A network public opinion reversal prediction system using AI technology

Abstract

The present invention provides a network public opinion reversal prediction system using AI technology, which relates to the field of electronic digital data processing, and includes an event collection module, an event analysis module, a public opinion training module and a reversal prediction module. The event collection module is used to collect public opinion events on the network, the event analysis module is used to classify the collected public opinion events and extract feature information of the events, the public opinion training module is used to train the collected events, and the reversal prediction module is used to predict the reversal of real-time events. By training and learning the reversal situations of existing events, the system can accurately predict and judge the reversal possibility of the current event, thereby reducing the degree of harm to the parties caused by public opinion.

Description

Technical Field

[0001] The present invention relates to the field of electronic digital data processing, and in particular to a network public opinion reversal prediction system using AI technology. Background Art

[0002] In the current development of the Internet, more and more people are participating in online activities, and the influence of online public opinion is also growing. However, due to some factors, events on the Internet tend to reverse, but the damage caused to the parties by public opinion before the reversal cannot be reversed. Therefore, a system is needed to predict the possibility of reversal of ongoing events. Based on the prediction results, public opinion participants can control their degree of participation, thereby reducing the damage to the parties caused by public opinion.

[0003] Now many public opinion prediction systems have been developed. After a lot of searching and reference, it is found that the existing public opinion prediction systems are such as the system disclosed by the publication number CN117407806A. These system methods generally include: obtaining multiple network public opinion event information from the supervision object library; wherein the network public opinion event information includes network public opinion reversal event information and network public opinion non-reversal event information; extracting the features of the network public opinion event information through the feature construction model, and converting the features into feature values ​​to obtain the public opinion event data of multiple public opinion events to be detected; inputting the public opinion event data of the public opinion event to be detected into the reversal prediction model, and outputting the prediction result of whether the public opinion event to be detected will be reversed; inputting the prediction result of the public opinion event to be detected into the SHAP explanation model, and outputting the main factors affecting the reversal of public opinion; and conducting public opinion control based on the public opinion supervision module and the main factors affecting the reversal of public opinion. However, the model in this system does not take into account the development process of the event, resulting in the accuracy of the prediction results to be improved. Summary of the invention

[0004] The purpose of the present invention is to propose a network public opinion reversal prediction system using AI technology to address the existing deficiencies.

[0005] The present invention adopts the following technical solution:

[0006] A network public opinion reversal prediction system using AI technology, including an event collection module, an event analysis module, a public opinion training module and a reversal prediction module;

[0007] The event collection module is used to collect public opinion events on the Internet, the event analysis module is used to classify the collected public opinion events and extract feature information of the events, the public opinion training module is used to train the collected events, and the reversal prediction module is used to perform reversal prediction on real-time events;

[0008] The event collection module includes an information collection unit, an event sorting unit and a data storage unit. The information collection unit is used to collect public opinion information on the Internet social platform. The event sorting unit is used to sort the collected public opinion information into multiple event information. The data storage unit is used to store the sorted event information.

[0009] The event analysis module includes a feature extraction unit, an event classification unit and a sample storage unit, wherein the feature extraction unit is used to extract feature data from event development information, the event classification unit classifies events into reversal events and non-reversal events based on event comment information, and the sample storage unit stores feature data and corresponding event types as samples;

[0010] The public opinion training module includes a sample input unit, a feature processing unit and a comparison optimization unit. The sample input unit is used to obtain sample data from the sample storage unit. The feature processing unit is used to perform nonlinear change processing on the feature data in the sample data and output a judgment result. The comparison optimization unit compares the event type in the sample with the judgment result and optimizes the parameters in the feature processing unit according to the comparison result.

[0011] The reversal prediction module includes a real-time data access unit and a prediction execution unit, wherein the real-time data access unit is used to receive real-time event information and perform preprocessing, and the prediction execution unit is used to obtain parameter information in the public opinion training module and perform reversal prediction on the preprocessed real-time event information;

[0012] Further, the feature extraction unit includes a node extraction processor and a feature conversion processor, the node extraction processor is used to extract event node text from event development information, and the feature conversion processor is used to convert the extracted event node text into feature information;

[0013] The process of converting the feature information obtained by the feature conversion processor comprises the following steps:

[0014] S1. Obtain keywords in the node text and convert each keyword into a spatial coordinate as a key point;

