Great emergency public opinion reversal evolution prediction model based on intelligent algorithm

Abstract

The invention relates to the technical field of public opinion analysis, in particular to an important emergency public opinion reversal evolution prediction model based on an intelligent algorithm, which comprises a data acquisition module, a data preprocessing module, an emotion analysis module, an evolution path recognition module and a reversal prediction module. Wherein the data acquisition module is used for acquiring public opinion data from multiple data sources in real time; the sentiment analysis module is used for classifying text contents in the public opinions into positive, negative or neutral sentiments; the evolution path identification module is used for constructing a public opinion evolution path network and identifying key nodes and potential rules of public opinion evolution; and the inversion prediction module is used for predicting a public opinion inversion time window and an inversion direction. According to the invention, through combining sentiment analysis, deep learning and public opinion evolution path network technologies, accurate prediction of public opinion inversion of major emergencies is realized, the time window and direction of public opinion inversion can be effectively identified, and scientific decision support is provided for crisis management.

Description

Technical Field

[0001] The present invention relates to the technical field of public opinion analysis, and particularly to a prediction model for the evolution and reversal of public opinion on major emergencies based on intelligent algorithms. Background Art

[0002] With the rapid development of information technology, digital platforms such as social media and online news platforms have become the main channels for the spread of public opinion information; when major emergencies occur, the spread of public opinion is extremely fast and has a wide range of social impacts, directly affecting public sentiment and social stability; especially in the initial stage of the event, the emotional trend and development trend of public opinion are often full of uncertainties and difficult to predict; however, how to timely grasp the dynamic changes of public opinion and predict the possibility and time window of the reversal of public opinion has become an important challenge in crisis management and public opinion response.

[0003] Existing technologies mainly rely on rule-based models or simple statistical methods in public opinion analysis and prediction. These methods are often difficult to process complex public opinion data, especially when facing emergencies, and the prediction accuracy is relatively low; traditional public opinion analysis methods are difficult to effectively capture the amplitude of emotional changes, the key nodes of public opinion dissemination, and potential laws; at the same time, existing public opinion reversal prediction models usually lack in-depth exploration of the development trend of public opinion and are difficult to accurately predict the time and direction of the reversal. In order to improve the accuracy of public opinion prediction, there is an urgent need for an intelligent algorithm model that can dynamically learn the evolution path of public opinion, combine sentiment analysis and dissemination patterns, and perform accurate prediction, so as to provide scientific decision-making support for crisis management. Summary of the Invention

[0004] Based on the above purpose, the present invention provides a prediction model for the evolution and reversal of public opinion on major emergencies based on intelligent algorithms.

[0005] The prediction model for the evolution and reversal of public opinion on major emergencies based on intelligent algorithms includes a data collection module, a data preprocessing module, a sentiment analysis module, an evolution path recognition module, and a reversal prediction module; wherein:

[0006] The data collection module: is used to collect public opinion data in real time from multiple data sources, and store the collected public opinion data in chronological order to form an original public opinion data set;

[0007] The data preprocessing module: is connected to the data collection module, and is used to receive the original public opinion data set and perform preprocessing on it, including data cleaning, data deduplication, and data normalization, to generate a standardized public opinion data set;

[0008] Sentiment Analysis Module: Connected to the data preprocessing module, it receives the standardized public opinion dataset, analyzes the sentiment tendency of public opinion data based on natural language processing technology, classifies the text content in public opinion into positive, negative or neutral sentiment, and outputs the sentiment analysis result;

[0009] Evolution Path Identification Module: Connected to the sentiment analysis module, it receives the sentiment analysis result, constructs a public opinion evolution path network based on graph analysis and graph neural network algorithms, identifies the key nodes and potential laws of public opinion evolution, and generates public opinion evolution path data;

[0010] Reversal Prediction Module: Connected to the evolution path identification module, it receives the public opinion evolution path data and historical public opinion data, and predicts the time window and direction of public opinion reversal based on machine learning algorithms.

[0011] Optionally, the data collection module includes a data source access unit, a data storage unit, and a data marking unit; where:

[0012] Data Source Access Unit: Used to access multiple public opinion data sources, including social media platforms, news websites, forums, and blogs, obtain public opinion data in real time through the API interface, and transmit the obtained original public opinion data to the data storage unit;

[0013] Data Storage Unit: Used to store the public opinion data from the data source access unit in chronological order to generate an original public opinion dataset;

[0014] Data Marking Unit: Used to mark the time stamps of the public opinion data stored in the data storage unit to ensure the clear time order of each piece of data, and mark each piece of public opinion data with a unique identifier.

