snapshot subgraph is smaller than the set dimension, the number of rows of the three matrices obtained by the singular value decomposition is filled to the set dimension by zero padding to obtain the graph topology feature of the snapshot subgraph.
[0020] Optionally, each snapshot subgraph is embedded with node features and then subjected to singular value decomposition to obtain information propagation features of each snapshot subgraph, specifically including:
[0021] Using formula Embed node features into snapshot subgraph;
[0022] After node feature embedding of the snapshot subgraph, singular value decomposition is performed. If the dimension of the matrix after node feature embedding of the snapshot subgraph is less than the set dimension, the number of rows of the three matrices obtained by the singular value decomposition is filled to the set dimension by zero padding to obtain the information propagation characteristics of the snapshot subgraph.
[0023] Among them, F i The matrix after node feature embedding for the i-th snapshot subgraph, MLP() is a multi-layer perceptron, embedding() is an embedding layer, is the i-th snapshot subgraph.
[0024] Optionally, the graph topology features and information propagation features of each snapshot subgraph are projected into the same dimension to obtain the fusion features of each snapshot subgraph, specifically including:
[0025] Using formula Project the graph topology features and information propagation features of the snapshot subgraph into the same dimension;
[0026] Among them, AF fusion is the fusion feature, LayerNorm() is the feature dimension normalization, U A ,∑ A and V A are the first orthogonal matrix, the first semi-positive definite matrix, and the second orthogonal matrix obtained by performing singular value decomposition on the matrix of the snapshot subgraph, respectively. F and∑ F are the third orthogonal matrix and the second semi-positive definite matrix obtained by performing singular value decomposition after embedding node features on the snapshot subgraph.
[0027] Optionally, the loss function used when pre-training the encoding part of the neural network using the first training set is:
[0028] Among them, KL loss is the contrastive divergence loss function value, KL div() is the contrastive divergence loss function, is the network topology feature of the i-th snapshot subgraph, AF fusion is the fusion feature, MLP() is the multi-layer perceptron, is the matrix of the i-th snapshot subgraph, and W is the weight matrix.
[0029] Optionally, the loss function when the pre-trained model is trained using the second training set is expressed as:
[0030]
[0031] Among them, L is the comprehensive loss value, L Entropy() is the cross entropy loss function, L MES () is the mean square error loss function, L b is the class-balanced loss function, KL loss is the contrast divergence loss function value, is the rumor source location prediction result, R is the true rumor source location, and λ is the loss coefficient.
[0032] In a second aspect, the present application provides a rumor source locating device for information propagation snapshots, the rumor source locating device for information propagation snapshots comprising:
[0033] The network topology snapshot collection module is used to collect network topology snapshots generated by multiple social networks at different times; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state;
[0034] A snapshot subgraph generation module, used to generate a snapshot subgraph set including a plurality of snapshot subgraphs from each network topology snapshot, and determine a matrix of each snapshot subgraph; the matrix of each snapshot subgraph is a matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network;
[0035] A graph topology feature determination module, used to perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology feature of each snapshot subgraph;
[0036] An information propagation feature determination module is used to embed node features of each snapshot subgraph and then perform singular value decomposition to obtain the information propagation features of each snapshot subgraph;
[0037] A fusion feature determination module is used to project the graph topology features and information propagation features of each snapshot subgraph into the same dimension to obtain the fusion features of each snapshot subgraph;
[0038] A pre-training module, which is used to form a first training set from the matrix and fusion features of each snapshot sub-graph, and use the first training set to pre-train the encoding part of the neural network to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot sub-graph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source positioning prediction result;
[0039] A second training module is used to train the pre-trained model using a second training set to obtain a rumor propagation source positioning model; the second training set includes a plurality of samples, each sample includes a matrix of a snapshot subgraph and a rumor propagation source positioning label corresponding to the matrix of the snapshot subgraph;
[0040] The rumor source positioning model application module is used to input the matrix of the snapshot subgraph to be used for rumor source positioning into the rumor source positioning model to obtain the rumor source positioning prediction result.
[0041] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-mentioned rumor source location methods for information propagation snapshots.
