An Opinion Publicity Joint Prediction Method and Device Combining a Fusion Mechanism Model and Deep Learning

Abstract

This application relates to an opinion joint prediction method and device that integrates a mechanism model and deep learning. The message passing mechanism of a graph neural network is introduced to replace the traditional social interaction mechanism to achieve automatic optimization of model parameters. During the information interaction process, an individual constructs an adaptive influence weight calculation mechanism based on a graph attention mechanism to adapt to the dynamic changes of influence weights between objects during the interaction process, thereby controlling the information flow in the social network. In addition, a real opinion, derived opinion, and environmental noise time series triple input is constructed, and on the basis of considering noise factors, the fusion of the model derivation result and real data is realized.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and particularly to an opinion sentiment joint prediction method and device integrating a mechanism model and deep learning. Background Art

[0002] The evolution of public opinion is a dynamic process in which people in society exchange opinions around various topics, such as the release of new products, social hot events, etc. In this process, individual opinions will change over time, thus influencing the development trend of group public opinion. At present, the exponential growth of social media has made the spread of public opinion show complex characteristics such as multi-modal, non-linear, and cross-scale. Deeply understanding the mechanism of public opinion evolution not only helps to accurately predict the trend of emotions, such as the spread trajectory of crisis events, but also provides strong support for social system decision-making such as commercial marketing and public opinion management. However, individual opinions in the public opinion system are affected by the coupling of multiple factors such as the social network topology structure, heterogeneous interaction rules, and external event disturbances. At present, there are mainly two methods in public opinion research: the model-driven method based on public opinion dynamics can depict the information interaction between individuals, but due to the large number of unknown parameters in the model, it faces the "curse of dimensionality", resulting in high computational costs and low efficiency; the data-driven method based on deep learning performs well in short-term prediction, but due to its "mechanism black box" characteristics, it is difficult to be widely promoted in practical applications. Summary of the Invention

[0003] Based on this, in view of the above technical problems, it is necessary to provide an opinion sentiment joint prediction method and device integrating a mechanism model and deep learning.

[0004] An opinion sentiment joint prediction method integrating a mechanism model and deep learning, the method comprising:

[0005] Constructing a public opinion propagation social network; the public opinion propagation social network includes individual opinions expressed by users as nodes, the relationships between users as edges, and the social weights of the influence of users on opinions;

[0006] Using a graph neural network to calculate the social weights between each pair of adjacent users in the public opinion propagation social network;

[0007] According to the individual opinion of each user and the social weight corresponding to the user, performing a single-step prediction on the opinion of the user at the next moment to obtain the predicted individual opinion of the user at the next moment;

[0008] Constructing a triple according to the predicted individual opinion, the individual opinion at the current moment, and a constant bias term;

[0009] Inputting the triple into a sequence model to predict the public opinion propagation result of the user.

[0010] In one embodiment, the public opinion dissemination social network is represented as ;

[0011] The individual opinion is represented as:

[0012] ;

[0013] Wherein, represents the set of individual opinions, represents the individual opinion of user at time , represents the set of users;

[0014] The social weight is represented as:

[0015] ;

[0016] Wherein, represents the social weight at time represents the influence degree of the individual opinion of user at time on user

[0017] In one embodiment, it further includes: calculating the social weight between each pair of adjacent users in the public opinion dissemination social network by using a graph neural network as:

[0018] ;

[0019] Wherein, represents the neighborhood of user in the public opinion dissemination social network, represents the scoring function with input .

[0020] In one embodiment, it further includes: performing a single-step prediction on the view of the user at the next moment according to the individual view of each user and the social weight corresponding to the user, and obtaining the predicted individual view of the user at the next moment as:

[0021] ;

[0022] Wherein, represents the predicted individual view of user , represents the influence degree of the individual view of user at time on user

[0023] In one embodiment, when all users in the public opinion dissemination social network adopt the individual opinions of other users, the GNN network is used to predict the individual opinions. Update to:

[0024] ;

[0025] where W represents the parameter matrix reflecting the relationship between users, represents the true opinion of user at time

[0026] In one embodiment, when the social weights of users in the public opinion dissemination social network are affected by the initial opinions of users, the FJ model is used to predict the individual opinions. Update to:

[0027] ;

[0028] where is a constant bias term, representing the degree of adherence to the initial opinions of users.

[0029] In one embodiment, it further includes: training the sequence model using the sliding window method to obtain a trained sequence model; inputting the triple into the sequence model to predict the public opinion dissemination result of users.