[0015] S2, select a keyword as the basis of the event node, and the corresponding spatial coordinates of the keyword as the base point;

[0016] S3, calculate the vector with the base point as the vector starting point and the other key points as the vector end points, and obtain the key vector set;

[0017] S4. Calculate the detail index DT of the event node according to the following formula:

[0018] ;

[0019] Among them, n is the number of key points, n0 is the detail base, c is the bias value, a i is the i-th key vector;

[0020] S5, select two key points closest to the base point, and connect the two key points and the base point to obtain a triangular plane;

[0021] S6. Calculate the area S and unit normal vector of the triangular plane ;

[0022] The feature conversion processor converts the area S, the vector , the vector corresponding to the base point coordinates and detail index DT as the characteristic information of the event node;

[0023] Furthermore, the feature processing unit includes a change analysis processor and a judgment output processor, wherein the change analysis processor is used to analyze the changes of feature data of different event nodes, and the judgment output processor is used to output the judgment result;

[0024] The change analysis processor calculates S, and The change value Vg is obtained by processing:

[0025] ;

[0026] Among them, S(1) is the S value of the first feature data, S(2) is the S value of the second feature data, is the vector of the first feature data , is the vector of the second feature data , is the vector of the first feature data , is the vector of the second feature data , DT(1) is the DT value of the first feature data, DT(2) is the DT value of the second feature data, , , and is the model variation parameter;

[0027] Furthermore, the judgment output processor calculates the judgment value Q according to the following formula:

[0028] ;

[0029] Where M is the number of event nodes, Vg i represents the i-th change value, and Determine parameters for the model;

[0030] When the Q value is greater than 0, the event is judged as a reversal event, otherwise the event is judged as a non-reversal event;

[0031] Furthermore, the event classification unit includes a comment scoring processor, a time series processing processor and a reversal judgment processor. The comment scoring processor is used to perform biased scoring on each comment information. The time series processing processor classifies and sorts the event nodes targeted by each comment to obtain a comprehensive score for each event node. The reversal judgment processor judges the event as a reversal event or a non-reversal event according to the change of the comprehensive score of the event node, and marks the reversal node at the same time.

[0032] The beneficial effects achieved by the present invention are:

[0033] This system can automatically divide the collected event information into reversal events and non-reversal events by analyzing the comments, thus improving the efficiency of pre-training preparation. In the process of processing event information, an event is split into multiple event nodes, and the feature information of each event node is extracted. By processing the feature information through the model, it can effectively reflect the impact of the development process of the event on the possibility of reversal, and finally improve the accuracy of the prediction through parameter optimization.

[0034] To further understand the features and technical contents of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are only for reference and description and are not intended to limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the overall structural framework of the present invention;

[0036] Figure 2 This is a schematic diagram of the event collection module of the present invention;

[0037] Figure 3 This is a schematic diagram of the event analysis module of the present invention;

[0038] Figure 4 This is a schematic diagram of the public opinion training module of the present invention;

[0039] Figure 5 It is a schematic diagram of the composition of the reversal prediction module of the present invention. DETAILED DESCRIPTION

[0040] The following is an explanation of the embodiments of the present invention through specific embodiments. Those skilled in the art can understand the advantages and effects of the present invention from the contents disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. In addition, the drawings of the present invention are only simple schematic illustrations and are not depicted according to actual sizes. It is stated in advance. The following embodiments will further explain the relevant technical contents of the present invention in detail, but the disclosed contents are not intended to limit the scope of protection of the present invention.

[0041] Embodiment 1: This embodiment provides a network public opinion reversal prediction system using AI technology, combined with Figure 1 , including event collection module, event analysis module, public opinion training module and reversal prediction module;

[0042] The event collection module is used to collect public opinion events on the Internet, the event analysis module is used to classify the collected public opinion events and extract feature information of the events, the public opinion training module is used to train the collected events, and the reversal prediction module is used to perform reversal prediction on real-time events;

[0043] The event collection module includes an information collection unit, an event sorting unit and a data storage unit. The information collection unit is used to collect public opinion information on the Internet social platform. The event sorting unit is used to sort the collected public opinion information into multiple event information. The data storage unit is used to store the sorted event information.

