Public opinion processing method and device, computer equipment and storage medium

Abstract

The invention discloses a public opinion processing method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the steps of obtaining a message text of a preset message group, and performing text local feature extraction on the message text to obtain a message local feature; obtaining a group member social relation graph and a group message time relation graph of the message group; inputting the group member social relation graph and the group message time relation graph into a pre-trained graph attention network to obtain global context features of the message text; carrying out feature fusion on the local features and the global context features of the messages to obtain fusion features; and performing time sequence analysis on the fused features to obtain a public opinion trend prediction result. The invention further relates to the technical field of block chains, and the message text is stored on the block chain node. According to the method, the workload and the time cost of operation personnel are greatly reduced, meanwhile, the monitoring and intervention efficiency of negative contents is improved, and potential public opinion risks can be quickly found and predicted.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a public opinion processing method, apparatus, computer equipment and storage medium. Background Art

[0002] Public opinion analysis is the process of conducting in-depth thinking and analysis of public opinion on a specific issue based on the needs of the issue, thereby obtaining relevant conclusions. Public opinion analysis involves objective, systematic, and quantitative analysis of information content, aiming to reveal the essential facts and trends in the information and provide a scientific basis for decision makers.

[0003] At present, some property and casualty insurance companies have opened up new operating channels based on messaging platforms, namely messaging community operations. According to the purchase stage of the customer, the customer is classified and added to different messaging groups (such as customer maintenance group, first day group, conversion group, etc.) for management. This is an innovative model of insurance operation. Property and casualty insurance companies arrange operators to manage messaging group operations. Operators can interact with customers in the messaging group, share insurance knowledge, release insurance product information, achieve more accurate personalized services, and respond to needs in a timely manner.

[0004] However, while daily message group operations bring opportunities, they also bring some challenges. For example, when customers send negative content in the group, if they cannot be intervened in time, it is easy to cause the message to ferment, resulting in major public opinion risks and affecting the company's image and reputation. In addition, the workload of maintaining message group operations is large, time-consuming and labor-intensive. Property and casualty insurance companies need to arrange more operators to manage them, which increases the company's human cost expenditure. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose a public opinion processing method, device, computer equipment and storage medium to solve the technical problems that the existing manual management of message community operations by operators is heavy in workload, time-consuming and labor-intensive, and when negative content cannot be intervened and processed in time, it is easy to cause news to ferment and generate major public opinion risks.

[0006] In order to solve the above technical problems, the present application embodiment provides a method for processing public opinion, which adopts the following technical solutions:

[0007] A method for handling public opinion, comprising:

[0008] Obtaining a message text of a preset message group, and extracting local features of the message text to obtain local features of the message;

[0009] Obtaining a social relationship graph of group members and a time relationship graph of group messages of a message group;

[0010] The social relationship graph of group members and the temporal relationship graph of group messages are input into the pre-trained graph attention network to obtain the global context features of the message text;

[0011] Perform feature fusion on the local features of the message and the global context features to obtain fused features;

[0012] Perform time series analysis on the fusion features to obtain the public opinion trend prediction results.

[0013] Further, the message text of the preset message group is obtained, and the local feature extraction of the message text is performed to obtain the local feature of the message, which specifically includes:

[0014] Obtain the message text of the preset message group and perform standardization on the message text;

[0015] Inputting the standardized message text into a pre-trained convolutional neural network, wherein the convolutional neural network includes an input layer, a convolutional layer, and an output layer;

[0016] The input layer extracts features from the standardized message text to obtain the initial features of the text;

[0017] Use the convolution layer to perform local convolution operations on the initial features of the text to obtain local convolution features;

[0018] The local convolutional features are nonlinearly mapped through the output layer to obtain the local features of the message.

[0019] Furthermore, before obtaining the message text of the preset message group and extracting the local features of the message text to obtain the local features of the message, the method further includes:

[0020] Obtain historical message texts and perform standardization on the historical message texts;

[0021] Perform feature annotation on the standardized historical message text to obtain standard historical local features;

[0022] Input the standardized historical message text into the preset initial convolutional neural network;

[0023] The input layer of the initial convolutional neural network is used to extract features of the standardized historical message text to obtain historical text features;

[0024] Use the convolution layer of the initial convolutional neural network to perform local convolution operations on the historical text features to obtain historical local convolution features;

[0025] The historical local convolution features are nonlinearly mapped through the output layer of the initial convolutional neural network to obtain the local features of the historical messages;

[0026] According to the standard historical local features and historical message local features, the back propagation algorithm is used to iteratively update the initial convolutional neural network to obtain a pre-trained convolutional neural network.

[0027] Furthermore, the step of obtaining the social relationship graph of group members and the time relationship graph of group messages of the message group specifically includes:

[0028] Obtaining social relationship data of group members of the message group, and constructing a social relationship graph of group members based on the social relationship data of group members;

[0029] The group message time relationship data of the message group is obtained, and a group message time relationship graph is constructed based on the group message time relationship data.

