Network rumor detection method based on spatial-temporal feature learning of double-dynamic graph convolutional network

Abstract

The invention provides a network rumor detection method based on spatial-temporal feature learning of a double-dynamic graph convolutional network. The method comprises the following steps: converting an event into a preset number of propagation structures according to time; forming a user interaction graph and an information propagation graph through the propagation structure; respectively learning the user interaction graph and the information propagation graph by using a convolutional network to obtain graph features; splicing the graph feature of the previous layer and the root node feature vector of the next layer to obtain an enhanced graph feature; obtaining an output feature of the time sequence fusion unit through the time sequence fusion unit; further obtaining the final feature of the event based on the output feature of the time sequence fusion unit; and finally, obtaining a classification result through a multi-layer perceptron. According to the method, spatial and temporal features of information propagation and user interaction can be comprehensively captured, content features, propagation features and user features can be effectively fused, rumor detection precision is improved, and the method especially has remarkable advantages in an early detection stage.

Description

Technical Field

[0001] The present invention relates to the fields of computer science and artificial intelligence, and in particular to a network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning. Background Art

[0002] The rapid development of social networks has brought about changes in the way information is disseminated, but this has also led to the problem of the spread of online rumors. Therefore, researching and developing online rumor detection technology, especially automated rumor identification based on social network data, has become an important research topic.

[0003] Most existing rumor detection methods focus on text content analysis, using techniques such as sentiment analysis and keyword extraction to identify rumors from message content. However, with the increasing ability of rumor spreaders to disguise information, it is difficult to effectively identify rumors simply by relying on content features. In recent years, rumor detection methods based on propagation features have gradually become a research hotspot. Graph Neural Networks (GNNs), as an emerging graph structure data processing method, can effectively learn the propagation structure features in the network and have achieved good results in social network rumor detection.

[0004] Existing GNN-based rumor detection methods mainly focus on the spatial structural features of graphs and capture the path of information propagation by learning the relationship between nodes and edges. However, most of these methods ignore the temporal structural features of information propagation and fail to fully consider the temporal evolution and dynamic changes of information during the propagation process. In addition, although some studies have attempted to combine spatiotemporal features and use the propagation temporal structure for rumor detection, these methods often ignore the spatiotemporal characteristics of user interactions and fail to fully explore the impact of user interaction patterns on rumor propagation. Summary of the invention

[0005] In view of the above situation, the main purpose of the present invention is to propose a network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning to solve the above technical problems.

[0006] The present invention proposes a network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning, and the method comprises the following steps: Step 1: Divide the event into a preset number of time series states according to time, and then transform each time series state into a communication structure by combining the forwarding and commenting relationships; Step 2: Obtain the user's social characteristics, and combine the user's social characteristics and communication structure to form a user interaction graph; Obtaining original posts and response information, and combining the obtained original posts, response information, and propagation structure to form an information propagation graph; Step 3: Use two different convolutional networks to learn the user interaction graph and the information propagation graph respectively, and obtain the user interaction graph features and the information propagation graph features respectively; Step 4: For the user interaction graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain enhanced user interaction graph features; For the information propagation graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain the enhanced information propagation graph features; Step 5: construct a temporal fusion unit based on the cross attention mechanism, input the enhanced user interaction graph features and the enhanced user interaction graph features as input sequences into the temporal fusion unit, and obtain the output features of the temporal fusion unit; Step 6: Perform global average pooling on the output features of the time series fusion unit to obtain output features after global average pooling; Based on the initial features in the information propagation graph, semantic features are obtained by combining the pre-trained preview model; The output features after global average pooling are concatenated with the semantic features to obtain the final features of the event; Step 7: Input the final features of the event into the multi-layer perceptron, and then process it through the activation function to obtain the classification result.

[0007] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a dual dynamic graph convolutional network, the present invention can comprehensively capture the spatiotemporal characteristics of information dissemination and user interaction, and effectively integrate text, dissemination and user characteristics, significantly improving the detection accuracy of online rumors; 2. The present invention introduces a cross-attention mechanism to perform time series fusion, which enhances the model's ability to perceive time series changes and is particularly suitable for early detection of rumors; 3. The present invention improves the robustness of the model through a dual dynamic graph convolutional network, can cope with complex social network data, and improves the generalization ability and stability of rumor detection, thereby effectively solving the problems of insufficient multimodal feature fusion and difficulty in capturing time series changes in existing methods.

[0008] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description or learned through embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a flow chart of the network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning proposed by the present invention; Figure 2 This is a model architecture diagram of the network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning proposed by the present invention; Figure 3 This is a cross-attention calculation flow chart of the network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning proposed in the present invention. DETAILED DESCRIPTION

[0010] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations on the present invention. Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations on the present invention.

[0011] These and other aspects of the embodiments of the present invention will be apparent with reference to the following description and accompanying drawings. In these descriptions and accompanying drawings, some specific implementations of the embodiments of the present invention are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present invention, but it should be understood that the scope of the embodiments of the present invention is not limited thereto.