[0015] Optionally, the data preprocessing module includes a data cleaning unit, a data deduplication unit, and a data normalization unit; where:

[0016] Data Cleaning Unit: Used to filter out noise and correct errors in the received original public opinion dataset, specifically including removing irrelevant information, filtering advertisements and spam content, and correcting spelling and grammar errors in the text to improve the accuracy of the data;

[0017] Data Deduplication Unit: Used to detect and remove duplicate data in the original public opinion dataset to ensure the uniqueness of each data record;

[0018] Data Normalization Unit: Used to convert the preprocessed public opinion dataset into a unified format and standard, specifically including unifying the time format, standardizing the numerical range, and unifying the text encoding method.

[0019] Optionally, the sentiment analysis module includes a sentiment classification unit and a sentiment intensity calculation unit; where:

[0020] Sentiment classification unit: Based on natural language processing technology, it performs sentiment tendency analysis on the text content in the received standardized public opinion dataset, and accurately classifies the text content into positive, negative or neutral sentiment;

[0021] Sentiment intensity calculation unit: It is used to quantitatively calculate the sentiment categories output by the sentiment classification unit, and determine the intensity values of various sentiments in each piece of public opinion data to reflect the strength of the sentiment.

[0022] Optionally, the sentiment intensity calculation unit includes:

[0023] Intensity weighted calculation of sentiment words: Evaluate the intensity of the sentiment categories output by the sentiment classification unit; according to the sentiment polarity value and word frequency statistics of sentiment words, use a weighted algorithm to calculate the intensity of each type of sentiment. Let the intensity of each sentiment word be S i , and the calculation formula is: S i =f(w i , t i )=E(w i )·f(t i ), where S i represents the intensity value of the i-th sentiment word, E(w i ) is the sentiment polarity value of the sentiment word w i , f(t i ) is the word frequency of the sentiment word w i in the text, and t i is the number of occurrences of the sentiment word;

[0024] Total intensity calculation of sentiment categories: Calculate the overall intensity of each sentiment according to the intensity values of each sentiment word in the text; assume that the text T contains n p positive sentiment words, n n negative sentiment words and n z neutral sentiment words, then the total intensities of positive, negative and neutral sentiments are calculated respectively as: Among them, S positive , S negative and S neutral respectively represent the total intensities of positive, negative and neutral sentiments in the text T;

[0025] Determination of the final sentiment intensity: By comparing the total intensity values of various sentiments, determine the dominant sentiment category of the text; if S positive is the largest, the dominant sentiment of this text is positive; if S negative is the largest, it is negative; if S neutral is the largest, it is neutral.

[0026] Optionally, the evolution path recognition module includes a graph construction unit, a deep learning analysis unit, and a path generation unit; where:

[0027] Graph construction unit: Based on the sentiment analysis results output by the sentiment analysis module, it uses graph analysis technology to construct a public opinion evolution path network. Different public opinion events are represented by nodes, and the propagation relationship between public opinion events is represented by edges, forming a structured public opinion evolution graph;

[0028] Deep learning analysis unit: Applies a graph neural network algorithm to analyze the public opinion evolution path network. By learning the complex relationships between public opinion events, it identifies key nodes and potential laws in the public opinion evolution process, and extracts relevant factors affecting public opinion evolution;

[0029] Path generation unit: Used to generate public opinion evolution path data according to the output results of the deep learning analysis unit. By analyzing key nodes and potential laws, it depicts the main direction and development trend of public opinion dissemination, forming public opinion evolution path data.

[0030] Optionally, the deep learning analysis unit includes:

[0031] Graph neural network model construction: Used to represent the public opinion evolution path network as a graph structure, where nodes represent public opinion events and edges represent the propagation relationship between public opinion events, and construct a graph neural network model;

[0032] Information dissemination and node update: In the graph neural network, the feature vector of the node is updated through the information dissemination mechanism. Specifically, let the node feature representation of the t-th layer be Then the new feature vector of node i Is updated by the transfer of the feature information of its neighbor nodes, and the formula is: Among them, Is the feature vector of node i at the t-th layer, N(i) is the set of neighbor nodes of node i, c ij Is the edge weight between node i and node j, W (t) And b (t) Are learnable parameters, and σ is the activation function;

[0033] Key node identification: Extract the final feature vector after updating the node, which is used to measure the importance of each node in the public opinion evolution; then the key node can be identified by calculating the centrality index of the node feature vector; if the centrality C(i) of node i is greater than the preset threshold, then this node is considered a key node, and the calculation formula is:

[0034] Among them, d iis the degree of node i, and shortest path(j, k) represents the length of the shortest path between nodes j and k;

[0035] Potential law extraction: By learning the variation law of node features in the graph neural network and combining the propagation path and the relationship between nodes, extract the potential laws affecting public opinion evolution; specifically, identify the clustering structure of similar nodes through the similarity calculation between node feature vectors, and extract the influencing factors of public opinion evolution from it.