[0030] Furthermore, the steps of obtaining social relationship data of group members of the message group and constructing a social relationship graph of group members based on the social relationship data of the group members specifically include:

[0031] Obtaining social relationship data of group members of the message group, wherein the social relationship data of group members includes group member names, group member interaction relationships, and group member interaction frequencies;

[0032] Creating a first graph structure according to the structure of the group member social relationship data, wherein the nodes of the first graph structure represent the group members, the edges of the first graph structure represent the interactive relationships among the group members, and the weights of the edges of the first graph structure represent the frequency of interaction among the group members;

[0033] The social relationship data of group members are sequentially imported into the first graph structure to generate a social relationship graph of group members.

[0034] Furthermore, the step of obtaining the group message time relationship data of the message group and constructing the group message time relationship graph based on the group message time relationship data specifically includes:

[0035] Acquire group message time relationship data of the message group, wherein the group message time relationship data includes message text, message text sending time and sending time correlation;

[0036] Creating a second graph structure according to the structure of the group message time relationship data, wherein the nodes of the second graph structure represent message texts, the edges of the second graph structure represent the message text sending time, and the weights of the edges of the second graph structure represent the sending time correlation;

[0037] The group message time relationship data is sequentially imported into the second graph structure to generate a group message time relationship graph.

[0038] Furthermore, the social relationship graph of group members and the temporal relationship graph of group messages are input into the pre-trained graph attention network to obtain the global context features of the message text, specifically including:

[0039] Obtain node feature representations of the group member social relationship graph to obtain a first node feature matrix;

[0040] Obtain feature representations of edges of the social relationship graph of group members to obtain a first adjacency matrix;

[0041] Obtaining node feature representation of the group message time relationship graph to obtain a second node feature matrix;

[0042] Obtain feature representations of edges of the group message time relationship graph to obtain a second adjacency matrix;

[0043] Based on the multi-head attention mechanism, the first node feature matrix, the second node feature matrix, the first adjacency matrix and the second adjacency matrix are concatenated to obtain the global context features of the message text.

[0044] In order to solve the above technical problems, the embodiment of the present application further provides a public opinion processing device, which adopts the following technical solution:

[0045] A public opinion processing device, comprising:

[0046] The local feature analysis module is used to obtain the message text of the preset message group and extract the local features of the message text to obtain the local features of the message;

[0047] A relationship graph acquisition module is used to obtain a social relationship graph of group members and a time relationship graph of group messages in a message group;

[0048] The global feature analysis module is used to input the social relationship graph of group members and the temporal relationship graph of group messages into the pre-trained graph attention network to obtain the global context features of the message text;

[0049] A feature fusion module is used to fuse local message features and global context features to obtain fused features;

[0050] The public opinion analysis module is used to perform time series analysis on the fusion features to obtain the public opinion trend prediction results.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0052] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the public opinion processing method as described in any one of the above items are implemented.

[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0054] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of any of the above-mentioned methods for processing public opinion.

[0055] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0056] The present application discloses a method, device, computer equipment and storage medium for public opinion processing, which belongs to the field of artificial intelligence technology. The present application realizes the multi-dimensional feature extraction and fusion of message text by combining local feature extraction of text and global context feature modeling based on graph attention network. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, the present application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content. It can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 An exemplary system architecture diagram is shown in which the present application can be applied;

[0059] Figure 2 A flowchart of an embodiment of a method for processing public opinion according to the present application is shown;

[0060] Figure 3 A schematic structural diagram of an embodiment of a public opinion processing device according to the present application is shown;

[0061] Figure 4 A schematic structural diagram of an embodiment of a computer device according to the present application is shown. DETAILED DESCRIPTION

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0063] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0065] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0066] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0067] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0068] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0069] It should be noted that the public opinion processing method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the public opinion processing device is generally arranged in the server / terminal device.

[0070] It should be understood that Figure 1 The numbers of terminal devices, networks and servers in the embodiment are merely illustrative. The above system may have any number of terminal devices, networks and servers according to implementation requirements.

[0071] Continue to refer Figure 2 , showing a flow chart of an embodiment of a method for processing public opinion according to the present application. The method for processing public opinion comprises the following steps:

[0072] S201, obtaining a message text of a preset message group, and extracting local features of the message text to obtain local features of the message.

[0073] Specifically, a preset message group refers to a social media chat group, forum discussion group or message platform cluster for a specific topic, scenario or target user group, such as an insurance company's customer rights protection group, claims consultation group, or a discussion group created by the public opinion monitoring team for a certain insurance event. The system uses collection tools to capture the message text in the group in real time or regularly to obtain the interactive content between users. These text data usually have the characteristics of diverse language expressions and scattered content, so the messages need to be cleaned and preprocessed, such as removing stop words and punctuation marks, and unifying semantic expressions. Afterwards, text feature extraction technology, such as word vector models (such as Word2Vec, GloVe) or deep learning methods (such as CNN, BERT, RoBERTa), is used to generate semantic embedding representations for message texts, extract local features from them, and capture the syntactic structure and the relationship between words in the message text to retain the core content information of a single message.