[0042] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned rumor source locating methods for information propagation snapshots.
[0043] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned rumor source locating methods for information propagation snapshots.
[0044] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0045] The present application provides a rumor source location method, apparatus, device and medium for information propagation snapshots, which reduces the problem of rumor source search space explosion by extracting snapshot subgraphs and adjacency matrices of network topology snapshots, and constructs a first training set consisting of matrices and fusion features of each snapshot subgraph. The encoding part of the neural network is pre-trained using the first training set to obtain a pre-trained model, thereby realizing topological structure feature comparative learning, improving the effectiveness of embedded features, and thus improving the effectiveness of rumor source location. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0047] Figure 1 A flowchart of a rumor source location method for information dissemination snapshots provided in an embodiment of the present application;
[0048] Figure 2 A schematic diagram of the principle of constructing a rumor propagation source positioning model provided in an embodiment of the present application;
[0049] Figure 3A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0051] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0052] This application provides a rumor source location method for information dissemination snapshots, such as Figure 1 As shown, the rumor source location method for information propagation snapshots includes:
[0053] Step 101: Collect network topology snapshots generated by multiple social networks at different times respectively; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state.
[0054] Step 102: Generate a snapshot subgraph set including multiple snapshot subgraphs for each network topology snapshot, and determine the matrix of each snapshot subgraph; the matrix of each snapshot subgraph is a matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network.
[0055] Step 103: Perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology feature of each snapshot subgraph.
[0056] Step 104: After embedding node features of each snapshot subgraph, perform singular value decomposition to obtain information propagation features of each snapshot subgraph.
[0057] Step 105: Project the graph topology features and information propagation features of each snapshot subgraph into the same dimension to obtain the fusion features of each snapshot subgraph.
[0058] Step 106: The matrix and fusion features of each snapshot subgraph constitute a first training set, and the encoding part of the neural network is pre-trained using the first training set to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot subgraph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source location prediction results.
[0059] Step 107: Train the pre-trained model with a second training set to obtain a rumor source location model; the second training set includes multiple samples, and each sample includes a matrix of a snapshot subgraph and a rumor source location label corresponding to the matrix of the snapshot subgraph.
[0060] Step 108: Input the matrix of the snapshot subgraph to be located for rumor source into the rumor source location model to obtain a rumor source location prediction result.
[0061] This application reduces the problem of the explosion of the rumor source search space by extracting the snapshot subgraph and the adjacency matrix of the network topology snapshot. The first training set is composed of the matrices of each snapshot subgraph and the fusion features. The encoding part of the neural network is pre-trained with the first training set to obtain a pre-trained model, realizing the contrastive learning of topological structure features, improving the effectiveness of the embedded features, and thus improving the effectiveness of rumor source location.
[0062] In a social network, the path of information dissemination can usually be modeled as a graph G(V, E), where V represents the set of nodes in the network, representing users in the social network, and E represents the set of edges between nodes, representing the connection relationships between users. In the early stage of rumor dissemination, since no operations such as rumor refutation are carried out, the dissemination mode of information diffusion in the social network conforms to the Linear Threshold (LT) model, the Independent Cascade (IC) model, or the Susceptible-Infective (SI) model.
[0063] In a social platform, there is generally one or more accounts that spread rumors or false information. To simulate the real information dissemination process, this application initializes K rumor sources in a given social network G(V, E), where 1 ≤ K < N, and N represents 10% of the total number of users (total number of nodes) in the social network.
[0064] Among them, for each social network in step 101, the number of rumor source nodes is initialized and K source nodes are randomly selected, that is, rumor dissemination nodes. After information spreads for a period of time, the nodes on the social network will show a rumor-infected state and an uninfected state, and then the snapshot s at time t t is saved, where s t ∈S(s 1 , s 2 ,..., s n ), 1 ≤ t < n, and n is the number of network topology snapshots. S() is the set of network topology snapshots.