[0044] The event analysis module includes a feature extraction unit, an event classification unit and a sample storage unit, wherein the feature extraction unit is used to extract feature data from event development information, the event classification unit classifies events into reversal events and non-reversal events based on event comment information, and the sample storage unit stores feature data and corresponding event types as samples;

[0045] The public opinion training module includes a sample input unit, a feature processing unit and a comparison optimization unit. The sample input unit is used to obtain sample data from the sample storage unit. The feature processing unit is used to perform nonlinear change processing on the feature data in the sample data and output a judgment result. The comparison optimization unit compares the event type in the sample with the judgment result and optimizes the parameters in the feature processing unit according to the comparison result.

[0046] The reversal prediction module includes a real-time data access unit and a prediction execution unit, wherein the real-time data access unit is used to receive real-time event information and perform preprocessing, and the prediction execution unit is used to obtain parameter information in the public opinion training module and perform reversal prediction on the preprocessed real-time event information;

[0047] The feature extraction unit includes a node extraction processor and a feature conversion processor, wherein the node extraction processor is used to extract event node text from event development information, and the feature conversion processor is used to convert the extracted event node text into feature information;

[0048] The process of converting the feature information obtained by the feature conversion processor comprises the following steps:

[0049] S1. Obtain keywords in the node text and convert each keyword into a spatial coordinate as a key point;

[0050] S2, select a keyword as the basis of the event node, and the corresponding spatial coordinates of the keyword as the base point;

[0051] S3, calculate the vector with the base point as the vector starting point and the other key points as the vector end points, and obtain the key vector set;

[0052] S4. Calculate the detail index DT of the event node according to the following formula:

[0053] ;

[0054] Among them, n is the number of key points, n0 is the detail base, c is the bias value, a i is the i-th key vector;

[0055] S5, select two key points closest to the base point, and connect the two key points and the base point to obtain a triangular plane;

[0056] S6. Calculate the area S and unit normal vector of the triangular plane ;

[0057] The feature conversion processor converts the area S, the vector , the vector corresponding to the base point coordinates and detail index DT as the characteristic information of the event node;

[0058] The feature processing unit includes a change analysis processor and a judgment output processor, wherein the change analysis processor is used to analyze the changes of feature data of different event nodes, and the judgment output processor is used to output the judgment result;

[0059] The change analysis processor calculates S, and The change value Vg is obtained by processing:

[0060] ;

[0061] Among them, S(1) is the S value of the first feature data, S(2) is the S value of the second feature data, is the vector of the first feature data , is the vector of the second feature data , is the vector of the first feature data , is the vector of the second feature data , DT(1) is the DT value of the first feature data, DT(2) is the DT value of the second feature data, , , and is the model variation parameter;

[0062] The judgment output processor calculates the judgment value Q according to the following formula:

[0063] ;

[0064] Where M is the number of event nodes, Vg i represents the i-th change value, and Determine parameters for the model;

[0065] When the Q value is greater than 0, the event is judged as a reversal event, otherwise the event is judged as a non-reversal event;

[0066] The event classification unit includes a comment scoring processor, a timing arrangement processor and a reversal judgment processor. The comment scoring processor is used to perform biased scoring on each comment information. The timing arrangement processor classifies and arranges the event nodes targeted by each comment to obtain a comprehensive score for each event node. The reversal judgment processor judges the event as a reversal event or a non-reversal event according to the change in the comprehensive score of the event node, and marks the reversal node at the same time.

[0067] Embodiment 2: This embodiment includes all the contents of Embodiment 1, and provides a network public opinion reversal prediction system using AI technology, including an event collection module, an event analysis module, a public opinion training module and a reversal prediction module;

[0068] The event collection module is used to collect public opinion events on the Internet, the event analysis module is used to classify the collected public opinion events and extract feature information of the events, the public opinion training module is used to train the collected events, and the reversal prediction module is used to perform reversal prediction on real-time events;

[0069] Combination Figure 2 The event collection module includes an information collection unit, an event sorting unit and a data storage unit. The information collection unit is used to collect public opinion information on the Internet social platform. The event sorting unit is used to sort the collected public opinion information into multiple event information. The data storage unit is used to store the sorted event information.

[0070] Combination Figure 3 The event analysis module includes a feature extraction unit, an event classification unit and a sample storage unit. The feature extraction unit is used to extract feature data from event development information. The event classification unit divides events into reversal events and non-reversal events based on event comment information. The sample storage unit stores feature data and corresponding event types as samples.

