Dynamic feature extraction method and system of social media account, electronic equipment and storage medium

Abstract

The invention discloses a dynamic feature extraction method and system for a social media account, electronic equipment and a storage medium, and the method comprises the steps: obtaining a text and a picture issued by a target user account, and coding the text and the picture into image-text dynamic data; extracting a multi-modal theme feature vector from the image-text dynamic data; constructing a graph neural network by taking the accounts as nodes and the connection between the accounts as edges, and constructing a dynamic feature extraction model in combination with a long-short term memory network; and inputting the multi-modal theme feature vector into a trained dynamic feature extraction model, and outputting a dynamic feature vector of the target user account. According to the method, the dynamic feature extraction model is constructed based on the graph neural network and the long-short term memory network, dynamic features can be extracted at the time level and the content level at the same time, meanwhile, the multi-modal data processing technology is combined, deep semantic information and dynamic changes of images and texts in a post published by a target user account can be captured, and the dynamic feature extraction efficiency is improved. And the accuracy and generalization of feature expression are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of text processing technology, and in particular to a method and system for extracting dynamic features of a social media account, an electronic device, and a computer-readable storage medium. Background Art

[0002] Social media account feature extraction refers to the use of specific technical means and algorithms to identify, analyze and construct features that can reflect user behavior patterns, interest preferences, social interactions and other characteristics from user activities and published content on social media in real time or regularly, so as to facilitate further analysis of social media accounts, such as account clustering, account behavior and speech consistency analysis, account opinion and position analysis, account circle affiliation identification analysis, etc.

[0003] The current social media account feature extraction method usually extracts features from text data of a single modality, such as using keyword matching, simple statistical analysis, or pre-trained semantic analysis and other technical means to extract features. However, posts published by existing social media platforms (such as WeChat, Facebook, Twitter, Instagram, etc.) not only contain text data, but also include a large amount of visual modality data, such as pictures, videos, etc. Single modality features cannot accurately reflect the three-dimensional features of the account, and thus cannot accurately reflect the social characteristics of the account. For example, patent CN119005199A discloses a social network forwarding behavior prediction method based on large model enhancement, which applies a large language model to the forwarding behavior prediction task, and injects user-related ID sequence information into the large language model, integrates the text content of social media and user-related ID sequence information, etc., and uses the powerful semantic understanding ability and extensive pre-trained knowledge of the large language model to align it with the prediction task through fine-tuning to help obtain better prediction results. This patent is to extract text features through a pre-trained semantic analysis model. In addition, existing social media account feature extraction methods usually perform static processing and analysis on the full data of a single account, while ignoring the extraction of dynamic features, and are unable to capture the dynamic changes of users in real time. Moreover, existing social media account feature extraction methods ignore the complexity and diversity of user behavior, as well as the relational representation of potential relationships between accounts, and are unable to accurately represent the three-dimensional features of accounts. Summary of the invention

[0004] The present invention provides a method and system for extracting dynamic features of a social media account, an electronic device, and a computer-readable storage medium, which can capture the deep semantic information and dynamic changes of images and texts in posts published by a target user account, comprehensively reflect the three-dimensional features of the target user account, accurately reflect the social characteristics of the target user account, and greatly improve the accuracy and generalization of feature expression.

[0005] According to one aspect of the present invention, a method for extracting dynamic features of a social media account is provided, comprising the following contents:

[0006] Obtain the text and pictures posted by the target user account, and encode the text and pictures into dynamic text and picture data;

[0007] Extract multimodal topic feature vectors from dynamic image and text data;

[0008] A graph neural network is constructed with accounts as nodes and connections between accounts as edges, and a dynamic feature extraction model is constructed in combination with a long short-term memory network.

[0009] The multimodal topic feature vector is input into the trained dynamic feature extraction model, and the dynamic feature vector of the target user account is output.

[0010] Furthermore, the process of encoding text and pictures into graphic and text dynamic data includes the following contents:

[0011] Use the pre-trained CLIP large model to encode text and pictures. The encoding process is as follows:

[0012]

[0013] in, represents an image-text pair, T and I represent text and image respectively, V represents a multimodal vector, and n represents the number of image-text pairs.

[0014] Furthermore, the multimodal topic feature vector is extracted from the graphic and text dynamic data using the following formula:

[0015]

[0016] in, Represents the multimodal topic feature vector of the target user account at time T, represents average pooling processing, Represents the nth multimodal vector of the target user account at time T.

