Method, apparatus and device for false information detection based on propagation reconstruction

Abstract

This disclosure relates to a method, apparatus, and device for detecting misinformation based on propagation reconstruction. The method includes: acquiring propagation data of blog posts in a social network; constructing an information propagation graph based on the propagation data; aggregating the node neighborhood features in the information propagation graph using a deep graph convolutional network to obtain a node representation V of the information propagation graph; estimating potential propagation interactions between nodes in the information propagation graph based on the node representation V to generate multiple potential propagation graphs; aggregating the node neighborhood features of each potential propagation graph using a deep graph convolutional network, and then combining the updated node representations of all updated potential propagation graphs to obtain a reconstructed node representation Z; performing task classification based on the node representation V and the reconstructed node representation Z to obtain a misinformation detection result for the blog post. This disclosure can improve the performance of misinformation detection tasks.

Description

Technical Field

[0001] This invention relates to the field of data mining technology, and specifically to a method, apparatus, and device for detecting false information based on propagation reconstruction. Background Technology

[0002] In the era of new media, social media websites have provided users with tremendous convenience in obtaining information, expressing opinions, and exchanging ideas. More and more users are enthusiastic about participating in discussions of trending topics on social media, and due to the platform's convenience, user-generated content can quickly reach a wide audience. However, social media has also bred a large amount of misinformation, posing a threat to social stability and seriously affecting people's daily lives and the healthy development of society. Therefore, how to automatically detect misinformation is an urgent research topic in the field of social network analysis, with significant research and practical application implications.

[0003] Over time, source blog posts develop a specific propagation structure through a series of forwarding or replying relationships. Existing research has found that, compared to genuine information, misinformation exhibits a faster, broader, and deeper propagation structure. These structural characteristics provide researchers with the possibility of detecting misinformation using propagation data. With the development of deep learning, data mining, and graph learning technologies, many methods model the propagation data of source blog posts as tree or graph structures, utilize deep learning tools to learn the feature representations of source blog posts, and construct classifiers for detection.

[0004] However, existing detection methods using shallow graph convolution can only aggregate local propagation dependencies, failing to adequately model deeper and broader propagation structures. Furthermore, in the era of big data, to protect the privacy and security of internet users, researchers find it difficult and costly to obtain complete information propagation data, and are unable to identify abnormal data such as false or irrelevant propagation, leading to significant uncertainty in network structure. These incomplete and unreliable propagation structures limit the model's learning ability, preventing it from focusing on key propagation structure features conducive to false information detection. Summary of the Invention

[0005] This invention addresses two main technical problems: first, how to fully model long-distance dependencies in deeper and broader propagation structures; and second, how to extract higher-order structural features from incomplete and unreliable propagation structures. This invention provides a method, apparatus, and device for detecting misinformation based on propagation reconstruction, thereby improving the performance of misinformation detection tasks.

[0006] The specific technical solution of the present invention is as follows:

[0007] According to a first aspect of the present disclosure, the present invention provides a method for detecting false information based on propagation reconstruction, the method comprising the following steps:

[0008] Obtain the propagation data of blog posts in social networks; wherein, the propagation data includes: the text content of the source blog post and subsequent propagated blog posts, the set of propagation relationships between the source blog post node and the subsequent propagated blog post node, and the set of propagation relationships between the propagated blog post nodes;

[0009] Based on the aforementioned propagation data, an information propagation graph is constructed;

[0010] The node representation V of the information propagation graph is obtained by aggregating the node neighborhood features in the information propagation graph using a depth graph convolutional network.

[0011] Based on the node representation V, estimate the potential propagation interactions between nodes in the information propagation graph to generate multiple potential propagation graphs;

[0012] After aggregating the node neighborhood features of each potential propagation graph using a depth graph convolutional network, the reconstructed node representation Z is obtained by combining the updated node representations of all updated potential propagation graphs.

[0013] Based on the node representation V and the reconstructed node representation Z, a task classification is performed to obtain the false information detection result of the blog post.

[0014] Furthermore, constructing the information propagation graph based on the propagation data includes:

[0015] The text features of the text content are used as the initialization features of the nodes in the information propagation graph;

[0016] and,

[0017] Based on the set of propagation relationships between the source blog post node and subsequent propagation blog post nodes, and the set of propagation relationships between the propagation blog post nodes, the adjacency matrix in the information propagation graph is obtained.