[0071] Combination Figure 4 The public opinion training module includes a sample input unit, a feature processing unit and a comparison optimization unit. The sample input unit is used to obtain sample data from the sample storage unit. The feature processing unit is used to perform nonlinear change processing on the feature data in the sample data and output a judgment result. The comparison optimization unit compares the event type in the sample with the judgment result and optimizes the parameters in the feature processing unit according to the comparison result.

[0072] Combination Figure 5 The reversal prediction module includes a real-time data access unit and a prediction execution unit. The real-time data access unit is used to receive real-time event information and perform preprocessing. The prediction execution unit is used to obtain parameter information in the public opinion training module and perform reversal prediction on the preprocessed real-time event information.

[0073] The information collection unit includes a platform connection processor and a text acquisition processor, wherein the platform connection processor is used to access a social platform on the Internet, and the text acquisition processor is used to collect public opinion information on the social platform;

[0074] The event sorting unit includes a label classification processor, an attribute classification processor and an event summarization processor. The label classification processor divides the public opinion information into multiple event information sets according to the labels in the public opinion information. The attribute classification processor is used to divide the public opinion information in the same event information set into factual information and commentary information. The event summarization processor is used to summarize the factual information in the same event information set to obtain development information.

[0075] The data storage unit includes a partition management processor and an initial data register, wherein the partition management processor establishes a partition index table according to the classification of the event sorting unit, and the initial data register is used to store the specific content of the development information and the comment information;

[0076] The feature extraction unit includes a node extraction processor and a feature conversion processor, wherein the node extraction processor is used to extract event node text from event development information, and the feature conversion processor is used to convert the extracted event node text into feature information;

[0077] The event classification unit includes a comment scoring processor, a time sequence sorting processor and a reversal judgment processor. The comment scoring processor is used to perform bias scoring on each comment information. The time sequence sorting processor classifies and sorts the event nodes targeted by each comment to obtain a comprehensive score for each event node. The reversal judgment processor judges the event as a reversal event or a non-reversal event according to the change of the comprehensive score of the event node, and marks the reversal node at the same time.

[0078] The sample storage unit includes a reversal event interception processor and a sample data register, wherein the reversal event interception processor is used to intercept feature information before the reversal node from the reversal event as a reversal sample, and the sample data register is used to store the reversal sample and the non-reversal sample;

[0079] The sample input unit includes an input control processor and a training buffer register, wherein the input control processor is used to obtain sample data from the sample storage unit and control the processing of the sample data, and the training buffer register is used to temporarily store the sample data obtained from the sample storage unit;

[0080] The input control processor sends the feature data in the sample data to the feature processing unit, and sends the event type in the sample data to the comparison optimization unit;

[0081] The feature processing unit includes a change analysis processor and a judgment output processor, wherein the change analysis processor is used to analyze the changes of feature data of different event nodes, and the judgment output processor is used to output the judgment result;

[0082] The comparison and optimization unit includes a result comparison register and an optimization feedback processor, wherein the result comparison register is used to receive event type information and judgment result information of sample data and perform comparison processing, and the optimization feedback processor optimizes and adjusts parameters in the feature processing unit based on the comparison result;

[0083] The real-time data access unit includes an instant collection processor and an information conversion processor, wherein the instant collection processor is used to collect factual information that is hotly discussed on the Internet, and the information conversion processor is used to send the collected factual information to the event sorting unit and the feature extraction unit to convert the information into a time node and corresponding feature data;

[0084] The prediction execution unit includes a model building processor and a model execution processor, wherein the model building processor is used to obtain the optimized parameters in the feature processing unit to build a prediction model, and the model execution processor is used to process the feature data in the real-time data access unit and output a prediction result;

[0085] The process of converting the feature information obtained by the feature conversion processor comprises the following steps:

[0086] S1. Obtain keywords in the node text and convert each keyword into a spatial coordinate as a key point;

[0087] S2, select a keyword as the basis of the event node, and the corresponding spatial coordinates of the keyword as the base point;

[0088] S3, calculate the vector with the base point as the vector starting point and the other key points as the vector end points, and obtain the key vector set;

[0089] S4. Calculate the detail index DT of the event node according to the following formula:

[0090] ;

[0091] Among them, n is the number of key points, n0 is the detail base, c is the bias value, a i is the i-th key vector;

[0092] S5, select two key points closest to the base point, and connect the two key points and the base point to obtain a triangular plane;

[0093] S6. Calculate the area S and unit normal vector of the triangular plane ;

[0094] The feature conversion processor converts the area S, the vector , the vector corresponding to the base point coordinates and detail index DT as the characteristic information of the event node;

[0095] The review scoring processor extracts review bias words from the review text, converts them into corresponding bias values, and calculates the bias score P of each review text according to the following formula:

[0096] ;

[0097] Where g is the target coefficient, max(Vb) represents the bias value with the largest absolute value, and Vb represents the set of bias values ​​extracted from the comment text;

[0098] When the comment object of the comment text is the first subject, g is 1, and when the comment object of the comment text is the second subject, g is -1. The first subject and the second subject are two subjects with opposite positions in the event.