[0017] Furthermore, the dynamic feature extraction model outputs the dynamic feature vector of the target user account using the following formula:

[0018]

[0019] in, represents the dynamic feature vector of the target user account at time T, represents a composite network combining graph neural network and long short-term memory network, Represents the multimodal topic feature vector of the target user account at time T, express The hidden vector obtained by the network at time (T-1).

[0020] Furthermore, the following contents are also included:

[0021] Obtain the user's behavioral dynamics data on the social media account, and input the behavioral dynamics data together with the multimodal topic feature vector into the trained dynamic feature extraction model to output the dynamic feature fusion vector of the target user account.

[0022] Furthermore, the dynamic feature extraction model outputs the dynamic feature fusion vector of the target user account using the following formula:

[0023]

[0024] in, represents the dynamic feature fusion vector of the target user account at time T, represents the dynamic feature vector of the target user account at time T, AVG() represents the average calculation, represents the maximum pooling process, Represents the dynamic feature vector list of the S neighboring nodes of the target user account node at time T.

[0025] Furthermore, behavioral dynamic data includes the number of likes, comments, reposts, and shares between accounts.

[0026] In addition, the present invention also provides a dynamic feature extraction system for social media accounts, comprising:

[0027] The image and text dynamic data encoding module is used to obtain the text and pictures posted by the target user account and encode the text and pictures into image and text dynamic data;

[0028] A multimodal topic feature vector extraction module is used to extract multimodal topic feature vectors from dynamic image and text data;

[0029] A dynamic feature extraction model building module is used to build a graph neural network with accounts as nodes and connections between accounts as edges, and to build a dynamic feature extraction model in combination with a long short-term memory network;

[0030] The dynamic feature vector extraction module is used to input the multimodal vector list into the trained dynamic feature extraction model and output the dynamic feature vector of the target user account.

[0031] In addition, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the above method by calling the computer program stored in the memory.

[0032] In addition, the present invention also provides a computer-readable storage medium for storing a computer program for extracting dynamic features from social media accounts, wherein the computer program executes the steps of the method described above when running on a computer.

[0033] The present invention has the following beneficial effects:

[0034] The dynamic feature extraction method of the social media account of the present invention, after obtaining the text and pictures posted by the target user account, encodes them into graphic and text dynamic data, extracts the multimodal theme feature vector from it, and then establishes a graph neural network with the account as the node and the connection between the accounts as the edge, and constructs a dynamic feature extraction model in combination with the long short-term memory network, and then inputs the multimodal theme feature vector into the trained dynamic feature extraction model, and the dynamic feature vector of the target user account can be output. The present invention constructs a dynamic feature extraction model based on the graph neural network and the long short-term memory network, and can extract dynamic features at the time level and the content level at the same time. At the same time, it combines the multimodal data processing technology, and can capture the deep semantic information and dynamic changes of the pictures and texts in the posts posted by the target user account, and comprehensively reflects the three-dimensional features of the target user account, and can accurately reflect the social characteristics of the target user account, greatly improving the accuracy and generalization of feature expression, and can be well applied to different social media platforms.

[0035] In addition, the dynamic feature extraction system of social media accounts of the present invention also has the above advantages.

[0036] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0038] Figure 1 It is a flowchart of a method for extracting dynamic features of a social media account according to a preferred embodiment of the present application;

[0039] Figure 2 is another flow chart of the method for extracting dynamic features of a social media account according to a preferred embodiment of the present application;

[0040] Figure 3 It is a schematic diagram of the principle of the dynamic feature extraction model outputting the dynamic feature fusion vector in the preferred embodiment of the present application;

[0041] Figure 4 It is a module diagram of a dynamic feature extraction system for a social media account according to another embodiment of the present application. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0043] Reference Figure 1 As shown, the preferred embodiment of the present application provides a method for extracting dynamic features of a social media account, comprising the following contents:

[0044] Step S1: Obtain the text and pictures posted by the target user account, and encode the text and pictures into dynamic text and picture data;

[0045] It can be understood that in step S1, a data interface is established through an API interface or data crawling technology of a social media platform (such as WeChat, Weibo, Facebook, Twitter, Instagram, etc.), the text and pictures in the post published by the target user account are obtained, and then the text and pictures are encoded into dynamic text and picture data. The process of encoding text and pictures into dynamic text and picture data includes the following:

[0046] The pre-trained CLIP model (Contrastive Language-Image Pretraining) is used to encode text and images. The specific encoding process is as follows:

[0047]

[0048] in, represents an image-text pair, T and I represent text and image respectively, V represents a multimodal vector, n represents the number of image-text pairs, and each image-text pair After input into the CLIP model, it will be encoded into a multimodal vector V. It should be noted that when the input image-text pair contains only text or pictures, the missing part is set to empty.