[0036] Optionally, the path generation unit includes:

[0037] Propagation direction determination: Identify the dominant path of public opinion dissemination by evaluating the propagation intensity and frequency between key nodes; let the propagation intensity be P ij , and its expression is: where P ij represents the propagation intensity from node i to node j, w ij is the propagation weight between nodes i and j, N(i) represents the set of neighbor nodes of node i, and β is a parameter controlling the change of propagation intensity;

[0038] Development trend analysis: Train the time series data of public opinion events through the time series analysis algorithm to predict the future development trend T of public opinion f , and the formula is: where T f represents the development trend of the public opinion event at the future time point f, α t is the weight coefficient in the time series, h t is the feature vector of the public opinion event corresponding to the time point t, and n1 is the time length of historical public opinion data;

[0039] Path data generation: Construct public opinion evolution path data according to the results of propagation direction determination and development trend analysis Its data structure includes event sequence, propagation intensity, propagation path, and time node; the path data generation formula is: where, represents the public opinion evolution path data, e i represents the public opinion event, P ij represents the propagation intensity from event e i to event e j , and t i is the time node when event e i occurs.

[0040] Optionally, the reverse prediction module includes a time window prediction unit and a reverse direction prediction unit; where:

[0041] Time window prediction unit: It is used to combine the public opinion evolution path data, historical public opinion data and current public opinion sentiment trend, and adopt time series analysis technology to predict the time window of public opinion reversal. By analyzing the trend of historical public opinion data and using the weighted average method to calculate the expected time point of reversal, the calculation formula is: Where, T rev is the predicted time point of public opinion reversal, T i is the reversal time node in the historical public opinion data, α i is the weight of the event, and n2 is the number of events in the historical data;

[0042] Reversal direction prediction unit: It is used to combine the public opinion evolution path data and the sentiment analysis result, and analyze the change trend of public opinion sentiment through a deep learning algorithm to identify the direction of public opinion reversal; specifically, based on the amplitude of sentiment change and the propagation mode, predict the potential direction of public opinion reversal, and output the positive or negative direction of public opinion reversal through a classification algorithm.

[0043] Optionally, the reversal direction prediction unit includes:

[0044] Sentiment change analysis: Based on the sentiment analysis result output by the sentiment analysis module, calculate the amplitude of sentiment change ΔS in the public opinion data, and the formula is: ΔS = |S end -S start |, where, S end represents the sentiment intensity before reversal, and S start represents the sentiment intensity after reversal;

[0045] Propagation mode analysis: Calculate the propagation probability between nodes to evaluate the propagation mode, and the formula is: Where, γ ij represents the propagation probability from node i to node j, w ij is the propagation weight between node i and node j, β is the parameter controlling the change of propagation intensity, and N(i) represents the set of neighbor nodes of node i;

[0046] Direction prediction: Combine the outputs of sentiment change analysis and propagation mode analysis, and use a classification algorithm to predict the direction of public opinion reversal. Adopt the logistic regression algorithm, and calculate the probability of the reversal direction through the following formula: Where, P direction is the probability of the reversal direction, θ0, θ1, θ2 are model parameters, ΔS is the amplitude of sentiment change, and γ ij is the propagation probability; when P direction is greater than 0.5, predict the reversal direction as positive; otherwise, predict it as negative.

[0047] Advantages of the present invention:

[0048] The present invention can analyze and predict the evolution process of public opinion in major emergencies in real time by combining deep learning technology and natural language processing technology; by performing sentiment analysis on public opinion data, it can accurately identify the trend of sentiment changes in public opinion and analyze the potential possibility of its reversal; this data-driven prediction method can precisely capture the critical moments of public opinion fluctuations.

[0049] The present invention further enhances the reliability and accuracy of prediction by constructing a public opinion evolution path network and combining propagation mode analysis; by dynamically learning the complex relationships between public opinion events, it can deeply explore the potential laws and influencing factors of public opinion reversal, thereby providing a scientific reversal time window and direction prediction; this method not only improves the accuracy of public opinion prediction but also provides effective decision-making support for crisis management. Brief Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0051] Figure 1 Schematic diagram of the public opinion reversal evolution prediction model for the embodiments of the present invention;

[0052] Figure 2 Schematic diagram of the evolution path recognition module for the embodiments of the present invention. Detailed Embodiments

[0053] The present invention will be described in detail below in combination with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0054] It should be pointed out that in the specification, when referring to "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc., it indicates that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. In addition, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0055] Generally, terms can be understood at least in part from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or property in a singular sense, or can be used to describe a combination of features, structures, or properties in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but rather can alternatively, depending at least in part on the context, allow for other factors that may not be explicitly described.