[0065] As Figure 2 shown, in the snapshot s tThere are many nodes that are not infected. In order to reduce the computational complexity of the search space for locating the rumor source, this application generates a snapshot S(s 1 ,s 2 , ..., s n ) is processed into a snapshot subgraph S sub () is a snapshot subgraph set, is the i-th snapshot subgraph. Since the number of infected nodes in different snapshot subgraphs is different, it is necessary to process the adjacency matrix A of the social network G into the neighbor matrix corresponding to the snapshot subgraph. I is the identity matrix corresponding to the adjacency matrix A, as shown in formula (1); when the node i=n-1 is in an uninfected state, the node needs to be removed, and the identity matrix U i，n As shown in formula (2).
[0066] The matrix representation of each snapshot subgraph is:
[0067]
[0068] in, Represents the identity matrix with dimension (n-1)×n after removing the node, where (n-1) is the number of rows and n is the number of columns. is the matrix of the i-th snapshot subgraph. The matrix of the snapshot subgraph is extracted from the adjacency matrix of the social network and corresponds to the snapshot subgraph. A is the adjacency matrix of the social network, I is the identity matrix, and U i，n is the eigenvector matrix representing the adjacency matrix corresponding to the i-th snapshot subgraph, U i，n is obtained by eigendecomposition, where the eigendecomposition is performed on (A+I). The eigenvector matrix U i，n Each column of U corresponds to a eigenvector. i，n and are mutually orthogonal.
[0069] Among them, step 103 specifically includes: performing singular value decomposition (SVD) on the matrix of the snapshot subgraph. If the dimension of the matrix of the snapshot subgraph is smaller than the set dimension, the number of rows of the three matrices obtained by the singular value decomposition is filled to the set dimension by zero padding to obtain the graph topology features of the snapshot subgraph.
[0070] Step 103 is the graph topology feature SVD projection step. Since the sizes of the adjacency matrices of different snapshot subgraphs are different, in order to project the adjacency matrices of different dimensions into the same latent space, this application uses the output dimension d of the latent space as latent The condition is divided into two cases. latent To set the dimension.
[0071] The matrix of the snapshot subgraph is expressed by singular value decomposition:
[0072] when m<d latent or <d latent When U A ,∑ A , V A The number of rows of the matrix is filled to d latent The matrix of m is The number of rows, n is The number of columns, Figure 2 Middle U A The dimension is m×m, ∑ A The dimension of is m×n, and the dimension of VA is n×m.
[0073] Wherein, step 104 specifically includes: using formula Embed node features into snapshot subgraph.
[0074] Among them, F i The matrix after node feature embedding for the i-th snapshot subgraph, MLP() is a multi-layer perceptron, embedding() represents the embedding layer in the neural network, is the i-th snapshot subgraph.
[0075] Step 104 is the information propagation feature SVD projection step. In , since the node state feature dimensions of different snapshot subgraphs are different, node feature embedding is required before SVD projection.
[0076] The position encoding of node feature embedding is expressed as
[0077]
[0078] U F ,∑ F , V F =SVD(F i (p, q)) (6).
[0079] Among them, PE j is the position code of the jth node, F i =F i (p, q), w j is an intermediate parameter, k is a dimension index ranging from 0 to d / 2, and d is the total dimension of the position encoding.
[0080] After embedding node features of the snapshot subgraph, singular value decomposition is performed. If the dimension of the matrix after embedding node features of the snapshot subgraph is less than the set dimension, the number of rows of the three matrices obtained by singular value decomposition is filled to the set dimension by zero padding to obtain the information propagation characteristics of the snapshot subgraph. That is to say, when p <d latent or q <d latent When U F ,∑ F , V F The number of rows of the matrix is filled to d latent The matrix of F i The number of rows of F i The number of columns, Figure 2 Middle U F The dimension is m×m, ∑ F The dimension of V is m×n. F The dimension is n×n.
[0081] Wherein, step 105 specifically includes: using formula The graph topology characteristics and information propagation characteristics of the snapshot subgraph are projected into the same dimension.
[0082] Among them, AF fusion is the fusion feature, LayerNorm() is the feature dimension normalization, U A ,∑ A and V A are the first orthogonal matrix, the first semi-positive definite matrix, and the second orthogonal matrix obtained by performing singular value decomposition on the matrix of the snapshot subgraph, respectively. F ,∑ F and V F They are respectively the third orthogonal matrix, the second semi-positive definite matrix and the fourth orthogonal matrix obtained by performing singular value decomposition after embedding node features on the snapshot subgraph.