[0049] It can be understood that encoding the input text and images into dynamic text and image data through the pre-trained CLIP large model is conducive to capturing the deep semantic information and dynamic changes of the text and images in the posts published by the target user account. Compared with the existing static feature extraction based on a single modality of text data, it can better reflect the three-dimensional characteristics of the target user account.

[0050] Step S2: extracting multimodal topic feature vectors from the graphic and text dynamic data;

[0051] It can be understood that in step S2, the multimodal topic feature vector is extracted using the following formula:

[0052]

[0053] in, Represents the multimodal topic feature vector of the target user account at time T, represents average pooling processing, Represents the nth multimodal vector of the target user account at time T.

[0054] It can be understood that the multimodal vector list obtained by encoding in step S1 Perform average pooling, that is, take the average value of the corresponding dimensions in all multimodal vectors, and you can get a multimodal topic feature vector , which can capture the deep semantic information of images and texts in posts published by the target user account.

[0055] Step S3: construct a graph neural network with accounts as nodes and connections between accounts as edges, and construct a dynamic feature extraction model in combination with a long short-term memory network;

[0056] It can be understood that in step S3, a GNN network (Graph Neural Network) is constructed with accounts as nodes and connections between accounts as edges, and a dynamic feature extraction model is constructed in combination with an LSTM network (Long Short-Term Memory). The dynamic feature extraction model is a combination of a graph neural network and a long short-term memory network, which can be represented as CNN-LSTM. Among them, the CNN-LSTM network belongs to an existing neural network, and its specific network structure is not described here. In addition, after the dynamic feature extraction model is constructed, it is necessary to construct a training sample set to train the model until the model converges. The specific model training process belongs to the existing technology and is not described here.

[0057] It can be understood that the present invention adopts GNN network and LSTM network to construct a dynamic feature extraction model, which has excellent dynamic feature extraction performance both at the time level and the content level.

[0058] Step S4: Input the multimodal topic feature vector into the trained dynamic feature extraction model, and output the dynamic feature vector of the target user account.

[0059] It can be understood that in step S4, the multimodal topic feature vector extracted in step S2 is input into the trained dynamic feature extraction model to output the dynamic feature vector of the target user account. The dynamic feature extraction model outputs the dynamic feature vector of the target user account using the following formula:

[0060]

[0061] in, represents the dynamic feature vector of the target user account at time T, represents a composite network combining graph neural network and long short-term memory network, Represents the multimodal topic feature vector of the target user account at time T, express The hidden vector obtained by the network at time (T-1). It can be seen that the update of the dynamic feature vector is obtained by the multimodal topic feature vector extracted at this moment and the hidden vector at the previous moment, so that dynamic features can be extracted in both the time dimension and the content dimension.

[0062] It can be understood that the dynamic feature extraction method of the social media account of this embodiment, after obtaining the text and pictures posted by the target user account, encodes them into graphic and text dynamic data, extracts the multimodal theme feature vector from them, and then establishes a graph neural network with the account as the node and the connection between the accounts as the edge, and combines the long short-term memory network to build a dynamic feature extraction model, and then inputs the multimodal theme feature vector into the trained dynamic feature extraction model, and the dynamic feature vector of the target user account can be output. The present invention constructs a dynamic feature extraction model based on the graph neural network and the long short-term memory network, which can extract dynamic features at the time level and the content level at the same time, and combines the multimodal data processing technology at the same time, which can capture the deep semantic information and dynamic changes of the pictures and texts in the posts posted by the target user account, and fully reflects the three-dimensional features of the target user account, which can accurately reflect the social characteristics of the target user account, greatly improve the accuracy and generalization of feature expression, and can be well applied to different social media platforms.

[0063] As a preferred example, Figure 2 As shown, the dynamic feature extraction method of the social media account also includes the following contents:

[0064] Step S5: Obtain the user's behavioral dynamic data on the social media account, and input the behavioral dynamic data together with the multimodal topic feature vector into the trained dynamic feature extraction model to output the dynamic feature fusion vector of the target user account.

[0065] The dynamic feature extraction model outputs the dynamic feature fusion vector of the target user account using the following formula:

[0066]

[0067] in, represents the dynamic feature fusion vector of the target user account at time T, represents the dynamic feature vector of the target user account at time T, AVG() represents the average calculation, represents the maximum pooling process, Represents the dynamic feature vector list of the S neighboring nodes of the target user account node at time T.