[0018] Furthermore, the node representation V, obtained by aggregating the node neighborhood features in the information propagation graph using a depth graph convolutional network, includes:

[0019] K stacked graph convolutional layers; where each graph convolutional layer introduces an initial residual and an identity mapping;

[0020] Based on the K-layer graph convolutional layer, the node neighborhood features in the information propagation graph are aggregated to obtain node representations. When the k-th graph convolutional layer performs aggregation, the updated node representation... σ(·) represents the activation function, αk Denotes the first hyperparameter, β k This represents the second hyperparameter. Represents the weight matrix. D denotes a diagonal matrix, and I denotes the identity matrix. This represents the node representation obtained by aggregating the source blog post nodes through K rounds of iterations. This represents the node representation obtained by aggregating the Nth propagation node through K rounds.

[0021] Further, the step of estimating the potential propagation interactions between nodes in the information propagation graph based on the node representation V to generate multiple potential propagation graphs includes:

[0022] Calculate the node representation Gaussian distributions from various modeling perspectives in, This represents the mean value from the m-th modeling perspective. Let M represent the variance of the m-th modeling perspective; where 1≤m≤M, and M represents the total number of modeling perspectives.

[0023] From the distribution After sampling, the distributed representation of node i in the m-th modeling perspective is obtained. Where ∈ represents a standard normal distribution with a mean of 0 and a variance of 1, ∈ ∈ N(0, I);

[0024] Integrating various distributed representations Obtain the node initialization feature Q from the m-th modeling perspective. m ;

[0025] For any modeling perspective m, calculate the potential propagation interactions between any two nodes.

[0026] Based on the potential propagation interactions Generate the adjacency matrix S from the m-th modeling perspective. m ;

[0027] Based on the node initialization feature Q m With the adjacency matrix S m Generate the potential propagation graph from the m-th modeling perspective.

[0028] Furthermore, after aggregating the node neighborhood features of each potential propagation graph using a depth graph convolutional network, the reconstructed node representation Z is obtained by integrating the node representations of all updated potential propagation graphs, including:

[0029] By using a two-layer graph convolutional network to aggregate the node neighborhood features of each node in the latent propagation graph, the node representation of the updated latent propagation graph is obtained. in, The adjacency matrix S represents m The regularized form, This represents the weights of the first layer of the graph convolutional network. This represents the weights of the second-layer graph convolutional network;

[0030] Calculate the reconstructed node representation Z = W z [U 1 ;…;U M ]+b z Among them, W z b is the first trainable parameter. z This represents the second trainable parameter.

[0031] Furthermore, the step of performing task classification based on the node representation V and the reconstructed node representation Z to obtain the false information detection result of the blog post includes:

[0032] Given the node representation V and the reconstructed node representation Z, the graph feature representation O = meanpooling([V; Z]) is computed using an average pooling layer;

[0033] A classification task is performed on the graph feature representation O to obtain the false information detection results of the blog post.

[0034] According to a second aspect of the present disclosure, the present invention provides a misinformation detection device based on propagation reconstruction, the device comprising:

[0035] The data acquisition module is used to acquire the propagation data of blog posts in social networks; wherein, the propagation data includes: the text content of the source blog post and subsequent propagated blog posts, the set of propagation relationships between the source blog post node and the subsequent propagated blog post node, and the set of propagation relationships between the propagated blog post nodes;

[0036] A propagation graph construction module is used to construct an information propagation graph based on the propagation data;

[0037] The first graph convolutional network module is used to aggregate the node neighborhood features in the information propagation graph using a depth graph convolutional network to obtain the node representation V of the information propagation graph;

[0038] The Gaussian propagation reconstruction module is used to estimate the potential propagation interactions between nodes in the information propagation graph based on the node representation V, so as to generate multiple potential propagation graphs;

[0039] The second graph convolutional network module is used to aggregate the node neighborhood features of each potential propagation graph using the depth graph convolutional network, and then synthesize the updated node representations of all updated potential propagation graphs to obtain the reconstructed node representation Z.

[0040] The false information classification module is used to perform task classification based on the node representation V and the reconstructed node representation Z to obtain the false information detection result of the blog post.