[0012] See also Figure 1 , Figure 2 The embodiment of the present invention proposes a network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning, and the method comprises the following steps: Step 1: Divide the event into a preset number of time series states according to time, and then transform each time series state into a communication structure by combining the forwarding and commenting relationships; Specifically, in this step, the publishing time of the last response information of the event and the publishing time of the original post are recorded, and based on the two time statistical time spans, the time span is evenly divided into a certain number of time intervals; then the time is divided into multiple time series states according to the time intervals, and each time series state is converted into a propagation structure according to the forwarding and commenting relationship.

[0013] Step 2: Obtain the user's social characteristics, and combine the user's social characteristics and communication structure to form a user interaction graph; Obtaining original posts and response information, and combining the obtained original posts, response information, and propagation structure to form an information propagation graph; Specifically, in this step, both the user interaction graph and the information propagation graph are composed of a node set, an edge set and an adjacency matrix.

[0014] Step 3: Use two different convolutional networks to learn the user interaction graph and the information propagation graph respectively, and obtain the user interaction graph features and the information propagation graph features respectively; In step 3, two different convolutional networks are used to learn the user interaction graph and the information propagation graph respectively, and the user interaction graph features and the information propagation graph features are obtained respectively. The relationship between the corresponding processes is: ; in, Indicates Layer user interaction graph features, Indicates Layer information propagation graph features, It means that it has been processed by the activation function. represents the normalized symmetric adjacency matrix, Indicates Layer user interaction graph features, Indicates Layer information propagation graph features, , , and Both indicate The trainable parameters in the layer convolutional network, Indicates an event, Indicates the timing status; The definition of the normalized symmetric adjacency matrix is: ; in, represents the adjacency matrix, represents the identity matrix, represents the degree matrix, Represents the matrix resulting from the addition of the adjacency matrix and the identity matrix.

[0015] Step 4: For the user interaction graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain enhanced user interaction graph features; For the information propagation graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain the enhanced information propagation graph features; In step 4, for the user interaction graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain the enhanced user interaction graph features. The relationship between the corresponding process is: ; in, Indicates The user interaction graph features after layer enhancement, Represents a splicing operation, Indicates The root node feature vector of the layer user interaction graph; For the information propagation graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain the enhanced information propagation graph features. The relationship between the corresponding process is: ; in, Indicates The information propagation graph features after layer enhancement, Indicates The root node feature vector of the layer information propagation graph; Specifically, in this step, enhanced user interaction graph features and enhanced information propagation graph features are obtained to fully highlight the role of user information and semantic information of the original post in rumor detection.

[0016] Step 5: construct a temporal fusion unit based on the cross attention mechanism, input the enhanced user interaction graph features and the enhanced user interaction graph features as input sequences into the temporal fusion unit, and obtain the output features of the temporal fusion unit; See also Figure 3 In step 5, a temporal fusion unit is constructed based on the cross attention mechanism. The definition of the temporal fusion unit is: ; in, It means that after cross attention processing, Represents the input sequence Query, Represents the input sequence The transposition of the keywords, Represents the input sequence The value of Represents the characteristic dimension of keywords. and Represent two different input sequences respectively; The enhanced user interaction graph features and the enhanced user interaction graph features are input into the time series fusion unit as the input sequence to obtain the output features of the time series fusion unit. The relationship between the corresponding process is: ; in, Indicates The output features of the timing fusion unit at the moment, Indicates The output features of the timing fusion unit at the moment, Indicates The enhanced user interaction graph features at each moment, Indicates Enhanced information diffusion graph features at each moment.

[0017] Specifically, this step fuses the two dynamic spatiotemporal features through a temporal fusion unit, and inputs the fusion result into the graph convolutional network and the temporal fusion unit at the next moment to realize the temporal interaction of the two dynamic graphs, further enhancing the model's perception of temporal changes.

[0018] Further, in Figure 3 middle, and Both represent the sequence length, , , and Both represent feature dimensions, , and Both represent learning parameters.

[0019] Step 6: Perform global average pooling on the output features of the time series fusion unit to obtain output features after global average pooling; Based on the initial features in the information propagation graph, semantic features are obtained by combining the pre-trained preview model; The output features after global average pooling are concatenated with the semantic features to obtain the final features of the event; In step 6, the output features of the time series fusion unit are subjected to global average pooling to obtain the output features after global average pooling. The relationship between the corresponding process is: ; in, represents the output feature after global average pooling, represents the average pooling operation; The output features after global average pooling are concatenated with the semantic features to obtain the final features of the event. The relationship between the corresponding process is: ; in, Represents the ultimate characteristic of time, Represents semantic features.

[0020] Step 7: Input the final features of the event into the multi-layer perceptron, and then process it through the activation function to obtain the classification result; In step 7, the final features of the event are input into the multi-layer perceptron, and then the classification result is obtained through activation function processing. The relationship between the corresponding process is: ; in, Represents the classification result, It means processed by multi-layer perceptron. It means that it has been processed by the activation function. represents the parameters of the linear layer, Represents the bias term of the linear layer.