[0068] Specifically, Figure 3 As shown in FIG. 1 , the user's behavior dynamic data on the social media account is obtained, including the number of likes, comments, reposts, and shares between accounts, which can accurately reflect the interactive behavior between accounts. Then, the behavior dynamic data and the multimodal topic feature vector extracted in step S2 are simultaneously used as the input of the dynamic feature extraction model. Due to the introduction of the behavior data input model, the behavior data of the target user account at time T has changed compared to the behavior data at time T-1, which makes the edges between nodes in the CNN-LSTM network change, so that dynamic features can be extracted more accurately. In addition, the hidden vector V at this time h T-1 It actually contains four parts, namely the neighbor node vector of the node at time T-1, the adjacent edge vector, the LSTM cell vector, and the LSTM hidden state vector.

[0069] It can be understood that the present invention also takes into account the complexity and diversity of user behaviors, as well as the relational representation of potential relationships between accounts, and introduces behavioral dynamic data into the dynamic feature extraction model. Since the behavioral dynamic data of the target user account at time T has changed compared to the behavioral dynamic data at time T-1, the edges between nodes in the CNN-LSTM network will also change, which can accurately capture various complex behavioral pattern information, thereby more accurately extracting dynamic features.

[0070] It can be understood that after the dynamic features are extracted on the basis of the present invention, they can be used for classification and identification in many scenarios, for example, user classification: users are divided into different categories, such as individuals, media, government or organizations, which helps to understand the attributes and behavior patterns of user groups; spam account detection: identifying and filtering out accounts that generate spam or engage in inappropriate behavior, which helps to maintain the health of the social media environment and the quality of user experience; personalized recommendation: based on the user's behavioral characteristics and preferences, recommending content, products or services of interest to users to improve user satisfaction and user stickiness of the platform; social influence analysis: analyzing the structure and behavioral communication patterns of users in social networks, predicting the user's influence and possible behavioral trends; public opinion monitoring: extracting users' views and emotions on specific topics or events, helping governments or enterprises to grasp public opinion in a timely manner and take corresponding response measures; security monitoring: monitoring abnormal account behavior, such as fraud, cyber attacks, etc., to ensure the user's network security and the stable operation of the platform; account topic identification: identifying the areas or topics that the account mainly focuses on, which helps content creators and advertisers to accurately locate the target audience; crisis response: when a public crisis event occurs, quickly identify the intensity and emotional tendency of related discussions so that relevant agencies can respond and take measures quickly.

[0071] In addition, if Figure 4 As shown, another embodiment of the present invention further provides a system for extracting dynamic features of a social media account, preferably using the above-mentioned method for extracting dynamic features of a social media account, including:

[0072] The image and text dynamic data encoding module is used to obtain the text and pictures posted by the target user account and encode the text and pictures into image and text dynamic data;

[0073] A multimodal topic feature vector extraction module is used to extract multimodal topic feature vectors from dynamic image and text data;

[0074] A dynamic feature extraction model building module is used to build a graph neural network with accounts as nodes and connections between accounts as edges, and to build a dynamic feature extraction model in combination with a long short-term memory network;

[0075] The dynamic feature vector extraction module is used to input the multimodal vector list into the trained dynamic feature extraction model and output the dynamic feature vector of the target user account.

[0076] It can be understood that the dynamic feature extraction system of the social media account of this embodiment, after obtaining the text and pictures posted by the target user account, encodes them into graphic and text dynamic data, extracts the multimodal theme feature vector from them, and then establishes a graph neural network with the account as the node and the connection between the accounts as the edge, and combines the long short-term memory network to build a dynamic feature extraction model, and then inputs the multimodal theme feature vector into the trained dynamic feature extraction model, and the dynamic feature vector of the target user account can be output. The present invention constructs a dynamic feature extraction model based on the graph neural network and the long short-term memory network, which can extract dynamic features at the time level and the content level at the same time, and combines the multimodal data processing technology at the same time, which can capture the deep semantic information and dynamic changes of the pictures and texts in the posts posted by the target user account, and fully reflects the three-dimensional features of the target user account, which can accurately reflect the social characteristics of the target user account, greatly improve the accuracy and generalization of feature expression, and can be well applied to different social media platforms.

[0077] In addition, the social media account dynamic feature extraction system also includes:

[0078] The behavior dynamics data collection module is used to obtain the user's behavior dynamics data on the social media account, and input the behavior dynamics data together with the multimodal topic feature vector into the trained dynamic feature extraction model.