[0099] The timing arrangement processor divides the bias score into corresponding event nodes and calculates the comprehensive score Pz of each event node according to the following formula:

[0100] ;

[0101] Where N is the number of bias scores in the event node, P i represents the i-th bias score, N1 is the number of positive bias scores, and N2 is the number of negative bias scores;

[0102] The process of performing reversal judgment by the reversal judgment processor comprises the following steps:

[0103] S21, arranging the comprehensive scores according to the order of event nodes to obtain a score sequence;

[0104] S22, obtaining a comprehensive score from the scoring sequence in order, and if all comprehensive scores have been obtained, determining that the event is a non-reversal event, and ending the process;

[0105] S23. If the absolute value of the comprehensive score is greater than the original positive limit value or negative limit value, the original positive limit value or negative limit value is replaced;

[0106] S24, calculate the difference between the positive limit value and the negative limit value. If it is greater than the reversal threshold, determine that the event is a reversal event, and mark the event node corresponding to the currently obtained comprehensive score as a reversal node, and end the process. If it is not greater than the reversal threshold, return to step S22;

[0107] The change analysis processor calculates S, and The change value Vg is obtained by processing:

[0108] ;

[0109] Among them, S(1) is the S value of the first feature data, S(2) is the S value of the second feature data, is the vector of the first feature data , is the vector of the second feature data , is the vector of the first feature data , is the vector of the second feature data , DT(1) is the DT value of the first feature data, DT(2) is the DT value of the second feature data, , , and is the model variation parameter;

[0110] The judgment output processor calculates the judgment value Q according to the following formula:

[0111] ;

[0112] Where M is the number of event nodes, Vg i represents the i-th change value, and Determine parameters for the model;

[0113] When the Q value is greater than 0, the event is judged as a reversal event, and the larger the value, the higher the possibility of reversal, otherwise the event is judged as a non-reversal event;

[0114] After the result comparison register receives the event type and judgment type of the training group, it screens out the events with wrong judgments and calculates the judgment deviation D according to the following formula:

[0115] ;

[0116] Where Ne is the number of events with incorrect judgments, is the judgment value of the i-th wrongly judged event;

[0117] The optimization feedback processor adjusts the model change parameters and the model judgment parameters so that the adjusted judgment deviation is reduced by at least half, and then formally replaces the parameters and replaces the new training group for training;

[0118] The training set refers to a fixed number of sample data;

[0119] The prediction method of the model execution processor is consistent with the processing method in the feature processing unit, which will not be repeated here;

[0120] In step S1, the keywords can be converted into spatial coordinates, and the conversion method is described below;

[0121] A large number of keywords are collected and trained using word embedding technology to obtain the mapping between each keyword and the spatial coordinates, which are then sorted into a mapping table. In step S1, the conversion is completed directly through the mapping table;

[0122] The method of obtaining the mapping table is not limited to the word embedding technology, and the mapping table can also be obtained by other methods. This process does not belong to the work content of this system;

[0123] As an existing technology, word embedding technology is briefly described below using the GloVe model in word embedding technology:

[0124] Construct a vocabulary containing all words and initialize a word co-occurrence matrix X, where X_{ij} represents the number of times word i and word j appear together in the corpus;

[0125] The window size can be set as needed, usually 5 to 10 words;

[0126] Word embedding is learned by optimizing an objective function. The design of the objective function takes the following points into consideration:

[0127] Log co-occurrence probability: The probability of two words appearing together is expressed as the logarithm of the number of co-occurrences, the purpose is to smooth the difference between high-frequency words and low-frequency words;

[0128] Loss function: A weighted square loss function is defined to measure the error of word vectors in predicting co-occurrence probabilities;

[0129] The learned word vectors are the corresponding spatial coordinates;

[0130] The i appearing in the above text is an ordinal number used to indicate a sequence number.