[0083] The model is used to generate network topology features and compare them with the topology of the snapshot subgraph, and the process is shown in formula (8) and formula (9). Existing methods require training from scratch for propagation data of different topological structures. For an indefinite number of propagation subgraphs, this method does not require separate training for specific topological structures. Even for propagation data of the same network topology, the adjacency matrix of different propagation subgraphs is different for the model. Therefore, before formally training the model, in order to better learn the topological features of the network, this application needs to pre-train the model using data from different data sets.
[0084]
[0085] Wherein, in step 106, the loss function used when pre-training the encoding part of the neural network using the first training set is:
[0086] Among them, KL loss is the contrastive divergence loss function value, KL div() is the contrastive divergence loss function, is the network topology feature of the i-th snapshot subgraph, AF fusion is the fusion feature, MLP() is the multi-layer perceptron, is the matrix of the i-th snapshot subgraph, and W is the learnable dynamic routing weight matrix.
[0087] The neural network of this application is a graph neural network, and the decoding part is Transformer. After the model is pre-trained, this application uses Transformer to fused feature AF fusion Decode and predict the source of rumor spread
[0088] In order to balance the imbalance between source nodes and non-source nodes, a class-balanced loss function is designed. Assuming that the label of the source node is 1 and the label of the non-source node is 0, the loss function Lb is:
[0089]
[0090] Among them, S represents the source node set, NS represents the non-source node set, and N s and N ns Represent the number of source nodes and non-source nodes respectively, i′ represents the probability that node i′ is predicted to be the source node, p j represents the probability that node j is predicted as a non-source node, and α is the imbalance coefficient used to adjust the loss weight between source nodes and non-source nodes.
[0091] Among them, in step 107, the loss function when the pre-trained model is trained using the second training set is expressed as:
[0092] Among them, L is the comprehensive loss value, L Entropy () is the cross entropy loss function, L MES () is the mean square error loss function, L b is the class-balanced loss function, KL loss is the contrast divergence loss function value, is the rumor source location prediction result, R is the true rumor source location, and λ is the loss coefficient.
[0093] Step 108 specifically includes: in actual application, firstly, a snapshot of information propagation is generated, then a snapshot subgraph is generated according to nodes in an infected state, and finally, the snapshot subgraph and the corresponding adjacency matrix are input into the rumor propagation source positioning model for prediction.
[0094] The technical effects of this application are as follows.
[0095] (1) Based on the social network topology diagram, this application uses the IC propagation model to simulate the early propagation process of rumors or false information, and obtains the infection snapshots of an uncertain number of rumor sources at each time step. By extracting the rumor propagation subgraph and adjacency matrix of the information propagation snapshot, the problem of explosion in the search space of the model to locate the rumor source is reduced;
[0096] (2) In order to solve the problem of feature embedding of snapshot subgraphs and adjacency matrices with an indefinite number of nodes, this application designs a graph topology feature SVD encoding module and an information propagation feature SVD encoding module, and introduces a topology structure feature comparison module to ensure the validity of the embedded features; through the SVD encoding module and the feature fusion module, the data is projected into the same latent space, and the Transformer model is introduced into the rumor source location task.
[0097] (3) According to existing research, the method of this application is the first to introduce the concept of an uncertain number of source nodes into the rumor source location task. Experiments have shown the effectiveness of the method of this application in the rumor source location task.
[0098] Based on the same inventive concept, the embodiment of the present application also provides a rumor source locating device for information dissemination snapshots for implementing the rumor source locating method for information dissemination snapshots involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the rumor source locating device for information dissemination snapshots provided below can refer to the limitations of the rumor source locating method for information dissemination snapshots above, and will not be repeated here.
[0099] In an exemplary embodiment, the present application provides a rumor source location device for information propagation snapshots, including:
[0100] The network topology snapshot collection module is used to collect network topology snapshots generated by multiple social networks at different times; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state.