[0074] Taking the CNN (Convolutional Neural Network) model as an example, the message text is converted into a message text sequence represented by a vector, and features are further extracted through multiple layers of convolutional layers and pooling layers. The convolutional layer is used to capture local features in the text, and the pooling layer is used to reduce the feature dimension. Then, a nonlinear activation function is applied to capture the correlation between features while retaining important information. Finally, the CNN model will output a fixed-length vector representing the feature representation of each message text, that is, the local features of the message.

[0075] S202, obtaining a social relationship graph of group members and a time relationship graph of group messages of the message group.

[0076] Specifically, the social relationship graph is used to characterize the interaction and connection between users in the message group. It generates graph structure data by counting the direct communication between users, the forwarding behavior of jointly participating in topics or messages. On the other hand, the time relationship graph records the distribution of messages in the time dimension and shows the generation and propagation mode of messages. By analyzing the timestamp of each message, as well as the forwarding and response intervals of messages between users, a time graph can be established in which nodes represent messages and edges represent the time series relationship between messages. The combination of the two provides key contextual information for the model, enabling it to extract more comprehensive features from user interaction behaviors and message propagation paths.

[0077] The social relationship graph of group members is a structured representation used to characterize the interactive relationship between members in a message group, where nodes represent each member in the group, edges represent the interactive relationship between members, and the weight of the edge reflects the frequency or intensity of the interaction. The feature definition of the edge includes the member's activity (such as the frequency of sending messages, the number of times participating in discussions), influence (such as whether they often initiate topics or are cited by others), social network characteristics (such as the degree of connection with other members and the size of the social circle), and text features (such as the length of the message sent by the member, emotional tendency, and the topic involved). The weight features of the edge include the type of interaction (such as whether it is a direct reply or @ someone) and the frequency of interaction. By comprehensively analyzing these features, the social relationship graph can effectively capture the behavioral characteristics and interaction patterns of members within the group, providing rich network and contextual information for public opinion analysis.

[0078] The group message time relationship graph is a graph structure representation used to describe the relevance of messages in the time dimension, where the node represents each message text in the group, the edge represents the time correlation between messages, and the edge weight reflects the temporal relevance of the message, which can be quantified by time continuity (such as the interval between message sending) or the interaction density within the time window. For example, if two messages are sent consecutively in a short period of time, or belong to the same discussion topic within the same time window, their relevance is high, and the edge weight will be correspondingly large. Through the time relationship graph, the temporal pattern of message propagation can be effectively captured, the time evolution law of discussion and the timing of hot spot generation can be revealed, and refined feature support on the time axis can be provided for public opinion trend analysis. This time relationship graph structure is particularly suitable for analyzing the key time nodes in the claims discussion group or complaint group in the field of property insurance, where users focus on discussing a certain topic.

[0079] S203, inputting the social relationship graph of group members and the temporal relationship graph of group messages into the pre-trained graph attention network to obtain the global context features of the message text.

[0080] Specifically, the Graph Attention Network (GAT) is a deep learning model based on graph structure data that can flexibly handle heterogeneous information of nodes and edges. The social relationship graph and the time relationship graph are input into the pre-trained GAT model, and the importance of nodes and edges in the graph is weighted through the multi-head attention mechanism, so that the model can focus on key users, important messages and their associations. This process not only extracts the explicit network structure features in the message group, but also mines the hidden social patterns and time series characteristics, thereby obtaining the global features of the message in the group context. This global context feature captures the overall trend and key driving factors of message dissemination, making up for the shortcomings of relying only on local text features.

[0081] S204, performing feature fusion on the local features of the message and the global context features to obtain fused features.

[0082] Specifically, feature fusion is to unify the features from two different sources, local and global, to obtain a more comprehensive message feature representation. By splicing, weighted averaging or fusing the local features extracted in step S201 and the global context features obtained in step S203, the model can comprehensively consider the text content, social relationships and time distribution information of the message. During the fusion process, redundancy and conflict between features can be processed, for example, by optimizing feature representation through dimensionality reduction techniques (such as principal component analysis PCA) or regularization methods. The final fused features not only retain the local semantic information of the message, but also include its dynamic changes in social networks and time series.

[0083] S205, performing time series analysis on the fusion features to obtain a public opinion trend prediction result.

[0084] Specifically, time series analysis is a technique for modeling and predicting the dynamic change patterns of fused features over time. By taking fused features as input, a time series prediction algorithm (such as long short-term memory network LSTM) is used to learn the evolution law of features in the time dimension. The model can capture the potential trends of public opinion dissemination, such as the growth of message group popularity, the spread of negative emotions, or the triggering time of key events. The prediction results can intuitively present the future direction of public opinion changes, provide operators with a basis for targeted intervention, and significantly improve the predictability and response capabilities of public opinion management.