[0021] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0022] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0023] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning, characterized in that: The method comprises the following steps: Step 1: Divide the event into a preset number of time series states according to time, and then transform each time series state into a communication structure by combining the forwarding and commenting relationships; Step 2: Obtain the user's social characteristics, and combine the user's social characteristics and communication structure to form a user interaction graph; Obtaining original posts and response information, and combining the obtained original posts, response information, and propagation structure to form an information propagation graph; Step 3: Use two different convolutional networks to learn the user interaction graph and the information propagation graph respectively, and obtain the user interaction graph features and the information propagation graph features respectively; Step 4: For the user interaction graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain enhanced user interaction graph features; For the information propagation graph features, the features of the previous layer are concatenated with the root node feature vectors of the next layer to obtain the enhanced information propagation graph features; Step 5: construct a temporal fusion unit based on the cross attention mechanism, input the enhanced user interaction graph features and the enhanced user interaction graph features as input sequences into the temporal fusion unit, and obtain the output features of the temporal fusion unit; Step 6: Perform global average pooling on the output features of the time series fusion unit to obtain output features after global average pooling; Based on the initial features in the information propagation graph, semantic features are obtained by combining the pre-trained preview model; The output features after global average pooling are concatenated with the semantic features to obtain the final features of the event; Step 7: Input the final features of the event into the multi-layer perceptron, and then process it through the activation function to obtain the classification result.

2. According to claim 1, the network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning is characterized in that: In step 3, two different convolutional networks are used to learn the user interaction graph and the information propagation graph respectively, and the user interaction graph features and the information propagation graph features are obtained respectively. The relationship between the corresponding processes is: ； in, Indicates Layer user interaction graph features, Indicates Layer information propagation graph features, It means that it has been processed by the activation function. represents the normalized symmetric adjacency matrix, Indicates Layer user interaction graph features, Indicates Layer information propagation graph features, , , and Both indicate The trainable parameters in the layer convolutional network, Indicates an event, Indicates the timing status.

3. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 2 is characterized in that: In the process of using two different convolutional networks to learn the user interaction graph and the information propagation graph respectively and obtain the features of the user interaction graph and the information propagation graph respectively, the definition of the normalized symmetric adjacency matrix is: ； in, represents the adjacency matrix, represents the identity matrix, represents the degree matrix, Represents the matrix resulting from the addition of the adjacency matrix and the identity matrix.

4. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 3 is characterized in that: In step 4, for the user interaction graph feature, the feature of the previous layer is concatenated with the root node feature vector of the next layer to obtain the enhanced user interaction graph feature. The relationship between the corresponding process is: ； in, Indicates The user interaction graph features after layer enhancement, Represents a splicing operation, Indicates The root node feature vector of the layer user interaction graph.

5. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 4 is characterized in that: In step 4, for the information propagation graph feature, the feature of the previous layer is concatenated with the root node feature vector of the next layer to obtain the enhanced information propagation graph feature. The relationship between the corresponding process is: ； in, Indicates The information propagation graph features after layer enhancement, Indicates The root node feature vector of the layer information propagation graph.

6. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 5 is characterized in that: In step 5, a time series fusion unit is constructed based on the cross attention mechanism, and the definition of the time series fusion unit is: ; in, It means that after cross attention processing, Represents the input sequence Query, Represents the input sequence The transposition of the keywords, Represents the input sequence The value of Represents the characteristic dimension of keywords. and Represent two different input sequences respectively.

7. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 6 is characterized in that: In step 5, the enhanced user interaction graph features and the enhanced user interaction graph features are input into the time series fusion unit as input sequences to obtain output features of the time series fusion unit. The corresponding process has the following relationship: ； in, Indicates The output features of the timing fusion unit at the moment, Indicates The output features of the timing fusion unit at the moment, Indicates The enhanced user interaction graph features at each moment, Indicates Enhanced information diffusion graph features at each moment.

8. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 7 is characterized in that: In step 6, the output features of the time series fusion unit are subjected to global average pooling processing to obtain the output features after global average pooling. The relationship between the corresponding process is: ； in, represents the output features after global average pooling, Represents an average pooling operation.

9. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 8 is characterized in that: In step 6, the output features after global average pooling are concatenated with the semantic features to obtain the final features of the event. The relationship between the corresponding process is: ； in, Represents the ultimate characteristic of time, Represents semantic features.

10. The network rumor detection method based on dual dynamic graph convolutional network spatiotemporal feature learning according to claim 9 is characterized in that: In step 7, the final features of the event are input into the multi-layer perceptron, and then the classification result is obtained through activation function processing. The relationship between the corresponding process is: ； in, Represents the classification result, It means processed by multi-layer perceptron. It means that it has been processed by the activation function. represents the parameters of the linear layer, Represents the bias term of the linear layer.

Images