[0056] As Figure 1 - Figure 2 shown, a major emergency public opinion reversal evolution prediction model based on intelligent algorithms includes a data collection module, a data preprocessing module, a sentiment analysis module, an evolution path identification module, and a reversal prediction module; wherein:

[0057] The data collection module: is used to collect public opinion data in real time from multiple data sources (including social media platforms, news websites, forums, blogs, etc.), and store the collected public opinion data in chronological order to form an original public opinion data set;

[0058] The data preprocessing module: is connected to the data collection module, and is used to receive the original public opinion data set and preprocess it, including data cleaning, data deduplication, and data normalization, to generate a standardized public opinion data set;

[0059] The sentiment analysis module: is connected to the data preprocessing module, receives the standardized public opinion data set, analyzes the sentiment tendency of the public opinion data based on natural language processing technology, and classifies the text content in the public opinion as positive, negative, or neutral sentiment, and outputs the sentiment analysis result;

[0060] The evolution path identification module: is connected to the sentiment analysis module, receives the sentiment analysis result, constructs a public opinion evolution path network based on graph analysis and graph neural network algorithms, identifies the key nodes and potential laws of public opinion evolution, and generates public opinion evolution path data;

[0061] The reversal prediction module: is connected to the evolution path identification module, receives the public opinion evolution path data and historical public opinion data, and predicts the time window and direction of public opinion reversal based on machine learning algorithms.

[0062] The data collection module includes a data source access unit, a data storage unit, and a data marking unit; wherein:

[0063] The data source access unit: is used to access multiple public opinion data sources, including social media platforms, news websites, forums, and blogs, obtain public opinion data in real time through the API interface method, and transmit the obtained original public opinion data to the data storage unit;

[0064] Data storage unit: It is used to store the public opinion data from the data source access unit in chronological order, generate the original public opinion data set, ensure the integrity and timeliness of the data, and facilitate subsequent analysis and processing;

[0065] Data marking unit: It is used to mark the time stamps of the public opinion data stored in the data storage unit, ensure the clear time order of each piece of data, and mark each piece of public opinion data with a unique identifier for subsequent traceability and analysis; Through the structural design of the above data acquisition module, public opinion information can be efficiently and comprehensively collected from multiple data sources in real time, and the comprehensiveness and timeliness of the data are ensured through the API interface. The introduction of the time sorting unit further guarantees the ordered storage of the data, providing a high-quality raw data basis for subsequent public opinion analysis and prediction.

[0066] The data preprocessing module includes a data cleaning unit, a data deduplication unit, and a data normalization unit; among them:

[0067] Data cleaning unit: It is used to filter out noise and correct errors in the received original public opinion data set, specifically including removing irrelevant information, filtering advertisements and spam content, and correcting spelling and grammar errors in the text to improve the accuracy and usability of the data;

[0068] Data deduplication unit: It is used to detect and remove duplicate data in the original public opinion data set, ensure the uniqueness of each data record, and thus avoid the impact of redundant information on subsequent analysis;

[0069] Data normalization unit: It is used to convert the preprocessed public opinion data set into a unified format and standard, specifically including unifying the time format, standardizing the numerical range, and unifying the text encoding method, so as to facilitate the processing and analysis of subsequent modules; Through the collaborative work of the above units, the data preprocessing module can effectively convert the original public opinion data set into a standardized public opinion data set with high quality and unified format, providing a reliable data basis for the subsequent sentiment analysis module.

[0070] The sentiment analysis module includes a sentiment classification unit and a sentiment intensity calculation unit; among them:

[0071] Sentiment classification unit: It is used to perform sentiment tendency analysis on the text content in the received standardized public opinion data set based on natural language processing technology, combined with a sentiment dictionary and a sentiment classification algorithm, and accurately classify the text content into positive, negative or neutral sentiment;

[0072] The specific steps include:

[0073] Use an emotion dictionary to identify emotion words in the text, and calculate the overall emotion tendency of the text based on the emotion polarity of each emotion word in the predefined emotion dictionary; each word in the emotion dictionary is associated with its corresponding emotion score (a positive value indicates a positive emotion, a negative value indicates a negative emotion, and a zero value indicates a neutral emotion);

[0074] Calculate the emotion score of the text based on the weighted summation method of emotion words. The formula is as follows:

[0075] where S(T) represents the overall emotion value of text T, w i is the weight of the i-th emotion word in the text, and E(w i ) is the emotion polarity value corresponding to the emotion word w i , and n is the total number of emotion words in the text; for the emotion value calculated, if S(T)>0, it is classified as a positive emotion; if S(T)<0, it is classified as a negative emotion; if S(T)=0, it is classified as a neutral emotion;

[0076] Emotion intensity calculation unit: used to perform quantitative calculation on the emotion categories output by the emotion classification unit, determine the intensity values of various emotions in each piece of public opinion data to reflect the strength of emotions; through the collaborative work of the above units, the emotion analysis module can accurately identify and classify the emotion tendencies in public opinion data and quantify the emotion intensity, providing detailed emotion analysis results for the subsequent evolution path recognition module.

