An analytical method for public opinion information content mining and dissemination monitoring

Abstract

The present application provides an analysis method for public opinion information content mining and dissemination monitoring, including: obtaining the last public opinion information; identifying the positive and negative aspects of the last public opinion event based on the last public opinion information; analyzing the factors that caused the last public opinion event; based on the factors that caused the last public opinion event, analyzing the degree of dissemination of the last public opinion event; based on the degree of dissemination of the last public opinion event, mining the self-media related to the report of the last public opinion event; predicting the attitude of the related self-media to the reappearance of the last public opinion event; and filing and authenticating the self-media that may cause a public opinion crisis based on the self-media attitude prediction results.

Description

[Technical field]

[0001] The present invention relates to the field of information technology, and in particular to an analysis method for public opinion information content mining and dissemination monitoring. [Background technology]

[0002] Public opinion information has the implicitness of words, and the corresponding public opinion events have positive and negative polarities. The occurrence of public opinion events is not always accidental, because some events will occur due to the occurrence of related events, and some will occur due to the occurrence of time periods. For example, the successful launch of the Shenzhou 13 manned spaceflight will bring a lot of public opinion information. However, for manned spaceflight, there must be landing after the launch. It can be predicted that the cycle of public opinion information will come after the planning cycle. At this time, the self-media that played a role in public opinion influence in the early stage will once again usher in propaganda with great influence. If the self-media itself has a negative impact after the last public opinion event is reported, its attitude towards the recurrence of the last public opinion event will also change greatly. Therefore, for the prediction of communication, the situation of the last public opinion event can be used to predict the recurrence of the last public opinion event. Some self-media will have bad remarks on the platform. How to conduct more stringent certification of self-media to prevent some self-media from making bad remarks and filing certification is a problem. [Summary of the invention]

[0003] The present invention provides an analysis method for public opinion information content mining and dissemination monitoring, which mainly includes:

[0004] Obtain the last public opinion information; identify the positive and negative aspects of the last public opinion event based on the last public opinion information; analyze the factors that led to the last public opinion event; analyze the degree of spread of the last public opinion event based on the factors that led to the last public opinion event; mine the self-media related to the last public opinion event report based on the degree of spread of the last public opinion event; predict the attitude of the related self-media towards the last public opinion event; and register and certify the self-media that may cause a public opinion crisis based on the predicted results of the self-media attitude;

[0005] Further optionally, the obtaining of the last public opinion information includes:

[0006] Obtaining the last public opinion information based on Internet public opinion information includes: obtaining preset conditions; the preset conditions include at least one of preset time, preset field, preset place and preset thing; based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information; based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information, including: according to the preset conditions, crawling at least one first information that is the same as the preset condition from an Internet database through a crawler; when at least one of the first information is one, determining the first information as the last public opinion information; or, when at least one of the first information is multiple, crawling the page views of multiple first information, and filtering out at least one second information with the highest page views from multiple first information; when at least one of the second information is one, determining the second information as the last public opinion information, or, when at least one of the second information is multiple, crawling the interaction volume of multiple second information, and determining the second information with the highest interaction volume among multiple second information as the last public opinion information; the interaction volume refers to the sum of the like volume and comment reply volume contained in the information.

[0007] Further optionally, identifying the positive or negative of the last public opinion event according to the last public opinion information includes:

[0008] The identification of the positive and negative of the last public opinion event refers to classifying the positive and negative polarity of the last public opinion information, and the method is as follows: first, feature extraction is performed on the last public opinion information to allow the computer system to distinguish the text content that expresses real subjective information; the feature extraction refers to the computer detecting the information text and consulting the standard vocabulary to screen out the paragraphs and sentences containing subjective information; the subjective information refers to words with emotional colors and positive and negative tendencies; then, based on the paragraphs and sentences containing subjective information, the opinions expressed in the paragraphs and sentences are further extracted, and one party's views on a specific topic are selected, including: topic extraction, opinion holder identification and statement selection; the topic extraction refers to extracting the specific aspects of the topic with comments and the expressed topic; the opinion holder identification refers to determining the person who holds the commentary opinion; the statement selection refers to identifying the opinions released by the opinion holder and removing the statements of other people; then, taking all historical public opinion events as samples, a single-classification SVM model is constructed based on the impact factor; the impact factor includes the extracted topic and the statement of the opinion holder; then, the single-classification SVM is activated using the activation function. The value interval of the distance from the sample of the intermediate output of the model to the center of the hypersphere is mapped to [0,1]; if the mapping result is equal to the first preset threshold, the polarity of the last public opinion event is neutral; if the mapping result is less than the first preset threshold, the polarity of the last public opinion event is opposite to the polarity of the historical public opinion event; otherwise, the polarity of the last public opinion event is the same as the polarity of the historical public opinion event.

[0009] Further optionally, the analysis of factors causing the last public opinion event includes:

[0010] According to the positive and negative aspects of the last public opinion event identified, the factors that caused the last public opinion event to occur are analyzed; the factors include associated events and associated time; the associated event refers to the event that caused the last public opinion event to occur; the associated time refers to the time period that caused the last public opinion event to occur; through the last public opinion event associated function, the last public opinion event associated degree is calculated, and the factors that caused it to occur are analyzed; including: the analysis of the associated events of the last public opinion event; the analysis of the associated time of the last public opinion event; the calculation of the last public opinion event associated degree according to the last public opinion event associated function;

[0011] The analysis of related events of the previous public opinion event specifically includes:

[0012] The associated events of the last public opinion event are analyzed through the public opinion associated event analysis system; the public opinion associated event analysis system includes a preprocessing module, a topology module and an expert module; the preprocessing module is used to preprocess the text of the last public opinion event; the text preprocessing refers to dividing the keywords of the text as sub-information, including place, thing and status; the topology module is used to classify the sub-information of the preprocessing module into different layers, and then perform topology on the sub-information of the same layer and the sub-information of different layers respectively, and generate multiple types of sub-information based on the information of the last public opinion event; the expert module crawls texts containing the same sub-information in the Internet big database according to the multiple types of sub-information generated by the topology module, takes the content corresponding to the text whose sum of views and interactions exceeds a second preset threshold as an associated event, and generates an analysis report.

[0013] The correlation time analysis of the last public opinion event specifically includes:

[0014] The associated time analysis is to analyze the time period; first, the data granularity is preset; the data granularity refers to the degree of data refinement of the previous public opinion event; the time interval is divided according to the data granularity, and a time interval distribution diagram is drawn, the horizontal axis is time, and the vertical axis is the associated event occurrence index. If the associated event occurs within the time interval, the vertical axis value of the time interval distribution diagram is 1, otherwise it is 0; if within the preset time period, the vertical axis value in the time interval distribution diagram is 1 within the preset time period, then the previous public opinion event is associated with the time period, and the horizontal axis value corresponding to the vertical axis value in the time interval distribution diagram is the associated time.

[0015] The calculation of the correlation degree of the last public opinion event based on the correlation function of the last public opinion event specifically includes:

[0016] Establish the correlation function of the last public opinion event, that is, R=Sr*W1+Tr*W2; where R represents the correlation degree of the last public opinion event; Sr represents the number of related events; Tr represents the number of related times; W1 and W2 represent the corresponding weights respectively, and W1+W2=1.

[0017] Further optionally, the analyzing the degree of propagation of the recurrence of the previous public opinion event based on the occurrence factors of the previous public opinion event includes:

[0018] The degree of spread is measured by the heat of the last public opinion event and the focus of the last public opinion event. The heat refers to the degree of popularity of the last public opinion event, reflecting the public's attention to the last public opinion event. The focus refers to the focus of public debate when the last public opinion event reappears. Based on the analyzed heat of the last public opinion event and the focus of the last public opinion event, a spread evaluation system for the reappearance of public opinion events is established, that is, SPR = (