[0041] According to a third aspect of the present disclosure, the present invention provides an electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement any of the above-described propagation-reconstruction-based false information detection methods.

[0042] According to a fourth aspect of the present disclosure, the present invention provides a computer-readable storage medium having stored thereon computer program instructions, characterized in that, when executed by a processor, the program instructions implement any of the above-described propagation-reconstruction-based false information detection methods.

[0043] Compared with the prior art, the technical solution provided by the present invention has at least the following beneficial effects:

[0044] The method provided by this invention can better learn incomplete propagation trees, explore more complex and potential propagation dependencies in the propagation process, and mine more valuable propagation structure features for detection, thereby effectively improving the detection accuracy of false information. This invention can effectively model long-distance propagation dependencies in propagation trees, and has more accurate detection performance, especially for tree structures that propagate deeper and wider.

[0045] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0046] Figure 1 The flowchart illustrates the implementation of the propagation-reconstruction-based false information detection method provided by this invention.

[0047] Figure 2 The structural diagram of the false information detection system based on propagation reconstruction provided by the present invention. Detailed Implementation

[0048] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the described embodiments are only intended to facilitate the understanding of the present invention and do not constitute any limitation thereof.

[0049] The method for detecting misinformation based on propagation reconstruction provided by this invention first constructs an information propagation graph using the text of the source blog post and some propagation data in a social network. Then, it captures long-distance propagation dependencies in the information propagation graph through a deep graph convolutional network. Based on the updated node representation, it generates multiple potential propagation graphs by estimating the potential propagation interactions between nodes. Finally, it obtains the misinformation detection result based on the updated node representations of the information propagation graph and the potential propagation graph.

[0050] Specifically, such as Figure 1 As shown in the implementation flowchart, the present invention includes the following steps:

[0051] Figure 1 This invention provides a flowchart of a method for detecting misinformation based on propagation reconstruction, which specifically includes the following steps:

[0052] Step A: Obtain the text and partial propagation data of the source post from the Twitter social network. Specifically, define the propagation data of the sample as G =<V,E> Where, V={r,x1,...,x n-1} represents the set of blog post information during the dissemination process, where r is the source blog post, and x1,...,x n-1 For subsequent blog posts; E = {e st |s,t=0,...,n-1} represents the set of propagation relations.

[0053] Step B: Extract text features based on the text data and construct a propagation graph structure using the propagation data. Specifically,

[0054] Step B1: For each sample, extract the source blog post r and the propagated blog posts x1,…,x n-1 TF-IDF text features are defined as follows: in, For the text features of the source blog post r, To disseminate blog posts, d0 represents the text feature dimension.

[0055] Step B2: Based on the information itself in social media and the dissemination behavior (such as comments and reposts) associated with that information, construct an undirected information propagation graph. The adjacency matrix corresponding to the graph structure is defined as follows: The initial value is defined as:

[0056]

[0057] The initial feature representation of all blog post nodes in the information propagation graph is constructed based on text features, denoted as X. TD =X BU =X.

[0058] Step C: Based on the propagation graph constructed in Step B, deep graph convolution is used to fully model the long-distance dependencies in the original propagation graph and extract the high-order structural features of the original propagation. To alleviate the oversmoothing problem of graph modeling, initial residuals and identity mappings are introduced. Multiple layers of graph convolution are stacked to aggregate the node neighborhood features in the original propagation graph, learning the feature representations of the nodes and fully modeling the long-distance dependencies between information propagation. The specific update method for nodes is as follows:

[0059]

[0060] in, This is the regularized graph Laplacian matrix. D is a diagonal matrix, I is the identity matrix, and α... k and β k These are two hyperparameters. It is the weight matrix, σ(·) is the activation function, and I n Represents the identity matrix.

[0061] By stacking K layers of graph convolution, the node representations in the original propagation graph are obtained.