[0030] In one embodiment, the sequence model includes: LSTM model, Transformer, Mamba, and xLSTM model.

[0031] A public opinion joint prediction device integrating a mechanism model and deep learning, the device includes:

[0032] A network construction module, used to construct a public opinion dissemination social network; the public opinion dissemination social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights of users' influence on opinions.

[0033] A weight calculation module, used to calculate the social weights between each pair of adjacent users in the public opinion dissemination social network using a graph neural network.

[0034] A single-step prediction module, used to perform a single-step prediction on the opinions of users at the next moment according to the individual opinions of each user and the social weights corresponding to the users, and obtain the predicted individual opinions of users at the next moment.

[0035] A triple construction module, used to construct a triple according to the predicted individual opinions, the individual opinions at the current moment, and the constant bias term.

[0036] A prediction module for inputting the triple into a sequence model to predict the user's public opinion dissemination result.

[0037] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0038] Construct a public opinion dissemination social network; the public opinion dissemination social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights representing the degree of influence of users on opinions.

[0039] Use a graph neural network to calculate the social weights between each pair of adjacent users in the public opinion dissemination social network.

[0040] Based on each user's individual opinion and the corresponding social weight of the user, perform a single-step prediction on the user's opinion at the next moment to obtain the predicted individual opinion of the user at the next moment.

[0041] Construct a triple based on the predicted individual opinion, the individual opinion at the current moment, and a constant bias term.

[0042] Input the triple into a sequence model to predict the user's public opinion dissemination result.

[0043] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0044] Construct a public opinion dissemination social network; the public opinion dissemination social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights representing the degree of influence of users on opinions.

[0045] Use a graph neural network to calculate the social weights between each pair of adjacent users in the public opinion dissemination social network.

[0046] Based on each user's individual opinion and the corresponding social weight of the user, perform a single-step prediction on the user's opinion at the next moment to obtain the predicted individual opinion of the user at the next moment.

[0047] Construct a triple based on the predicted individual opinion, the individual opinion at the current moment, and a constant bias term.

[0048] Input the triple into a sequence model to predict the user's public opinion dissemination result.

[0049] The above-mentioned combined prediction method and device for public opinion of the fusion mechanism model and deep learning introduce the message passing mechanism of the graph neural network to replace the traditional social interaction mechanism to achieve automatic optimization of model parameters. During the information interaction process, individuals construct an adaptive influence weight calculation mechanism based on the graph attention mechanism to adapt to the dynamic changes of the influence weights between objects during the interaction process, thereby controlling the information flow in the social network. In addition, a real view, a derived view, and an environmental noise time series triple input are constructed, and on the basis of considering noise factors, the fusion of the model derivation result and real data is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flowchart of the combined prediction method for public opinion of the fusion mechanism model and deep learning in an embodiment;

[0051] Figure 2 It is a schematic diagram of the time series model in an embodiment;

[0052] Figure 3 It is a structural block diagram of the combined prediction device for public opinion of the fusion mechanism model and deep learning in an embodiment;

[0053] Figure 4 It is an internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] In one embodiment, as Figure 1 shown, a combined prediction method for public opinion of the fusion mechanism model and deep learning is provided, including the following steps:

[0056] Step 102, construct a social network for public opinion dissemination.

[0057] The social network for public opinion dissemination includes individual views expressed by users as nodes, relationships between users as edges, and social weights of the influence degree of users on the views.

[0058] Step 104, use the graph neural network to calculate the social weights between each pair of adjacent users in the social network for public opinion dissemination.

[0059] Step 106, based on the individual views of each user and the social weights corresponding to the users, perform a single-step prediction on the views of the users at the next moment to obtain the predicted individual views of the users at the next moment.

[0060] Step 108: Construct a triple based on the predicted individual view, the individual view at the current moment, and the constant bias term, and input the triple into the sequence model to predict the result of user public opinion dissemination.

[0061] In the above-mentioned public opinion joint prediction method that combines the fusion mechanism model and deep learning, the message passing mechanism of the graph neural network is introduced to replace the traditional social interaction mechanism to achieve the automatic optimization of model parameters. During the information interaction process, individuals construct an adaptive influence weight calculation mechanism based on the graph attention mechanism to adapt to the dynamic changes of the influence weights between objects during the interaction process, thereby controlling the information flow in the social network. In addition, a time series triple input of real views, derived views, and environmental noise is constructed to achieve the fusion of the model derivation results and real data while considering noise factors.