[0085] In the above embodiments, this application combines local text feature extraction and global context feature modeling based on graph attention network to construct an overall architecture based on GAT-CNN model + LSTM model to perform public opinion trend prediction and analysis, and realizes multi-dimensional feature extraction and fusion of message text. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, this application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content, and can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management.

[0086] Further, the message text of the preset message group is obtained, and the local feature extraction of the message text is performed to obtain the local feature of the message, which specifically includes:

[0087] Obtain the message text of the preset message group and perform standardization on the message text;

[0088] Inputting the standardized message text into a pre-trained convolutional neural network, wherein the convolutional neural network includes an input layer, a convolutional layer, and an output layer;

[0089] The input layer extracts features from the standardized message text to obtain the initial features of the text;

[0090] Use the convolution layer to perform local convolution operations on the initial features of the text to obtain local convolution features;

[0091] The local convolutional features are nonlinearly mapped through the output layer to obtain the local features of the message.

[0092] In this embodiment, first obtain the text data of the preset message group, such as user interaction content from WeChat group, forum discussion group or social platform message group. The obtained text is standardized, including removing redundant punctuation, unifying text format (such as case conversion), removing meaningless words (stop words), and performing operations such as word segmentation or word form restoration on unstructured text to ensure the standardization and consistency of text data. The processed text is input into a pre-trained convolutional neural network (CNN), and the initial features of the text are extracted through the input layer, which are usually the basic semantic representation of the text. Then the convolution layer performs local convolution operations on the text features on this basis to extract the local context information of the text, such as phrase patterns, grammatical structures and semantic associations in sentences. Finally, the extracted local convolution features are nonlinearly mapped through the output layer, for example, the nonlinear expression ability of the model is enhanced by an activation function (such as ReLU), and the final local features of the message are obtained. These local features can not only retain the local semantic details of the message, but also provide accurate feature representation for global feature extraction and fusion.

[0093] Through the above steps, efficient local feature extraction of message text can be achieved, significantly improving the ability to capture text detail information, providing a solid foundation for complex public opinion analysis, and ensuring the accuracy and stability of processing results.

[0094] Furthermore, before obtaining the message text of the preset message group and extracting the local features of the message text to obtain the local features of the message, the method further includes:

[0095] Obtain historical message texts and perform standardization on the historical message texts;

[0096] Perform feature annotation on the standardized historical message text to obtain standard historical local features;

[0097] Input the standardized historical message text into the preset initial convolutional neural network;

[0098] The input layer of the initial convolutional neural network is used to extract features of the standardized historical message text to obtain historical text features;

[0099] Use the convolution layer of the initial convolutional neural network to perform local convolution operations on the historical text features to obtain historical local convolution features;

[0100] The historical local convolution features are nonlinearly mapped through the output layer of the initial convolutional neural network to obtain the local features of the historical messages;

[0101] According to the standard historical local features and historical message local features, the back propagation algorithm is used to iteratively update the initial convolutional neural network to obtain a pre-trained convolutional neural network.

[0102] In this embodiment, a basic data set is first constructed by collecting historical message texts, which come from message groups or discussion groups similar to the target analysis scenario, covering various topics and interactive content. The collected historical message texts are standardized, such as text cleaning, word segmentation, stop word removal and unified format operations to ensure the consistency and quality of the input data. After standardization, the text is feature-annotated, such as by annotating emotional tendencies, topic categories, message lengths, etc. through manual or automated tools to generate standard historical local features. Next, the historical text is input into an initial convolutional neural network (CNN), and the initial features of the text are extracted through the input layer of the network. The convolution layer performs local convolution operations on these features to capture the local context pattern of the message, and the output layer generates a local feature representation of the historical message through nonlinear mapping. Subsequently, the back-propagation algorithm is used to calculate the loss based on the annotated standard historical local features and the generated historical local features, and the parameters of the initial CNN are iteratively updated so that the model can better learn the semantic features of the historical data. Finally, the pre-trained convolutional neural network obtained after the training is completed can efficiently process new message texts to provide accurate local feature extraction capabilities.

[0103] Through the above steps, historical data can be effectively used to optimize the convolutional neural network, so that the model has stronger semantic understanding and generalization capabilities, providing a high-quality basic model for feature extraction of new message texts, thereby significantly improving analysis efficiency and accuracy.

[0104] Furthermore, the step of obtaining the social relationship graph of group members and the time relationship graph of group messages of the message group specifically includes:

[0105] Obtaining social relationship data of group members of the message group, and constructing a social relationship graph of group members based on the social relationship data of group members;

[0106] The group message time relationship data of the message group is obtained, and a group message time relationship graph is constructed based on the group message time relationship data.