[0077] The emotion intensity calculation unit includes:

[0078] Intensity weighting calculation of emotion words: evaluate the intensity of the emotion categories output by the emotion classification unit; according to the emotion polarity value and word frequency statistics of emotion words, use a weighting algorithm to calculate the intensity of each type of emotion. Let the intensity of each emotion word be S i , and the calculation formula is: S i =f(w i , t i )=E(w i )·f(t i ), where S i represents the intensity value of the i-th emotion word, E(w i ) is the emotion polarity value of the emotion word w i , f(t i ) is the word frequency of the emotion word w i in the text, and t i is the number of occurrences of the emotion word; this formula shows that the intensity value of an emotion word is not only related to its polarity but also proportional to the word frequency (i.e., the frequency of occurrence);

[0079] Calculation of the total intensity of emotion categories: Based on the intensity values of each emotion word in the text, calculate the overall intensity of each emotion (positive, negative, or neutral); assume that the text T contains n p positive emotion words, n n negative emotion words, and n z neutral emotion words, then the total intensities of positive, negative, and neutral emotions are calculated respectively as follows: where S positive , S negative , and S neutral represent the total intensities of positive, negative, and neutral emotions in the text T respectively;

[0080] Determination of the final emotion intensity: By comparing the total intensity values of various emotions, determine the dominant emotion category of the text; if S positive is the largest, the dominant emotion of the text is positive; if S negative is the largest, it is negative; if S neutral is the largest, it is neutral; through the design of the above emotion intensity calculation unit, the emotion intensity in the text can be accurately quantified, avoiding the errors brought by simple emotion classification methods, and providing more detailed emotion analysis data for subsequent public opinion evolution path recognition and reversal prediction.

[0081] The evolution path recognition module includes a graph construction unit, a deep learning analysis unit, and a path generation unit; among them:

[0082] Graph construction unit: Used to construct a public opinion evolution path network based on the emotion analysis results output by the emotion analysis module, using graph analysis technology. Different public opinion events are represented by nodes, and the propagation relationship between public opinion events is represented by edges, forming a structured public opinion evolution graph;

[0083] Deep learning analysis unit: Apply graph neural network algorithms to analyze the public opinion evolution path network, learn the complex relationships between public opinion events, identify key nodes and potential laws in the public opinion evolution process, and extract relevant factors affecting public opinion evolution;

[0084] Path generation unit: Used to generate public opinion evolution path data according to the output results of the deep learning analysis unit. By analyzing key nodes and potential laws, depict the main direction and development trend of public opinion dissemination, and form public opinion evolution path data.

[0085] The deep learning analysis unit includes:

[0086] Graph Neural Network Model Construction: It is used to represent the public opinion evolution path network as a graph structure, where nodes represent public opinion events, edges represent the propagation relationships between public opinion events, a graph neural network model is constructed, the feature vector of each node represents information such as the sentiment feature and propagation intensity of the public opinion event, and the weight of the edge represents the association intensity between public opinion events;

[0087] Information Propagation and Node Update: In the graph neural network, the feature vector of the node is updated through the information propagation mechanism. Specifically, let the node feature representation of the t-th layer be Then the new feature vector of node i is updated by the transfer of the feature information of its neighbor nodes, and the formula is: where, is the feature vector of node i at the t-th layer, N(i) is the set of neighbor nodes of node i, c ij is the edge weight between node i and node j, W (t) and b (t) are learnable parameters, and σ is the activation function (such as ReLU);

[0088] Key Node Identification: Extract the final feature vector after updating the node, which is used to measure the importance of each node in the public opinion evolution; then the key nodes can be identified by calculating the centrality index of the node feature vector. The centrality index includes the clustering coefficient, degree centrality or betweenness centrality of the node; if the centrality C(i) of node i is greater than the preset threshold, then this node is considered a key node, and the calculation formula is:

[0089] where, d i is the degree of node i, and shortest path(j, k) represents the shortest path length between node j and node k;

[0090] Potential Law Extraction: By learning the change law of node features in the graph neural network, combined with the propagation path and the relationship between nodes, extract the potential laws affecting public opinion evolution; specifically, by calculating the similarity between node feature vectors, identify the clustering structure of similar nodes, and extract the influencing factors of public opinion evolution from them, such as sentiment fluctuations, time delays in public opinion propagation, etc. The similarity sim(i, j) between node feature vectors can be calculated using cosine similarity, and the formula is: where, h i and h j represent the feature vectors of node i and node j respectively, and ‖·‖ is the norm of the vector; through the above steps, the deep learning analysis unit can effectively learn the complex relationships between public opinion events, identify the key nodes in public opinion evolution, and extract potential laws, providing accurate support for public opinion reversal prediction.