[0062] In one embodiment, the public opinion dissemination social network is represented as ; The individual view is represented as:

[0063] ;

[0064] Among them, represents the set of individual views, represents the individual view of user at time , represents the set of users; The social weight is represented as:

[0065] ;

[0066] Among them, represents the social weight at time t, represents the degree of influence of the individual view of user at time on user represents the degree of adherence of the user to their own view.

[0067] Specifically, the individual view can be obtained by deriving through the posts published by the individual on social media (such as Weibo) . The individual views are divided into five categories: extremely disgusted , disgusted , neutral , in favor , and very much in favor . The set of disturbances of external events during the evolution process is denoted as , aiming to predict the view value of users at any time within a given time window .

[0068] In one of the embodiments, the social weights between each pair of adjacent users in the social network of public opinion dissemination are calculated using a graph neural network as follows:

[0069] ;

[0070] where, represents the neighborhood of user in the social network of public opinion dissemination, and represents a scoring function with an input of .

[0071] In this embodiment, by calculating the social weights between users in real time, accurate derivation is achieved.

[0072] In one of the embodiments, based on the individual opinions of each user and the corresponding social weights of the users, a single-step prediction of the opinions of the users at the next moment is performed, and the predicted individual opinions of the users at the next moment are obtained as follows:

[0073] ;

[0074] where, represents the predicted individual opinion of user , and represents the influence degree of the individual opinion of user at time on user

[0075] In this embodiment, taking user 2 as an example, this user has social relationships with user 1 and user 3. Therefore, the opinion of user 2 will be directly affected by these two and itself. Thus, the predicted value of the opinion of user 2 at time is the weighted sum of the three, that is, .

[0076] For the above single-step prediction model, the loss function is the mean square error between the deduced opinion value and the true value at each moment, that is:

[0077] .

[0078] To further simulate various situations, in one of the embodiments, when all users in the social network of public opinion dissemination adopt the individual opinions of other users, the GNN network is used to update the predicted individual opinion to:

[0079] ;

[0080] where, W represents a parameter matrix reflecting the relationships between users, and represents the user at time True view.

[0081] In this embodiment, the opinion of an individual is represented by a real number, which can represent a certain decision preference or probability estimate. The individual updates his opinion based on the influence weight of his neighbors on himself and the opinions of his neighbors, that is:

[0082] ;

[0083] The general way of GNN:

[0084] .

[0085] Denote the -th layer of GNN aggregates information from the neighbor nodes of each node, while denotes the function of the -th layer of GNN that fuses the neighbor node and the current node representation to update the node criterion. And is consistent with the initial node representation of the node. This means that when , and the initial node representation is the initial opinion of the node , the GNN network can be used to replace the Degroot model.

[0086] In another embodiment, when in the public opinion dissemination social network, the social weight of a user is affected by the user's initial opinion, the FJ model is used to predict the individual opinion and update it to:

[0087] ;

[0088] where is a constant bias term, representing the degree of persistence of the user's initial opinion.

[0089] In this embodiment, when in the public opinion dissemination social network, the social weight of a user is affected by the user's initial opinion, it is expressed as:

[0090] ;

[0091] Therefore, the expression of FJ can be considered as the required formula. Since FJ has nothing to do with the current opinion of the individual, it can be considered as a weighted sum operation, and the weight of the current opinion is 0.

[0092] In another embodiment, different from the above linear opinion dynamics model, the bounded confidence model is a class of non-linear opinion dynamics models used to describe that individuals are only influenced by those with similar opinions during the opinion update process. In this type of model, the opinion update of each individual depends on the opinions of its neighbors within its confidence interval, denoted as , taking the HK model as an example, , the opinion update rule of the HK model is

[0093] ;

[0094] It is not difficult to find that the essence of the confidence interval is to control the flow of information. Therefore, it can be constructed by imitating the graph attention mechanism, by introducing attention weights for each pair of nodes ,

[0095] ;

[0096] where, due to the existence of external noise, is not fixed either. If this parameter is set to change over time, then in fact, the constraint can be removed. The value range of itself is . When , it can be regarded as exceeding the information threshold. Therefore, this hyperparameter can be ignored.

[0097] In one of the embodiments, the sequence model is trained using the sliding window method to obtain a trained sequence model; the triple is input into the sequence model to predict the user public opinion propagation result.

[0098] Specifically, after realizing the single-step opinion speculation, the time series part will synthesize the existing information on this basis to realize the reasoning and prediction of time series data. When performing time series calculation, there are mainly three inputs: the opinion value at the current moment , the opinion at the next moment speculated by the GNN-based Surrogate model , and the bias term including external interference factors and the user's own initial intention. Therefore, the input of each time step of the time series module is a triple, where ANMP represents Adaptive Neural Message Passing, as Figure 2 shown.