[0107] In this embodiment, first, the social relationship data of the group members is obtained through the data interface or record log of the message group, including the interaction records between members, such as private chat, reply, @ someone, etc., as well as the number and frequency of message forwarding, etc. After processing, these data are used to construct a social relationship graph of group members. The nodes in the graph represent group members, the edges represent the interactive relationships between members, and the weights of the edges reflect the intensity or frequency of interaction (for example, members with frequent interactions have higher connection weights). At the same time, the time relationship data of group messages is obtained, such as the sending time of each message, the time interval between adjacent messages, and the message aggregation pattern in the same time window. Based on these data, a group message time relationship graph is constructed. The nodes in the graph represent message texts, the edges represent time associations, and the weights of the edges quantify the time association between messages (for example, the shorter the time interval, the higher the weight). Through the construction of these two graphs, the social network characteristics of member interactions in the message group and the temporal dynamic characteristics of message propagation are fully captured, providing rich structured information.

[0108] Through the above steps, we can make full use of the interactive relationship between group members and the time correlation characteristics of messages to construct two types of relationship diagrams, comprehensively characterize the multi-dimensional characteristics of message dissemination, and provide in-depth support for public opinion analysis and trend prediction.

[0109] Furthermore, the steps of obtaining social relationship data of group members of the message group and constructing a social relationship graph of group members based on the social relationship data of the group members specifically include:

[0110] Obtaining social relationship data of group members of the message group, wherein the social relationship data of group members includes group member names, group member interaction relationships, and group member interaction frequencies;

[0111] Creating a first graph structure according to the structure of the group member social relationship data, wherein the nodes of the first graph structure represent the group members, the edges of the first graph structure represent the interactive relationships among the group members, and the weights of the edges of the first graph structure represent the frequency of interaction among the group members;

[0112] The social relationship data of group members are sequentially imported into the first graph structure to generate a social relationship graph of group members.

[0113] In this embodiment, the social relationship data of the group members is first obtained through the data interface or group chat log. The social relationship data includes the name of each group member, the interactive relationship with other members (such as reply, forwarding, @, etc.), and the frequency of interaction (that is, the number of times or intensity of a member interacting with other members within a certain period of time). These data will be organized into a structured format. Then, a first graph structure is created based on these structured data, in which each node represents a group member, the edge represents the interactive relationship between members, and the weight of the edge is assigned according to the frequency of interaction between group members. The edge weight between members with a higher interaction frequency is larger. By gradually importing all the social relationship data of group members into this graph structure, a complete social relationship graph of group members is finally constructed, which can intuitively show the interaction mode between group members and the strength of the relationship between members. In addition, the social relationship graph of group members can also reflect the key nodes of information flow within the group and the potential path of public opinion propagation.

[0114] Through the above steps, the social relationship diagram of group members can be accurately constructed, clearly revealing the interactive relationship and intensity between group members.

[0115] Furthermore, the step of obtaining the group message time relationship data of the message group and constructing the group message time relationship graph based on the group message time relationship data specifically includes:

[0116] Acquire group message time relationship data of the message group, wherein the group message time relationship data includes message text, message text sending time and sending time correlation;

[0117] Creating a second graph structure according to the structure of the group message time relationship data, wherein the nodes of the second graph structure represent message texts, the edges of the second graph structure represent the message text sending time, and the weights of the edges of the second graph structure represent the sending time correlation;

[0118] The group message time relationship data is sequentially imported into the second graph structure to generate a group message time relationship graph.

[0119] In this embodiment, the group message time relationship data is first obtained through the group chat log or message record, which includes the text content of each message, the sending time, and the correlation between the sending times (i.e., the time interval between adjacent messages or the frequency of messages in the same time window). By analyzing these data, the system can better understand the temporal relationship between messages. Then, according to the structure of the group message time relationship data, a second graph structure is created, in which each node represents a message text, and the edge represents the sending time correlation relationship of the message. The weight of the edge quantifies the temporal correlation between the messages. If the sending time between the two messages is very close, the weight of the edge between them is large, reflecting the close connection of the messages in time. By constructing the time relationship graph in this way, the system can capture the propagation sequence of the message and the evolution trend of the group discussion. Finally, the group message time relationship data is gradually imported into the second graph structure, and a complete group message time relationship graph is finally generated.

[0120] Through the above steps, a time relationship diagram of group messages can be accurately constructed to reveal the propagation pattern of messages in time and their mutual correlation.

[0121] Furthermore, the social relationship graph of group members and the temporal relationship graph of group messages are input into the pre-trained graph attention network to obtain the global context features of the message text, specifically including:

[0122] Obtain node feature representations of the group member social relationship graph to obtain a first node feature matrix;

[0123] Obtain feature representations of edges of the social relationship graph of group members to obtain a first adjacency matrix;

[0124] Obtaining node feature representation of the group message time relationship graph to obtain a second node feature matrix;

[0125] Obtain feature representations of edges of the group message time relationship graph to obtain a second adjacency matrix;

[0126] Based on the multi-head attention mechanism, the first node feature matrix, the second node feature matrix, the first adjacency matrix and the second adjacency matrix are concatenated to obtain the global context features of the message text.