[0091] The path generation unit includes:

[0092] Determination of the propagation direction: Using the key nodes and potential patterns identified by the deep learning analysis unit, analyze the connection relationships and propagation paths between the key nodes to determine the main direction of public opinion propagation. Specifically, by evaluating the propagation intensity and frequency between each key node, identify the dominant path of public opinion propagation; let the propagation intensity be P ij , and its expression is: Among them, P ij represents the propagation intensity from node i to node j, w ij is the propagation weight between node i and node j, N(i) represents the set of neighbor nodes of node i, and β is a parameter that controls the change of propagation intensity; through this formula, the intensity of propagation between nodes can be evaluated, and the main propagation direction can be identified;

[0093] Analysis of the development trend: Based on the potential patterns extracted by the deep learning analysis unit, predict the future development trend of public opinion propagation. Train the time series data of public opinion events through a time series analysis algorithm (such as Long Short-Term Memory Network LSTM) to predict the future development trend T of public opinion f , and the formula is: Among them, T f represents the development trend of the public opinion event at the future time point f, α t is the weight coefficient in the time series, h t is the feature vector of the public opinion event corresponding to the time point t, and n1 is the time length of historical public opinion data; this formula predicts the future trend of public opinion evolution by weighted summation of the influence of historical data;

[0094] Path data generation: According to the results of the determination of the propagation direction and the analysis of the development trend, construct public opinion evolution path data Its data structure includes event sequence, propagation intensity, propagation path, and time node; the path data generation formula is: Among them, represents the public opinion evolution path data, e i represents the public opinion event, P ij represents the propagation intensity from event e i to event e j , and t i is the time node when event e i occurs; this formula generates complete public opinion evolution path data by collecting the propagation paths, intensities, and time information of all events; through the above steps, the path generation unit can accurately depict the main direction and development trend of public opinion propagation, generate detailed public opinion evolution path data, and provide a reliable path basis for public opinion reversal prediction.

[0095] The reverse prediction module includes a time window prediction unit and a reverse direction prediction unit; where:

[0096] Time window prediction unit: used to combine the public opinion evolution path data, historical public opinion data and the current public opinion sentiment trend, and adopt time series analysis technology to predict the time window of public opinion reversal. By analyzing the trend of historical public opinion data and using the weighted average method to calculate the expected time point of reversal, the calculation formula is: Where, T rev is the predicted time point of public opinion reversal, T i is the reversal time node in the historical public opinion data, α i is the weight of the event, indicating the impact of each reversal event in the historical data on the current prediction, and n2 is the number of events in the historical data;

[0097] Reverse direction prediction unit: used to combine the public opinion evolution path data and the sentiment analysis result, and analyze the change trend of public opinion sentiment through a deep learning algorithm to identify the direction of public opinion reversal; specifically, based on the amplitude and propagation mode of sentiment change, predict the potential direction of public opinion reversal, and output the positive or negative direction of public opinion reversal through a classification algorithm; through the above units, the reverse prediction module can accurately predict the time window and direction of public opinion reversal, and then provide early warning and support decision-making. This module combines historical public opinion data with the current sentiment trend, improving the accuracy and real-time performance of reverse prediction.

[0098] The reverse direction prediction unit includes:

[0099] Sentiment change analysis: Based on the sentiment analysis result output by the sentiment analysis module, calculate the amplitude of sentiment change ΔS in the public opinion data, and the formula is: ΔS = |S end -S start |, where, S end represents the sentiment intensity before reversal, S start represents the sentiment intensity after reversal. This formula is used to quantify the absolute value of sentiment change and reflect the degree of sentiment change;

[0100] Propagation mode analysis: Calculate the propagation probability between nodes to evaluate the propagation mode, and the formula is: Where, γ ij represents the propagation probability from node i to node j, w ij is the propagation weight between node i and node j, β is the parameter controlling the change of propagation intensity, and N(i) represents the set of neighbor nodes of node i. Through this formula, the propagation relationship between public opinion events can be quantified to help analyze the public opinion propagation mode;

[0101] Direction prediction: Combining the outputs of sentiment change analysis and propagation pattern analysis, a classification algorithm is used to predict the direction (positive or negative) of public opinion reversal. The logistic regression algorithm is adopted to calculate the probability of the reversal direction through the following formula: where P direction is the probability of the reversal direction, θ0, θ1, θ2 are model parameters, ΔS is the amplitude of sentiment change, and γ ij is the propagation probability; when P direction is greater than 0.5, the predicted reversal direction is positive; otherwise, it is predicted to be negative. Through the collaborative work of the above steps, the reversal direction prediction unit can accurately predict the direction (positive or negative) of public opinion reversal based on the amplitude of sentiment change and the propagation pattern, and output the corresponding prediction result.