[0019] HOT+FOC) / 2; wherein, SPR represents the evaluation index of the degree of propagation of the last public opinion event; HOT represents the heat evaluation index of the last public opinion event; if the heat of the last public opinion event exceeds the third preset threshold, HOT=1, otherwise HOT=0; FOC represents the focus evaluation index of the last public opinion event; if the probability of the keyword of the focus of the last public opinion event in the crawled big data vocabulary is greater than the fourth preset threshold, FOC=1, otherwise, FOC=0; Analyze the degree of propagation of the last public opinion event according to the BP neural network; determine the training sample and the test sample according to the public opinion event heat function and the focus prediction model, the training sample is used for neural network training, and the test sample is used to detect the relative error between the actual value and the predicted value; including: analyzing the heat of the last public opinion event based on the number of related events; calculating the heat of the last public opinion event according to the public opinion event heat function; predicting the focus of the last public opinion event;

[0020] The analysis of the popularity of the last public opinion event based on the number of related events specifically includes:

[0021] Based on the number of related events, the heat of the last public opinion event is analyzed; if there are more related events, the higher the total attention to the related events and the total attention to the last public opinion event, the higher the heat of the last public opinion event, and the higher the heat of the last public opinion event reappearance; the heat of the last public opinion event calculated according to the total attention to related events and the heat function of the public opinion event is used as the original data, and the ARIMA regression model is established: the change curve of the total attention to related events and the heat of the last public opinion event is plotted in chronological order; if the curve is not stable, the sequence corresponding to the curve is differentiated, and the difference line graph is drawn to determine the order d; then the autocorrelation function graph and the partial autocorrelation function graph of the difference line graph are continued to be drawn, and the order p and q of the model are determined according to the shape of the graph; finally, according to the ARIMA (p, d, q) output curve, the predicted value after moving back in time is read, which is the heat of the last public opinion event reappearance.

[0022] The calculation of the heat of the previous public opinion event according to the public opinion event heat function specifically includes:

[0023] Establish a public opinion event heat function, that is, Hpr = Sr * Ca1 + Ca2; where Hpr represents the heat of the last public opinion event; Sr represents the number of related events; Ca1 represents the total attention to the related event; the total attention to the related event refers to the sum of the number of views, likes, comments and reposts of the public on the related event in the statistical data of the event publishing system background. Ca2 represents the total attention to the last public opinion event; the total attention to the last public opinion event refers to the sum of the number of views, likes, comments and reposts of the public on the last public opinion event in the statistical data of the event publishing system background.

[0024] The focus of predicting the recurrence of the previous public opinion event specifically includes:

[0025] By crawling the Internet big data to obtain all the historical focuses, the focus prediction model is used to predict the focus of the last public opinion event. The content of the focus prediction model is as follows: take all the historical focuses as sample data, analyze them through the K-means clustering algorithm, and select multiple centroids according to the keywords of all the historical focuses; the keywords refer to words that appear in the big data with a frequency greater than the fifth preset threshold; then calculate the Euclidean distance from each sample to the centroid, and classify each sample to the nearest centroid; then take the average of the sample data of each category to find the new centroid of each category; finally, iterate the sample classification and find the new centroid until the clustering converges and the new and old centroids no longer change. The last public opinion event and the associated event are input as parameters, and the keyword type of the focus that is most similar to it is classified by the clustering algorithm as the algorithm output; finally, according to the keywords of the focus and the last public opinion event, analyze and obtain the focus of the last public opinion event.

[0026] Further optionally, mining the self-media related to the report of the last public opinion event according to the degree of dissemination of the reappearance of the last public opinion event includes:

[0027] Based on the degree of spread of the last public opinion event, the types and quantity of related self-media reporting on the last public opinion event are mined; the related self-media include: graphic self-media, video self-media, audio self-media and live self-media; the more related self-media reporting on the last public opinion event, the higher the popularity of the last public opinion event, and the greater the degree of spread of the last public opinion event; the association rule mining Apriori algorithm is used to mine the correlation between the degree of spread of the related self-media corresponding to the last public opinion event; the association rule algorithm process is as follows: through iterative input, the types and quantity of related self-media in the network big data platform are retrieved. All frequent item sets in the two factors, that is, the item sets whose support is not less than the set threshold; the support is the percentage of the number of times each factor item set appears in the big data platform to the total number of each item set; then the frequent item sets are used to construct rules that meet the confidence level; the confidence level is the percentage of the total item set of each factor to all item sets in the database; by comparing the support and confidence levels, the strength of the correlation with the degree of propagation of the last public opinion event reappearance is judged, and a network diagram of the degree of propagation of the last public opinion event reappearance is drawn according to the association rules; the network diagram of the degree of propagation of the last public opinion event reappearance indicates the relevant self-media reporting on the last public opinion event.