[0127] In this embodiment, the feature representation of each node is first obtained from the social relationship graph of group members to obtain the first node feature matrix, which includes the behavioral characteristics of each member, such as activity, influence, etc.; at the same time, the features of the edges in the social relationship graph of group members are obtained to obtain the first adjacency matrix, which reflects the intensity and frequency of interaction between members. Similarly, node features are extracted from the group message time relationship graph to obtain the second node feature matrix, which represents the text features of each message, such as emotional tendency, message type, etc.; at the same time, the feature matrix of the edge of the group message time relationship graph is obtained to obtain the second adjacency matrix, which represents the correlation of the message sending time and the time adjacent relationship of the message. Then, based on the multi-head attention mechanism of the graph attention network, the first node feature matrix, the second node feature matrix, the first adjacency matrix and the second adjacency matrix are matrix spliced. Through the multi-head attention mechanism, the network can pay attention to the features of different nodes and edges, and capture information from different graph structures at the same time. Finally, these spliced ​​feature vectors will generate global context features of the message text, which can fully combine the social interaction relationship between group members and the time dimension of the message.

[0128] For example, suppose there are five members in a message group, namely A, B, C, D, and E. When the system obtains the node feature representation of the social relationship graph of group members, it extracts the behavior data of each member in the group, such as the frequency of sending messages, the number of times they participate in discussions, and whether they are the main topic initiators in the group. These behavior data are converted into feature vectors to form the first node feature matrix. For example, the features of A may include "activity = 0.8" and "influence = 0.7", and the features of B may include "activity = 0.6" and "influence = 0.5", and so on.

[0129] At the same time, the edge feature representation of the social relationship graph of group members reflects the interaction relationship and interaction frequency between members. For example, A and B interact frequently and may have a strong interaction relationship; while C and D interact less and have a lower edge weight. These interaction frequencies and intensities form the first adjacency matrix. Assuming that there are 5 interactions between A and B, 2 interactions between A and C, and 3 interactions between B and D, the adjacency matrix is ​​constructed accordingly to represent the relationship strength between members.

[0130] Assume that the above message group has 5 messages in one day, namely M1, M2, M3, M4, and M5. The system first obtains the node feature representation of the group message time relationship graph, where each node represents a message, and the node features include the sentiment tendency of the message (such as positive, negative, or neutral), the topic type (such as entertainment, news, discussion, etc.), and the length of the message or the number of interactions. For example, the sentiment tendency of M1 is "positive", the topic type is "news", and the length is "long"; the sentiment tendency of M2 is "negative", the topic type is "discussion", and the length is "short". These features constitute the second node feature matrix, which represents the basic information of each message.

[0131] At the same time, the feature representation of the edge of the group message time relationship graph reflects the temporal correlation between messages. For example, the time interval between M1 and M2 is 5 minutes, the time interval between M2 and M3 is 30 minutes, and the time interval between M4 and M5 is 1 minute. Based on these time intervals, weights can be assigned to the edges between each pair of adjacent messages. For example, the weights between messages with shorter time intervals are larger, reflecting the close correlation between the messages. These time interval information constitutes the second adjacency matrix, indicating the temporal relationship between messages.

[0132] Through the above steps, the features of the social relationship graph and the time relationship graph can be effectively integrated, and the global context features can be extracted through the graph attention network, thereby providing rich contextual information for public opinion analysis and improving analysis accuracy and prediction ability.

[0133] In the above embodiment, the present application discloses a method for processing public opinion, which belongs to the field of artificial intelligence technology. The present application realizes the multi-dimensional feature extraction and fusion of message text by combining local feature extraction of text and global context feature modeling based on graph attention network. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, the present application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content. It can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management.

[0134] In this embodiment, the electronic device (eg, Figure 1 The server shown in the figure) can receive instructions or obtain data through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0135] It should be emphasized that in order to further ensure the privacy and security of the above message text information, the above message text information can also be stored in a blockchain node.

[0136] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0137] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0138] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0139] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0140] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0141] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a public opinion processing device, and the device embodiment is Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0142] like Figure 3 As shown, the public opinion processing device 300 described in this embodiment includes:

[0143] The local feature analysis module 301 is used to obtain the message text of the preset message group, and extract the local features of the message text to obtain the local features of the message;

[0144] A relationship graph acquisition module 302 is used to acquire a social relationship graph of group members and a time relationship graph of group messages of a message group;

[0145] A global feature analysis module 303 is used to input the social relationship graph of group members and the temporal relationship graph of group messages into a pre-trained graph attention network to obtain global context features of the message text;

[0146] A feature fusion module 304 is used to fuse the local features of the message and the global context features to obtain fused features;

[0147] The public opinion analysis module 305 is used to perform time series analysis on the fusion features to obtain public opinion trend prediction results.