[0102] This invention covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc. are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0103] The above description is only a preferred embodiment of this invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements can be made without departing from the principle of this invention, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. A prediction model for the reversal evolution of public opinion in major emergencies based on intelligent algorithms, characterized by: It includes data collection module, data preprocessing module, sentiment analysis module, evolution path identification module and reversal prediction module; among which: Data collection module: used to collect public opinion data from multiple data sources in real time, and store the collected public opinion data in chronological order to form the original public opinion data set; Data preprocessing module: connected to the data acquisition module, used to receive the original public opinion data set and preprocess it, including data cleaning, data deduplication and data normalization, to generate a standardized public opinion data set; Sentiment analysis module: connected to the data preprocessing module, receiving the standardized public opinion data set, analyzing the sentiment tendency of the public opinion data based on natural language processing technology, and classifying the text content in the public opinion into positive, negative or neutral sentiment, and outputting the sentiment analysis results; Evolution path identification module: connected to the sentiment analysis module, receives sentiment analysis results, builds a public opinion evolution path network based on graph analysis and graph neural network algorithms, identifies key nodes and potential laws of public opinion evolution, and generates public opinion evolution path data; Reversal prediction module: connected to the evolution path identification module, receiving public opinion evolution path data and historical public opinion data, and predicting the time window and direction of public opinion reversal based on machine learning algorithm.

2. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 1 is characterized in that: The data acquisition module includes a data source access unit, a data storage unit and a data marking unit; wherein: Data source access unit: used to access multiple public opinion data sources, including social media platforms, news websites, forums and blogs, obtain public opinion data in real time through API interfaces, and transmit the obtained raw public opinion data to the data storage unit; Data storage unit: used to store the public opinion data from the data source access unit in chronological order to generate an original public opinion data set; Data marking unit: used to timestamp the public opinion data stored in the data storage unit to ensure that the time sequence of each piece of data is clear and mark each piece of public opinion data with a unique identifier.

3. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 1 is characterized in that: The data preprocessing module includes a data cleaning unit, a data deduplication unit and a data normalization unit; wherein: Data cleaning unit: used to filter noise and correct errors in the received raw public opinion data set, including removing irrelevant information, filtering advertisements and junk content, and correcting spelling errors and grammatical errors in the text to improve the accuracy of the data; Data deduplication unit: used to detect and remove duplicate data in the original public opinion data set to ensure the uniqueness of each data record; Data normalization unit: used to convert the preprocessed public opinion dataset into a unified format and standard, including a unified time format, standardized numerical range, and unified text encoding method.

4. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 1 is characterized in that: The sentiment analysis module includes a sentiment classification unit and a sentiment intensity calculation unit; wherein: Sentiment classification unit: used to perform sentiment analysis on the text content in the received standardized public opinion data set based on natural language processing technology, and accurately classify the text content into positive, negative or neutral sentiment; Sentiment intensity calculation unit: used to quantify the sentiment categories output by the sentiment classification unit, determine the intensity values ​​of each type of sentiment in each piece of public opinion data, and reflect the strength of the sentiment.

5. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 4 is characterized in that: The emotion intensity calculation unit comprises: Weighted calculation of the intensity of sentiment words: evaluate the intensity of the sentiment categories output by the sentiment classification unit; calculate the intensity of each type of sentiment using a weighted algorithm based on the sentiment polarity value and word frequency statistics of the sentiment words, assuming that the intensity of each sentiment word is S i , the calculation formula is: S i =f(w i , t i )=E(w i )·f(t i ), where S i represents the intensity value of the i-th sentiment word, E(w i ) is a sentiment word w i The sentiment polarity value, f(t i ) is a sentiment word w i The frequency of the word in the text, t i is the number of occurrences of sentiment words; Calculation of the total intensity of sentiment categories: Calculate the overall intensity of each sentiment based on the intensity value of each sentiment word in the text; suppose that the text T contains n p positive sentiment words, n n negative sentiment words and n z neutral sentiment words, the total strength of positive, negative and neutral sentiments are calculated as: Among them, S positive , S negative and S neutral Represent the total intensity of positive, negative and neutral sentiment in text T respectively; Determination of the final emotional intensity: By comparing the total intensity values ​​of various emotions, the dominant emotional category of the text is determined; if S positive The largest, then the dominant sentiment of the text is positive; if S negative Maximum, it is negative; if S neutral Maximum is neutral.

6. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 1 is characterized in that: The evolution path identification module includes a graph construction unit, a deep learning analysis unit and a path generation unit; wherein: Graph construction unit: used to construct a public opinion evolution path network based on the sentiment analysis results output by the sentiment analysis module using graph analysis technology. Nodes represent different public opinion events, and edges represent the propagation relationship between public opinion events, thus forming a structured public opinion evolution graph. Deep learning analysis unit: Applying graph neural network algorithm to analyze the public opinion evolution path network, identifying key nodes and potential laws in the public opinion evolution process by learning the complex relationship between public opinion events, and extracting relevant factors affecting the evolution of public opinion; Path generation unit: used to generate public opinion evolution path data based on the output results of the deep learning analysis unit, and to describe the main direction and development trend of public opinion dissemination by analyzing key nodes and potential rules, thus forming public opinion evolution path data.

7. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 6 is characterized in that: The deep learning analysis unit comprises: Graph neural network model construction: used to represent the public opinion evolution path network as a graph structure, where nodes represent public opinion events and edges represent the propagation relationship between public opinion events, and to build a graph neural network model; Information propagation and node update: In graph neural networks, the feature vector of the node is updated through the information propagation mechanism. Specifically, the feature vector of the node in the tth layer is represented as Then the new feature vector of node i is It is updated by transmitting the characteristic information of its neighboring nodes, and the formula is: in, is the feature vector of node i at layer t, N(i) is the set of neighbor nodes of node i, c ij is the edge weight between node i and node j, W (t) and b (t) is a learnable parameter, σ is an activation function; Key node identification: extract the final feature vector after updating the node to measure the importance of each node in the evolution of public opinion; the key node can be identified by calculating the centrality index of the node feature vector; if the centrality C(i) of node i is greater than the preset threshold, the node is considered to be a key node, and the calculation formula is: Among them, d i is the degree of node i, shortest path (j, k) represents the shortest path length between node j and node k; Potential law extraction: By learning the changing rules of node features in graph neural networks and combining the propagation path and the relationship between nodes, the potential laws that affect the evolution of public opinion are extracted; specifically, by calculating the similarity between node feature vectors, the clustering structure of similar nodes is identified, and the influencing factors of public opinion evolution are extracted from them.

8. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 7 is characterized in that: The path generation unit comprises: Determine the direction of propagation: By evaluating the propagation intensity and frequency between key nodes, identify the dominant path of public opinion propagation; let the propagation intensity be P ij , whose expression is: Among them, P ij represents the transmission intensity from node i to node j, w ij is the propagation weight between node i and node j, N(i) represents the set of neighbor nodes of node i, and β is the parameter that controls the change of propagation strength; Development trend analysis: Use time series analysis algorithms to train the time series data of public opinion events and predict the development trend of future public opinion. f , the formula is: Among them, T f represents the development trend of public opinion events at the future time point f, α t is the weight coefficient in the time series, h t is the feature vector of the public opinion event corresponding to time point t, and n1 is the time length of the historical public opinion data; Path data generation: Based on the results of the propagation direction determination and development trend analysis, the public opinion evolution path data D is constructed. path , its data structure includes event sequence, propagation intensity, propagation path and time node; the path data generation formula is: D path ={(e1,P 12 , t1),(e2,P 23 , t2),...,(e n , P (n-1)n , t n )}, where D path represents the public opinion evolution path data, e i represents a public opinion event, P ij Indicates that from event e i To event e j The propagation intensity, t i It is event i The time point when it occurs.

9. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 1 is characterized in that: The reversal prediction module includes a time window prediction unit and a reversal direction prediction unit; wherein: Time window prediction unit: It is used to combine the public opinion evolution path data, historical public opinion data and current public opinion sentiment trend, and use time series analysis technology to predict the time window of public opinion reversal. The expected time point of reversal is calculated by trend analysis of historical public opinion data and weighted average method. The calculation formula is: Among them, T rev T is the time point for predicting the reversal of public opinion. i is the reversal time node in the historical public opinion data, α i is the weight of the event, n2 is the number of events in the historical data; Reversal direction prediction unit: used to combine the public opinion evolution path data and sentiment analysis results, analyze the changing trend of public opinion sentiment through deep learning algorithms, and identify the direction of public opinion reversal; specifically, based on the amplitude and propagation pattern of sentiment changes, predict the potential direction of public opinion reversal, and output the positive or negative direction of public opinion reversal through classification algorithms.

10. The prediction model for the reversal evolution of public opinion on major emergencies based on intelligent algorithms according to claim 9 is characterized in that: The reverse direction prediction unit comprises: Sentiment change analysis: Based on the sentiment analysis results output by the sentiment analysis module, the amplitude of sentiment change in public opinion data, ΔS, is calculated. The formula is: ΔS = |S end -S start |, where S end Indicates the emotional intensity before the reversal, S start Indicates the intensity of emotion after the reversal; Propagation mode analysis: The propagation probability between nodes is calculated to evaluate the propagation mode. The formula is: Among them, γ ij represents the probability of propagation from node i to node j, w ij is the propagation weight between node i and node j, β is the parameter that controls the change of propagation strength, and N(i) represents the set of neighbor nodes of node i; Direction prediction: Combining the outputs of sentiment change analysis and communication pattern analysis, a classification algorithm is used to predict the direction of public opinion reversal. A logistic regression algorithm is used to calculate the probability of the reversal direction using the following formula: Among them, P direction is the probability of reversal direction, θ0, θ1, θ2 are model parameters, ΔS is the amplitude of emotion change, γ ij is the propagation probability; when P direction When it is greater than 0.5, the predicted reversal direction is positive; otherwise, the prediction is negative.

Images