[0062] Step D: After modeling the original propagation in Step C, the Gaussian propagation reconstruction module is used to define the potential propagation between nodes in the original propagation graph from multiple perspectives. This example, based on the node representation obtained in Step C, designs a Gaussian propagation reconstruction module to model the implicit propagation in the actual information propagation process from multiple angles, estimating the potential propagation interactions between pairs of nodes; specifically:

[0063] Step D1: Based on the node features in the original propagation graph, the Gaussian propagation reconstruction module models it as a distributed representation, which can more accurately and comprehensively reflect potential propagation dependencies. The distributed representation calculation method for nodes is as follows:

[0064]

[0065]

[0066] Where M represents the number of modeling perspectives, g θ and g′ θ Let be two trainable neural networks, where φ is a non-linear activation function. The mean, The variance describes the multi-dimensional uncertainty of this propagation node. From the distribution... After sampling, a distributed representation of the nodes is obtained:

[0067]

[0068] Where ∈ represents a standard normal distribution with a mean of 0 and a variance of 1, and I is the identity matrix.

[0069] Step D2: Obtain node representations based on step D1. This yields a new node representation. Specifically, for any modeling viewpoint m, the corresponding node representation is denoted as Q. m .

[0070] Based on the distribution used in step D1 For any modeling perspective m, calculate the potential propagation interactions between any two nodes:

[0071]

[0072] Based on the above calculations, the potential propagation graphs after M reconstructions can be obtained, with the corresponding node representations and adjacency matrices being {Q}. 1 ,…,Q M} and {S 1 ,…,S M}

[0073] Step E: Based on the multiple potential propagation graphs reconstructed in Step D, design a graph convolutional network based on root node enhancement to learn the node representations of the reconstructed graphs; specifically,

[0074] For each reconstructed latent propagation graph, given node representations and adjacency matrices, a two-layer graph convolutional network is used to further aggregate the node neighborhood features in the latent propagation. The message propagation formula is defined as follows:

[0075]

[0076] in, This is the regularized form of the adjacency matrix S. and It is a trainable parameter matrix.

[0077] Finally, considering all M reconstructed potential propagation graphs, the calculation for each node is as follows:

[0078] Z = W z [U 1 ;…;U M ]+b z

[0079] Among them, W z and b z These are trainable parameters.

[0080] Step F: Based on the multiple sets of node representations obtained in steps C and E, the final feature representation of the source post is calculated using an average pooling layer, and then input into the classifier to complete the fake information detection. Specifically:

[0081] Step F1: Given the node representations learned in the original propagation and potential propagation graphs, the graph feature representation is calculated using an average pooling layer. The specific calculation method is as follows:

[0082] O = meanpooling([V; Z])

[0083] Here, meanpooling(·) is the average pooling function.

[0084] Step F2: The fake information detection task is essentially a classification task. Based on the final feature representation of the sample obtained in Step F1, calculate the probability of the fake information category label for that sample, i.e.:

[0085]

[0086] Among them, W c ,b c These are trainable parameters.

[0087] In summary, the present invention has the following technical effects:

[0088] I. The method provided by this invention can explore more complex and potential propagation dependencies in the propagation process and uncover more valuable propagation structural features for detection;

[0089] Second, this invention can effectively improve the accuracy of detecting fake information, achieving an improvement of 5.3% and 1.9% in accuracy compared to existing methods on publicly available fake news datasets in the political and entertainment categories, respectively.

[0090] Third, this invention can effectively model long-distance propagation dependencies in propagation trees, and has more accurate detection performance, especially for tree structures with deeper and wider propagation. In public datasets, it can improve the detection accuracy by about 50% for tree structures with a propagation depth greater than 6.

[0091] like Figure 2 As shown, the present invention also provides a false information detection device based on propagation reconstruction. The system includes: a data acquisition module 100, an information propagation graph construction module 200, a first graph convolutional network module 300, a Gaussian propagation reconstruction module 400, a second graph convolutional network module 500, and a false information classification module 600.

[0092] The data acquisition module 100 is used to acquire the propagation data of blog posts in social networks; wherein, the propagation data includes: the text content of the source blog post and subsequent propagation blog posts, the set of propagation relationships between the source blog post node and the subsequent propagation blog post node, and the set of propagation relationships between the propagation blog post nodes;

[0093] The propagation graph construction module 200 is used to construct an information propagation graph based on the propagation data;

[0094] The first graph convolutional network module 300 is used to aggregate the node neighborhood features in the information propagation graph using a depth graph convolutional network to obtain the node representation V of the information propagation graph;

[0095] The Gaussian propagation reconstruction module 400 is used to estimate the potential propagation interactions between nodes in the information propagation graph based on the node representation V, so as to generate multiple potential propagation graphs;

[0096] The second graph convolutional network module 500 is used to aggregate the node neighborhood features of each potential propagation graph using the depth graph convolutional network, and then synthesize the updated node representations of all updated potential propagation graphs to obtain the reconstructed node representation Z.