[0099] To make full use of the data, the sliding window method is used for training, that is, given a window width and a prediction step size , and represent the moment and The predicted value of the time view, and respectively represent the time and the bias term at time. The sliding window divides the time series data into multiple input-output pairs. For time step , the input feature contains the time step to of data points, while the output contains to of data points. Thus, the data set can be converted into input-output pairs through the sliding window, and the goal of the sequence model is to predict the output window through the input window, where are the time series model parameters. Therefore, the training function is the mean square error of the two, that is:

[0100] ;

[0101] Since there is a label for the individual at the final time, the loss function of the final time series module should also include the cross-entropy function, that is .

[0102] In one embodiment, the sequence model includes: LSTM model, Transformer, Mamba, and xLSTM model.

[0103] Specifically, the LSTM model: The LSTM model is an improved recurrent neural network, mainly solving the problem of long-term dependencies in predicting time series. This method provides long-term memory storage function by introducing memory units and controls the data flow through gating mechanisms (including input gate, forget gate, and output gate) to alleviate the problem of gradient disappearance.

[0104] Transformer model: The Transformer model is a model built entirely using the attention mechanism based on the encoder-decoder framework. This model is the basis of many current large models, such as GPT, BERT, etc.

[0105] Mamba model: Mamba is an efficient general-purpose sequence modeling architecture designed specifically to address the computational efficiency problem of Transformer in processing long sequences. Mamba is based on an improved structured state space model (SSM), and by making the SSM parameters dynamic (as a function of the input), it realizes the selective propagation or forgetting of sequence information.

[0106] xLSTM Model: xLSTM is an extended LSTM model that enhances the dynamic representation ability of LSTM by introducing the Exponential Gating mechanism and combining normalization and stability techniques. At the same time, the model improves the memory structure to adapt to the design concept of modern large language models.

[0107] It should be understood that although Figure 1 each step in the flowchart is shown in sequence according to the indication of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise clearly stated in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 at least a part of the steps in

[0108] In one embodiment, as Figure 3 shown, an opinion joint prediction device that combines a mechanism model and deep learning is provided, including: a network construction module 302, a weight calculation module 304, a single-step prediction module 306, a triple construction module 308, and a prediction module 310, where:

[0109] The network construction module 302 is used to construct an opinion propagation social network; the opinion propagation social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights of the influence degree of users on opinions;

[0110] The weight calculation module 304 is used to calculate the social weights between each pair of adjacent users in the opinion propagation social network by using a graph neural network;

[0111] The single-step prediction module 306 is used to perform a single-step prediction on the opinions of users at the next moment according to the individual opinions of each user and the social weights corresponding to the users, and obtain the predicted individual opinions of users at the next moment;

[0112] The triple construction module 308 is used to construct a triple according to the predicted individual opinions, the individual opinions at the current moment, and a constant bias term;

[0113] The prediction module 310 is used to input the triple into a sequence model to predict the opinion propagation result of users.

[0114] In one of the embodiments, the opinion propagation social network is represented as ;

[0115] The individual view is expressed as:

[0116] ;

[0117] Among them, represents the set of individual views, represents the individual view of user at time , represents the set of users;

[0118] The social weight is expressed as:

[0119] ;

[0120] Among them, represents the social weight at time t, represents the influence degree of the individual view of user at time on user

[0121] In one embodiment, a graph neural network is used to calculate the social weight between each pair of adjacent users in the public opinion propagation social network as:

[0122] ;

[0123] Among them, represents the neighborhood of user in the public opinion propagation social network, represents the scoring function with input .

[0124] In one embodiment, according to the individual views of each user and the social weights corresponding to the users, a single-step prediction is made on the views of the users at the next moment, and the predicted individual views of the users at the next moment are obtained as:

[0125] ;

[0126] Among them, represents the predicted individual view of user , represents the influence degree of the individual view of user at time on user

[0127] In one embodiment, when all users in the public opinion propagation social network adopt the individual views of other users, the GNN network is used to make the predicted individual views Updated to:

[0128] ;

[0129] Wherein, W represents a parameter matrix reflecting the relationship between users, represents the true view of user at time

[0130] In one embodiment, when in the public opinion dissemination social network, the social weight of a user is affected by the initial opinion of the user, the FJ model is used to predict the individual view Updated to:

[0131] ;

[0132] Wherein, is a constant bias term, representing the degree of persistence of the initial opinion of the user.