[0131] The contents disclosed above are only preferred feasible embodiments of the present invention, and do not limit the protection scope of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention specification and drawings are included in the protection scope of the present invention. In addition, the elements therein can be updated as technology develops.

Claims

1. A network public opinion reversal prediction system using AI technology, characterized in that: It includes event collection module, event analysis module, public opinion training module and reversal prediction module; The event collection module is used to collect public opinion events on the Internet, the event analysis module is used to classify the collected public opinion events and extract feature information of the events, the public opinion training module is used to train the collected events, and the reversal prediction module is used to perform reversal prediction on real-time events; The event collection module includes an information collection unit, an event sorting unit and a data storage unit. The information collection unit is used to collect public opinion information on the Internet social platform. The event sorting unit is used to sort the collected public opinion information into multiple event information. The data storage unit is used to store the sorted event information. The event analysis module includes a feature extraction unit, an event classification unit and a sample storage unit, wherein the feature extraction unit is used to extract feature data from event development information, the event classification unit classifies events into reversal events and non-reversal events based on event comment information, and the sample storage unit stores feature data and corresponding event types as samples; The public opinion training module includes a sample input unit, a feature processing unit and a comparison optimization unit. The sample input unit is used to obtain sample data from the sample storage unit. The feature processing unit is used to perform nonlinear change processing on the feature data in the sample data and output a judgment result. The comparison optimization unit compares the event type in the sample with the judgment result and optimizes the parameters in the feature processing unit according to the comparison result. The reversal prediction module includes a real-time data access unit and a prediction execution unit, wherein the real-time data access unit is used to receive real-time event information and perform preprocessing, and the prediction execution unit is used to obtain parameter information in the public opinion training module and perform reversal prediction on the preprocessed real-time event information; wherein the feature extraction unit includes a node extraction processor and a feature conversion processor, wherein the node extraction processor is used to extract event node text from event development information, and the feature conversion processor is used to convert the extracted event node text into feature information; The process of converting the feature information obtained by the feature conversion processor comprises the following steps: S1. Obtain keywords in the node text and convert each keyword into a spatial coordinate as a key point; S2, select a keyword as the basis of the event node, and the corresponding spatial coordinates of the keyword as the base point; S3, calculate the vector with the base point as the vector starting point and the other key points as the vector end points, and obtain the key vector set; S4. Calculate the detail index DT of the event node according to the following formula: ； Among them, n is the number of key points, n0 is the detail base, c is the bias value, a i is the i-th key vector; S5, select two key points closest to the base point, and connect the two key points and the base point to obtain a triangular plane; S6. Calculate the area S and unit normal vector of the triangular plane ; The feature conversion processor converts the area S, the vector , the vector corresponding to the base point coordinates And the detail index DT is used as the characteristic information of the event node.

2. The network public opinion reversal prediction system using AI technology as claimed in claim 1, characterized in that: The feature processing unit includes a change analysis processor and a judgment output processor, wherein the change analysis processor is used to analyze the changes of feature data of different event nodes, and the judgment output processor is used to output the judgment result; The change analysis processor calculates S, and The change value Vg is obtained by processing: ； Among them, S(1) is the S value of the first feature data, S(2) is the S value of the second feature data, is the vector of the first feature data , is the vector of the second feature data , is the vector of the first feature data , is the vector of the second feature data , DT(1) is the DT value of the first feature data, DT(2) is the DT value of the second feature data, , , and is the model variation parameter.

3. The network public opinion reversal prediction system using AI technology as claimed in claim 2, characterized in that: The judgment output processor calculates the judgment value Q according to the following formula: ； Where M is the number of event nodes, Vg i represents the i-th change value, and Determine parameters for the model; When the Q value is greater than 0, the event is judged to be a reversal event, otherwise the event is judged to be a non-reversal event.

4. The network public opinion reversal prediction system using AI technology as claimed in claim 3, characterized in that: The event classification unit includes a comment scoring processor, a timing arrangement processor and a reversal judgment processor. The comment scoring processor is used to perform biased scoring on each comment information. The timing arrangement processor classifies and arranges the event nodes targeted by each comment to obtain a comprehensive score for each event node. The reversal judgment processor judges the event as a reversal event or a non-reversal event according to the change in the comprehensive score of the event node, and marks the reversal node at the same time.

Images