[0097] The false information classification module 600 is used to perform task classification based on the node representation V and the reconstructed node representation Z to obtain the false information detection result of the blog post.

[0098] The propagation-reconstruction-based fake information detection device provided in this disclosure can implement all the processes implemented in the above-described propagation-reconstruction-based fake information detection method embodiments. To avoid repetition, these processes will not be described again here.

[0099] The present invention also provides an electronic device. This electronic device may be a computer device, a laptop computer, a server, or other types of electronic equipment.

[0100] An electronic device may include at least one processor and a memory. The processor can execute instructions stored in the memory. The processor is communicatively connected to the memory via a data bus. In addition to the memory, the processor may also be communicatively connected to input devices, output devices, and communication devices via the data bus.

[0101] The processor can be any conventional processor. Processors may include central processing units (CPUs), graphics processing units (GPUs), field-programmable gate arrays (FPGAs), systems on chips (SoCs), application-specific integrated circuits (ASICs), or combinations thereof.

[0102] Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0103] In this embodiment of the disclosure, executable instructions are stored in the memory. The processor can read the executable instructions from the memory and execute the instructions to implement all or part of the steps of the above-described propagation-based reconstruction-based false information detection method.

[0104] In addition to the methods and apparatus described above, exemplary embodiments of this disclosure also include a computer program product or a computer-readable storage medium storing the computer program product. The computer product includes computer program instructions that can be executed by a processor to perform all or part of the steps described in the exemplary embodiments above.

[0105] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. Programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages, and scripting languages ​​(e.g., Python). The program code can be executed entirely on a user's computing device, partially on a user's device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0106] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media include: static random access memory (SRAM) having one or more electrically connected wires, electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk, or any suitable combination thereof.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting misinformation based on propagation reconstruction, characterized in that, The method includes: Obtain the propagation data of blog posts in social networks; wherein, the propagation data includes: the text content of the source blog post and subsequent propagated blog posts, the set of propagation relationships between the source blog post node and the subsequent propagated blog post node, and the set of propagation relationships between the propagated blog post nodes; Based on the aforementioned propagation data, an information propagation graph is constructed; The node representation V of the information propagation graph is obtained by aggregating the node neighborhood features in the information propagation graph using a depth graph convolutional network. Based on the node representation V, estimate the potential propagation interactions between nodes in the information propagation graph to generate multiple potential propagation graphs; After aggregating the node neighborhood features of each potential propagation graph using a depth graph convolutional network, the reconstructed node representation Z is obtained by combining the updated node representations of all updated potential propagation graphs. Based on the node representation V and the reconstructed node representation Z, a task classification is performed to obtain the false information detection result of the blog post; Specifically, the node representation V is obtained by aggregating the node neighborhood features in the information propagation graph based on a depth graph convolutional network, including: K stacked graph convolutional layers; where each graph convolutional layer introduces an initial residual and an identity mapping; Based on the K-layer graph convolutional layer, the node neighborhood features in the information propagation graph are aggregated to obtain node representations. When the k-th graph convolutional layer performs aggregation, the updated node representation... σ(·) represents the activation function, α k Denotes the first hyperparameter, β k This represents the second hyperparameter. Represents the weight matrix. D denotes a diagonal matrix, and I denotes the identity matrix. This represents the node representation obtained by aggregating the source blog post nodes through K rounds of iterations. This represents the node representation obtained by aggregating the Nth propagation node through K rounds.

2. The method for detecting false information based on propagation reconstruction as described in claim 1, characterized in that, The construction of the information propagation graph based on the propagation data includes: The text features of the text content are used as the initialization features of the nodes in the information propagation graph; and, Based on the set of propagation relationships between the source blog post node and subsequent propagation blog post nodes, and the set of propagation relationships between the propagation blog post nodes, the adjacency matrix in the information propagation graph is obtained.