[0133] In one embodiment, the sliding window method is used to train the sequence model to obtain a trained sequence model; the triple is input into the sequence model to predict the user public opinion dissemination result.

[0134] In one embodiment, the sequence model includes: LSTM model, Transformer, Mamba, and xLSTM model.

[0135] For the specific limitations of the public opinion joint prediction device regarding the fusion mechanism model and deep learning, reference can be made to the limitations of the public opinion joint prediction method for the fusion mechanism model and deep learning in the above text, which will not be elaborated here. Each module in the above public opinion joint prediction device for the fusion mechanism model and deep learning can be implemented in whole or in part through software, hardware, and their combination. The above modules can be embedded in the processor in the computer device in hardware form or be independent of it, or be stored in the memory in the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.

[0136] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 4As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a public opinion joint prediction method that combines a fusion mechanism model and deep learning. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0137] Those skilled in the art can understand that Figure 4 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0138] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the method in the above embodiment.

[0139] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, it implements the steps of the method in the above embodiment.

[0140] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0141] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0142] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. An opinion sentiment joint prediction method integrating a mechanism model and deep learning, characterized in that, The method includes: Construct an opinion dissemination social network; the opinion dissemination social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights representing the degree of influence of users on opinions; Use a graph neural network to calculate the social weights between each pair of adjacent users in the opinion dissemination social network; Based on the individual opinions of each user and the corresponding social weights of the users, perform a single-step prediction on the opinions of users at the next moment to obtain the predicted individual opinions of users at the next moment; Construct a triple based on the predicted individual opinions, the individual opinions at the current moment, and a constant bias term; Input the triple into a sequence model to predict the result of user opinion dissemination; Based on the individual opinions of each user and the corresponding social weights of the users, perform a single-step prediction on the opinions of users at the next moment to obtain the predicted individual opinions of users at the next moment, including: Based on the individual opinions of each user and the corresponding social weights of the users, perform a single-step prediction on the opinions of users at the next moment, and the predicted individual opinions of users at the next moment are: Among them, represents the predicted individual view of the user i , represents t the influence degree of the individual view of the user j at the moment on the user i . When all users in the public opinion dissemination social network adopt the individual opinions of other users, the GNN network will be used to predict individual opinions Update to: where W represents a parameter matrix reflecting the relationships between users; When the social weight of a user in the public opinion dissemination social network is affected by the user's initial opinion, the FJ model is used to predict individual opinions Update to: Among them, is a constant bias term, indicating the degree of adherence to the user's initial opinion.

2. The method according to claim 1, wherein The public opinion dissemination social network is expressed as ; The individual opinion is expressed as: Among them, represents the set of individual opinions, represents t the individual opinion of the user at the moment i ; , represents the set of users; The social weight is expressed as: Among them, represents the social weight at time t, represents t the degree of influence of the individual opinion of user j at time on user i.

3. The method according to claim 1, wherein Using a graph neural network to calculate the social weights between each pair of adjacent users in the opinion dissemination social network, including: Using a graph neural network to calculate the social weights between each pair of adjacent users in the opinion dissemination social network is: Among them, represents the neighborhood of user i in the public opinion dissemination social network, represents the input as scoring function.

4. The method according to any one of claims 1 to 3, characterized in that, Input the triple into a sequence model to predict the result of user opinion dissemination, including: Use the sliding window method to train the sequence model to obtain a trained sequence model; Input the triple into a sequence model to predict the result of user opinion dissemination.

5. The method according to claim 4, wherein The sequence model includes: LSTM model, Transformer, Mamba, and xLSTM model.

6. An opinion sentiment joint prediction device integrating a mechanism model and deep learning, which is used to implement the opinion sentiment joint prediction method integrating a mechanism model and deep learning according to any one of claims 1 to 5, and is characterized in that, The device includes: A network construction module for constructing an opinion dissemination social network; the opinion dissemination social network includes individual opinions expressed by users as nodes, relationships between users as edges, and social weights representing the degree of influence of users on opinions; A weight calculation module for using a graph neural network to calculate the social weights between each pair of adjacent users in the opinion dissemination social network; A single-step prediction module for performing a single-step prediction on the opinions of users at the next moment based on the individual opinions of each user and the corresponding social weights of the users to obtain the predicted individual opinions of users at the next moment; A triple construction module for constructing a triple based on the predicted individual opinions, the individual opinions at the current moment, and a bias term; A prediction module for inputting the triple into a sequence model to predict the result of user opinion dissemination.

Images