[0079] It can be understood that after the behavior dynamics data collection module inputs the collected behavior dynamics data and the extracted multimodal topic feature vector into the trained dynamic feature extraction model, the dynamic feature extraction model construction module can output the dynamic feature fusion vector of the target user account.

[0080] It can be understood that the various modules of the system embodiment correspond to the various steps of the above method embodiment, so the specific working principles of each module are not repeated here, and the corresponding references can be made to the various steps of the above method embodiment.

[0081] In addition, another embodiment of the present invention further provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the above method by calling the computer program stored in the memory.

[0082] In addition, another embodiment of the present invention further provides a computer-readable storage medium for storing a computer program for extracting dynamic features from social media accounts, wherein the computer program executes the steps of the method described above when running on a computer.

[0083] Common computer-readable storage media include: floppy disks, flexible disks, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, any other optical media, punch cards, paper tapes, any other physical media with patterns of holes, random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), flash-erasable programmable read-only memory (FLASH-EPROM), any other memory chip or cartridge, or any other medium that can be read by a computer. Instructions can further be transmitted or received by a transmission medium. The term transmission medium may include any tangible or intangible medium that can be used to store, encode or carry instructions for execution by a machine, and includes digital or analog communication signals or intangible media that facilitate communication of the above instructions. Transmission media include coaxial cables, copper wire, and optical fiber, including the wires of a bus used to transmit a computer data signal.

[0084] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0085] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0086] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0088] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0089] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting dynamic features of social media accounts, characterized in that: Includes the following: Obtain the text and pictures posted by the target user account, and encode the text and pictures into dynamic text and picture data; Extract multimodal topic feature vectors from dynamic image and text data; A graph neural network is constructed with accounts as nodes and connections between accounts as edges, and a dynamic feature extraction model is constructed in combination with a long short-term memory network. The multimodal topic feature vector is input into the trained dynamic feature extraction model, and the dynamic feature vector of the target user account is output.

2. The method for extracting dynamic features of a social media account according to claim 1, wherein: The process of encoding text and pictures into dynamic graphic data includes the following: Use the pre-trained CLIP large model to encode text and pictures. The encoding process is as follows: in, represents an image-text pair, T and I represent text and image respectively, V represents a multimodal vector, and n represents the number of image-text pairs.

3. The method for extracting dynamic features of social media accounts according to claim 1, characterized in that: The multimodal topic feature vector is extracted from the graphic and text dynamic data using the following formula: in, Represents the multimodal topic feature vector of the target user account at time T, represents average pooling processing, Represents the nth multimodal vector of the target user account at time T.

4. The method for extracting dynamic features of social media accounts according to claim 1, characterized in that: The dynamic feature extraction model outputs the dynamic feature vector of the target user account using the following formula: in, represents the dynamic feature vector of the target user account at time T, represents a composite network combining graph neural network and long short-term memory network, Represents the multimodal topic feature vector of the target user account at time T, express The hidden vector obtained by the network at time (T-1).

5. The method for extracting dynamic features of social media accounts according to claim 1, characterized in that: Also included are the following: Obtain the user's behavioral dynamics data on the social media account, and input the behavioral dynamics data together with the multimodal topic feature vector into the trained dynamic feature extraction model to output the dynamic feature fusion vector of the target user account.

6. The method for extracting dynamic features of social media accounts according to claim 5, characterized in that: The dynamic feature extraction model outputs the dynamic feature fusion vector of the target user account using the following formula: in, represents the dynamic feature fusion vector of the target user account at time T, represents the dynamic feature vector of the target user account at time T, AVG() represents the average calculation, represents the maximum pooling process, Represents the dynamic feature vector list of the S neighboring nodes of the target user account node at time T.

7. The method for extracting dynamic features of a social media account according to claim 5, wherein: Behavioral dynamics data includes the number of likes, comments, reposts, and shares between accounts.

8. A dynamic feature extraction system for social media accounts, characterized in that: include: The image and text dynamic data encoding module is used to obtain the text and pictures posted by the target user account and encode the text and pictures into image and text dynamic data; A multimodal topic feature vector extraction module is used to extract multimodal topic feature vectors from dynamic image and text data; A dynamic feature extraction model building module is used to build a graph neural network with accounts as nodes and connections between accounts as edges, and to build a dynamic feature extraction model in combination with a long short-term memory network; The dynamic feature vector extraction module is used to input the multimodal vector list into the trained dynamic feature extraction model and output the dynamic feature vector of the target user account.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method according to any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium for storing a computer program for extracting dynamic features from social media accounts, characterized in that: When the computer program is run on a computer, the steps of the method according to any one of claims 1 to 7 are executed.

Images