[0148] Furthermore, the local feature analysis module 301 is specifically used for:

[0149] The message text of a preset message group is obtained, and the message text is standardized; the standardized message text is input into a pre-trained convolutional neural network, wherein the convolutional neural network includes an input layer, a convolutional layer and an output layer; the input layer is used to extract features of the standardized message text to obtain initial features of the text; the convolutional layer is used to perform local convolution operations on the initial features of the text to obtain local convolution features; the output layer is used to perform nonlinear mapping on the local convolution features to obtain local features of the message.

[0150] Furthermore, the public opinion processing device 300 also includes a model training module, which is specifically used for:

[0151] Obtain historical message text and perform standardization on the historical message text; perform feature annotation on the standardized historical message text to obtain standard historical local features; input the standardized historical message text into a preset initial convolutional neural network; perform feature extraction on the standardized historical message text through the input layer of the initial convolutional neural network to obtain historical text features; perform local convolution operations on the historical text features using the convolutional layer of the initial convolutional neural network to obtain historical local convolutional features; perform nonlinear mapping on the historical local convolutional features through the output layer of the initial convolutional neural network to obtain historical message local features; based on the standard historical local features and the historical message local features, use the back propagation algorithm to iteratively update the initial convolutional neural network to obtain a pre-trained convolutional neural network.

[0152] Furthermore, the relationship graph acquisition module 302 is specifically used for:

[0153] The social relationship data of the group members of the message group are obtained, and a social relationship graph of the group members is constructed based on the social relationship data of the group members; the group message time relationship data of the message group are obtained, and a group message time relationship graph is constructed based on the group message time relationship data.

[0154] Furthermore, the relationship diagram acquisition module 302 is further used for:

[0155] Acquire social relationship data of group members of a message group, wherein the social relationship data of group members include group member names, group member interaction relationships and group member interaction frequencies; create a first graph structure according to the structure of the social relationship data of group members, wherein nodes of the first graph structure represent group members, edges of the first graph structure represent group member interaction relationships, and weights of edges of the first graph structure represent group member interaction frequencies; import the social relationship data of group members into the first graph structure in sequence to generate a social relationship graph of group members.

[0156] Furthermore, the relationship graph acquisition module 302 is further used to:

[0157] Acquire group message time relationship data of a message group, wherein the group message time relationship data includes message text, message text sending time and sending time correlation; create a second graph structure according to the structure of the group message time relationship data, wherein the nodes of the second graph structure represent message texts, the edges of the second graph structure represent message text sending time, and the weights of the edges of the second graph structure represent sending time correlation; import the group message time relationship data into the second graph structure in sequence to generate a group message time relationship graph.

[0158] Furthermore, the global feature analysis module 303 is specifically used for:

[0159] Obtain node feature representations of the social relationship graph of group members to obtain a first node feature matrix; obtain feature representations of edges of the social relationship graph of group members to obtain a first adjacency matrix; obtain node feature representations of the group message time relationship graph to obtain a second node feature matrix; obtain feature representations of edges of the group message time relationship graph to obtain a second adjacency matrix; perform matrix concatenation on the first node feature matrix, the second node feature matrix, the first adjacency matrix and the second adjacency matrix based on a multi-head attention mechanism to obtain global context features of the message text.

[0160] In the above embodiment, the present application discloses a public opinion processing device, which belongs to the field of artificial intelligence technology. The present application realizes the multi-dimensional feature extraction and fusion of message text by combining local feature extraction of text and global context feature modeling based on graph attention network. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, the present application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content. It can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management.

[0161] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0162] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 4 with a memory 41, a processor 42, and a network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0163] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0164] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the public opinion processing method, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0165] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the public opinion processing method.

[0166] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0167] In the above embodiment, the present application discloses a computer device, which belongs to the field of artificial intelligence technology. The present application realizes the multi-dimensional feature extraction and fusion of message text by combining local feature extraction of text and global context feature modeling based on graph attention network. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, the present application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content. It can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management.

[0168] The present application also provides another implementation, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the public opinion processing method as described above.

[0169] In the above embodiment, the present application discloses a computer-readable storage medium, which belongs to the field of artificial intelligence technology. The present application realizes the multi-dimensional feature extraction and fusion of message text by combining local feature extraction of text and global context feature modeling based on graph attention network. By introducing the social relationship graph of group members and the time relationship graph of group messages, the propagation characteristics of messages in social networks and time series are captured, and time series analysis is used to predict public opinion trends. Compared with the traditional method that relies on manual management, the present application greatly reduces the workload and time cost of operators, while improving the efficiency of monitoring and intervention of negative content. It can quickly discover and predict potential public opinion risks, effectively curb the fermentation of messages, and significantly improve the intelligence and precision of public opinion management.