[0101] The snapshot subgraph generation module is used to generate a snapshot subgraph set including multiple snapshot subgraphs from each network topology snapshot, and determine the matrix of each snapshot subgraph; the matrix of each snapshot subgraph is the matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network.
[0102] The graph topology feature determination module is used to perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology feature of each snapshot subgraph.
[0103] The information propagation feature determination module is used to embed the node features of each snapshot subgraph and then perform singular value decomposition to obtain the information propagation features of each snapshot subgraph.
[0104] The fusion feature determination module is used to project the graph topology features and information propagation features of each snapshot subgraph into the same dimension to obtain the fusion features of each snapshot subgraph.
[0105] A pre-training module is used to form a first training set from the matrix and fusion features of each snapshot sub-graph, and use the first training set to pre-train the encoding part of the neural network to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot sub-graph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source positioning prediction results.
[0106] The second training module is used to train the pre-trained model with a second training set to obtain a rumor propagation source positioning model; the second training set includes multiple samples, each sample includes a matrix of a snapshot subgraph and a rumor propagation source positioning label corresponding to the matrix of the snapshot subgraph.
[0107] The rumor source positioning model application module is used to input the matrix of the snapshot subgraph to be used for rumor source positioning into the rumor source positioning model to obtain the rumor source positioning prediction result.
[0108] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store rumor source location data for information propagation snapshots. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a rumor source location method for information propagation snapshots is implemented.
[0109] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned embodiments of the rumor source location method for information propagation snapshots are implemented.
[0110] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps in the above-mentioned rumor source location method embodiments for information propagation snapshots.
[0111] In an exemplary embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps in the above-mentioned rumor source location method embodiments for information propagation snapshots.
[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0113] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0114] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a data processing logic of a programmable logic device, etc., but is not limited thereto.
[0115] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0116] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A rumor source location method for information dissemination snapshots, characterized in that: The rumor source location method for information dissemination snapshots includes: The network topology snapshots generated by multiple social networks at different times are collected respectively; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state; Generate a snapshot subgraph set including multiple snapshot subgraphs from each network topology snapshot, and determine the matrix of each snapshot subgraph; the matrix of each snapshot subgraph is a matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network; Perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology features of each snapshot subgraph; After embedding the node features of each snapshot subgraph, perform singular value decomposition to obtain the information propagation features of each snapshot subgraph; The graph topology features and information propagation features of each snapshot subgraph are projected into the same dimension to obtain the fusion features of each snapshot subgraph; The matrix and fusion features of each snapshot subgraph constitute a first training set, and the encoding part of the neural network is pre-trained using the first training set to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot subgraph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source positioning prediction result; The pre-trained model is trained using a second training set to obtain a rumor propagation source positioning model; the second training set includes a plurality of samples, each sample includes a matrix of a snapshot subgraph and a rumor propagation source positioning label corresponding to the matrix of the snapshot subgraph; The matrix of the snapshot subgraph for rumor source location is input into the rumor source location model to obtain a rumor source location prediction result.
2. The rumor source location method for information dissemination snapshot according to claim 1 is characterized in that: The matrix representation of each snapshot subgraph is: in, is the matrix of the i-th snapshot subgraph, A is the adjacency matrix of the social network, I is the identity matrix, U i,n is a matrix indicating whether each node in the i-th snapshot subgraph is in an infected state. If the node is in a rumor infected state, the value is 1, and if the node is not infected, the value is 0.
3. The rumor source location method for information dissemination snapshot according to claim 1 is characterized in that: Perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology features of each snapshot subgraph, including: The matrix of the snapshot subgraph is subjected to singular value decomposition. If the dimension of the matrix of the snapshot subgraph is smaller than the set dimension, the number of rows of the three matrices obtained by the singular value decomposition is filled to the set dimension by zero padding to obtain the graph topology feature of the snapshot subgraph.
4. The rumor source location method for information dissemination snapshot according to claim 1 is characterized in that: After embedding the node features of each snapshot subgraph, singular value decomposition is performed to obtain the information propagation features of each snapshot subgraph, including: Using formula Embed node features into snapshot subgraph; After node feature embedding of the snapshot subgraph, singular value decomposition is performed. If the dimension of the matrix after node feature embedding of the snapshot subgraph is less than the set dimension, the number of rows of the three matrices obtained by the singular value decomposition is filled to the set dimension by zero padding to obtain the information propagation characteristics of the snapshot subgraph. Among them, F i The matrix after node feature embedding for the i-th snapshot subgraph, MLP() is a multi-layer perceptron, embedding() is an embedding layer, is the i-th snapshot subgraph.