3. The method for detecting false information based on propagation reconstruction as described in claim 1, characterized in that, The step of estimating the potential propagation interactions between nodes in the information propagation graph based on the node representation V, to generate multiple potential propagation graphs, includes: Calculate the node representation Gaussian distributions from various modeling perspectives in, This represents the mean value from the m-th modeling perspective. Let M represent the variance of the m-th modeling perspective; where 1≤m≤M, and M represents the total number of modeling perspectives. From the distribution After sampling, the distributed representation of node i in the m-th modeling perspective is obtained. Where, 𝜖 represents a standard normal distribution with a mean of 0 and a variance of 1, 𝜖∈N(0,I); Integrating various distributed representations Obtain the node initialization feature Q from the m-th modeling perspective. m ; For any modeling perspective m, calculate the potential propagation interactions between any two nodes. Based on the potential propagation interactions Generate the adjacency matrix S from the m-th modeling perspective. m ; Based on the node initialization feature Q m With the adjacency matrix S m Generate the potential propagation graph from the m-th modeling perspective.

4. The method for detecting false information based on propagation reconstruction as described in claim 3, characterized in that, After aggregating the node neighborhood features of each potential propagation graph using a depth graph convolutional network, the reconstructed node representation Z is obtained by integrating the node representations of all updated potential propagation graphs, including: By using a two-layer graph convolutional network to aggregate the node neighborhood features of each node in the latent propagation graph, the node representation of the updated latent propagation graph is obtained. in, The adjacency matrix S represents m The regularized form, This represents the weights of the first layer of the graph convolutional network. This represents the weights of the second-layer graph convolutional network; Calculate the reconstructed node representation Z = W z [U 1 ;…;U M ]+b z Among them, W z b is the first trainable parameter. z This represents the second trainable parameter.

5. The method for detecting false information based on propagation reconstruction as described in claim 1, characterized in that, The process of performing task classification based on the node representation V and the reconstructed node representation Z to obtain the false information detection results of the blog post includes: Given the node representation V and the reconstructed node representation Z, the graph feature representation O = meanpooling([V; Z]) is computed using an average pooling layer; A classification task is performed on the graph feature representation O to obtain the false information detection results of the blog post.

6. A misinformation detection device based on propagation reconstruction, characterized in that, The device includes: The data acquisition module is used to acquire the propagation data of blog posts in social networks; wherein, the propagation data includes: the text content of the source blog post and subsequent propagated blog posts, the set of propagation relationships between the source blog post node and the subsequent propagated blog post node, and the set of propagation relationships between the propagated blog post nodes; A propagation graph construction module is used to construct an information propagation graph based on the propagation data; The first graph convolutional network module is used to aggregate the node neighborhood features in the information propagation graph using a depth graph convolutional network to obtain the node representation V of the information propagation graph; The Gaussian propagation reconstruction module is used to estimate the potential propagation interactions between nodes in the information propagation graph based on the node representation V, so as to generate multiple potential propagation graphs; The second graph convolutional network module is used to aggregate the node neighborhood features of each potential propagation graph using the depth graph convolutional network, and then synthesize the updated node representations of all updated potential propagation graphs to obtain the reconstructed node representation Z. The false information classification module is used to perform task classification based on the node representation V and the reconstructed node representation Z to obtain the false information detection result of the blog post; Specifically, the node representation V is obtained by aggregating the node neighborhood features in the information propagation graph based on a depth graph convolutional network, including: K stacked graph convolutional layers; where each graph convolutional layer introduces an initial residual and an identity mapping; Based on the K-layer graph convolutional layer, the node neighborhood features in the information propagation graph are aggregated to obtain node representations. When the k-th graph convolutional layer performs aggregation, the updated node representation... σ(·) represents the activation function, α k Denotes the first hyperparameter, β k This represents the second hyperparameter. Represents the weight matrix. D denotes a diagonal matrix, and I denotes the identity matrix. This represents the node representation obtained by aggregating the source blog post nodes through K rounds of iterations. This represents the node representation obtained by aggregating the Nth propagation node through K rounds.

7. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the propagation-reconstruction-based false information detection method according to any one of claims 1-5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the propagation-reconstruction-based false information detection method as described in any one of claims 1-5.

Images