[0170] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0171] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0172] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for processing public opinion, characterized in that: include: Obtaining a message text of a preset message group, and performing text local feature extraction on the message text to obtain a message local feature; Obtaining a social relationship graph of group members and a time relationship graph of group messages of the message group; Inputting the group member social relationship graph and the group message time relationship graph into a pre-trained graph attention network to obtain global context features of the message text; Performing feature fusion on the local features of the message and the global context features to obtain fused features; A time series analysis is performed on the fusion features to obtain a public opinion trend prediction result.

2. The method for processing public opinion according to claim 1, characterized in that: The step of obtaining the message text of the preset message group and extracting local features of the message text to obtain local features of the message specifically includes: Obtaining a message text of a preset message group and performing standardization processing on the message text; Inputting the standardized message text into a pre-trained convolutional neural network, wherein the convolutional neural network includes an input layer, a convolutional layer, and an output layer; Performing feature extraction on the standardized message text through the input layer to obtain initial features of the text; Using the convolution layer to perform a local convolution operation on the initial text feature to obtain a local convolution feature; The local convolutional features are nonlinearly mapped through the output layer to obtain the local features of the message.

3. The method for processing public opinion according to claim 2, characterized in that: Before the step of obtaining the message text of the preset message group and extracting local features of the message text to obtain the local features of the message, the method further includes: Acquire historical message texts, and perform standardization processing on the historical message texts; Performing feature annotation on the standardized historical message text to obtain standard historical local features; Inputting the standardized historical message text into a preset initial convolutional neural network; Performing feature extraction on the standardized historical message text through the input layer of the initial convolutional neural network to obtain historical text features; Using the convolution layer of the initial convolutional neural network to perform a local convolution operation on the historical text features to obtain historical local convolution features; Performing nonlinear mapping on the historical local convolutional features through the output layer of the initial convolutional neural network to obtain local features of historical messages; According to the standard historical local features and the historical message local features, the initial convolutional neural network is iteratively updated using a back propagation algorithm to obtain the pre-trained convolutional neural network.

4. The method for processing public opinion according to claim 1, characterized in that: The step of obtaining the social relationship graph of group members and the time relationship graph of group messages of the message group specifically includes: Acquire social relationship data of group members of the message group, and construct a social relationship graph of the group members based on the social relationship data of the group members; The group message time relationship data of the message group is obtained, and the group message time relationship graph is constructed based on the group message time relationship data.

5. The method for processing public opinion according to claim 4, characterized in that: The step of obtaining the social relationship data of the group members of the message group and constructing the social relationship graph of the group members based on the social relationship data of the group members specifically includes: Acquire social relationship data of group members of the message group, wherein the social relationship data of group members includes group member names, group member interaction relationships, and group member interaction frequencies; Creating a first graph structure according to the structure of the group member social relationship data, wherein the nodes of the first graph structure represent group members, the edges of the first graph structure represent interactive relationships among group members, and the weights of the edges of the first graph structure represent the frequency of interaction among group members; The group member social relationship data is sequentially imported into the first graph structure to generate the group member social relationship graph.

6. The method for processing public opinion according to claim 4, characterized in that: The step of obtaining the group message time relationship data of the message group and constructing the group message time relationship graph based on the group message time relationship data specifically includes: Acquire group message time relationship data of the message group, wherein the group message time relationship data includes message text, message text sending time and sending time correlation; Creating a second graph structure according to the structure of the group message time relationship data, wherein the nodes of the second graph structure represent message texts, the edges of the second graph structure represent the message text sending time, and the weights of the edges of the second graph structure represent the sending time correlation; The group message time relationship data is sequentially imported into the second graph structure to generate the group message time relationship graph.

7. The method for processing public opinion according to claim 1, characterized in that: The step of inputting the group member social relationship graph and the group message time relationship graph into a pre-trained graph attention network to obtain the global context features of the message text specifically includes: Obtaining node feature representations of the group member social relationship graph to obtain a first node feature matrix; Acquire feature representations of edges of the social relationship graph of group members to obtain a first adjacency matrix; Obtaining a node feature representation of the group message time relationship graph to obtain a second node feature matrix; Obtaining feature representations of edges of the group message time relationship graph to obtain a second adjacency matrix; Based on a multi-head attention mechanism, matrix concatenation is performed on the first node feature matrix, the second node feature matrix, the first adjacency matrix, and the second adjacency matrix to obtain global context features of the message text.

8. A public opinion processing device, characterized in that: include: A local feature analysis module is used to obtain the message text of a preset message group, and extract local features of the message text to obtain local features of the message; A relationship graph acquisition module, used to acquire a social relationship graph of group members and a time relationship graph of group messages of the message group; A global feature analysis module, used to input the group member social relationship graph and the group message time relationship graph into a pre-trained graph attention network to obtain the global context features of the message text; A feature fusion module, used to fuse the local features of the message and the global context features to obtain fused features; The public opinion analysis module is used to perform time series analysis on the fusion features to obtain public opinion trend prediction results.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the public opinion processing method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the public opinion processing method according to any one of claims 1 to 7 are implemented.

Images