5. The rumor source location method for information propagation snapshot according to claim 1 is characterized in that: The graph topology features and information propagation features of each snapshot subgraph are projected into the same dimension to obtain the fusion features of each snapshot subgraph, including: Using formula Project the graph topology features and information propagation features of the snapshot subgraph into the same dimension; Among them, AF fusion is the fusion feature, LayerNorm() is the feature dimension normalization, U A ,∑ A and V A are the first orthogonal matrix, the first semi-positive definite matrix, and the second orthogonal matrix obtained by performing singular value decomposition on the matrix of the snapshot subgraph, respectively. F and∑ F are the third orthogonal matrix and the second semi-positive definite matrix obtained by performing singular value decomposition after embedding node features on the snapshot subgraph.
6. The rumor source location method for information propagation snapshot according to claim 1 is characterized in that: The loss function used when pre-training the encoding part of the neural network using the first training set is: Among them, KL loss is the contrastive divergence loss function value, KL div() is the contrastive divergence loss function, is the network topology feature of the i-th snapshot subgraph, AF fusion is the fusion feature, MLP() is the multi-layer perceptron, is the matrix of the i-th snapshot subgraph, and W is the weight matrix.
7. The rumor source location method for information dissemination snapshot according to claim 1 is characterized in that: The loss function when training the pre-trained model using the second training set is expressed as: Among them, L is the comprehensive loss value, L Entropy () is the cross entropy loss function, L MES () is the mean square error loss function, L b is the class-balanced loss function, KL loss is the contrast divergence loss function value, is the rumor source location prediction result, R is the true rumor source location, and λ is the loss coefficient.
8. A rumor source location device for information dissemination snapshot, characterized in that: The rumor source locating device for information dissemination snapshots includes: The network topology snapshot collection module is used to collect network topology snapshots generated by multiple social networks at different times; each social network includes a node set and an edge set, and the edge set is a collection of connection relationships between nodes; the number of rumor propagation sources in each social network is greater than or equal to 1; the state of each node in the network topology snapshot is a rumor infection state or an uninfected state; A snapshot subgraph generation module, used to generate a snapshot subgraph set including a plurality of snapshot subgraphs from each network topology snapshot, and determine a matrix of each snapshot subgraph; the matrix of each snapshot subgraph is a matrix corresponding to the snapshot subgraph in the adjacency matrix of the social network; A graph topology feature determination module, used to perform singular value decomposition on the matrix of each snapshot subgraph to obtain the graph topology feature of each snapshot subgraph; An information propagation feature determination module is used to embed node features of each snapshot subgraph and then perform singular value decomposition to obtain the information propagation features of each snapshot subgraph; A fusion feature determination module is used to project the graph topology features and information propagation features of each snapshot subgraph into the same dimension to obtain the fusion features of each snapshot subgraph; A pre-training module, which is used to form a first training set from the matrix and fusion features of each snapshot sub-graph, and use the first training set to pre-train the encoding part of the neural network to obtain a pre-trained model; the neural network includes an encoding part and a decoding part, the encoding part is used to input the matrix of the snapshot sub-graph and output the fusion features, and the decoding part is used to input the fusion features and output the rumor propagation source positioning prediction result; A second training module is used to train the pre-trained model using a second training set to obtain a rumor propagation source positioning model; the second training set includes a plurality of samples, each sample includes a matrix of a snapshot subgraph and a rumor propagation source positioning label corresponding to the matrix of the snapshot subgraph; The rumor source positioning model application module is used to input the matrix of the snapshot subgraph to be used for rumor source positioning into the rumor source positioning model to obtain the rumor source positioning prediction result.
9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the rumor source location method for information propagation snapshots as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the rumor source location method for information propagation snapshot described in any one of claims 1 to 7 is implemented.