[0028] Further optionally, the predicting of the attitude of the relevant self-media towards the recurrence of the previous public opinion event includes:

[0029] Based on the deep learning method, a relevant self-media attitude prediction model is established to predict the attitude of relevant self-media towards the reappearance of the last public opinion event; first, a large amount of public opinion influence of relevant self-media is collected, including self-media popularity, self-media platform activity, self-media influence and the impact of the last public opinion event; the self-media popularity refers to the number of fans of the self-media, and the more fans, the higher the self-media popularity; the self-media platform activity includes the frequency and total number of releases of all public opinion events on the self-media platform; the self-media influence includes the number of likes, reposts and comments after the self-media reports all public opinion events; the impact of the last public opinion event refers to the positive or negative impact of the self-media itself on the last public opinion event; the positive impact is measured by the number of likes after the last public opinion event is reported, and the more likes, the greater the positive impact; the negative impact is measured by the number of reports after the last public opinion event is reported, and the more reports, the greater the negative impact; crawl the content data of the public opinion influence of the relevant self-media on the Internet big data platform through crawlers, Including the number of fans of the self-media, the frequency and total number of all public opinion events posted on the self-media platform, the number of likes, reposts and comments after the self-media reported all public opinion events, and the number of likes and reports after the last public opinion event was reported; the content data of the public opinion influence of the relevant self-media is divided into a training set and a test set; then the number of likes and reports after the last public opinion event was reported are pre-processed, and if the number of likes is greater than the number of reports and the value exceeds the sixth preset threshold, it is determined to be a positive influence as the pre-processing result; if the number of reports is greater than the number of likes and the value exceeds the seventh preset threshold, it is determined to be a negative influence as the pre-processing result; the self-media popularity, self-media platform activity, and self-media influence are used as features, and the pre-processing results are used as labels to train the relevant self-media attitude prediction neural network model, judge the attitude of the relevant self-media towards the recurrence of the last public opinion event, and record the judgment result in a table file in a preset format and return it to the system; finally, the data of the test set is used to continuously adjust the parameters of the model to improve the accuracy of the relevant self-media attitude prediction model.

[0030] According to the predicted results of the self-media attitude, the self-media that may cause a public opinion crisis will be registered and certified, including:

[0031] The public opinion crisis refers to the negative impact on society caused by the release of bad remarks by self-media; the self-media filing and certification includes a data processing module, a neural network model construction module and a self-media early warning generation module; according to the attitude prediction results of the self-media, the self-media that may cause the public opinion crisis is filed and certified, including: judging whether it is necessary to conduct bad speech certification based on the number of reports after the self-media reported the last public opinion event; if the number of reports is higher than the eighth preset threshold, it is determined that bad speech certification is required, and the last public opinion information reported by the self-media is certified as the bad speech; the content data of the last public opinion event is obtained through the data processing module, and the deep learning RNN algorithm is used to analyze the number of reports and the attitude prediction results of the relevant self-media for the recurrence of the public opinion event, find out the bad speech, and construct a bad speech vector; the neural network model construction module uses the bad speech vector to construct and train a convolutional neural network model; the self-media early warning generation module uses the trained convolutional neural model to issue an early warning, so as to realize the filing and certification of the self-media that may cause the public opinion crisis.

[0032] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0033] The present invention can identify the positive and negative aspects of the last public opinion event, analyze the related events and related time of the last public opinion event, and thus predict the heat and focus of the last public opinion event. At the same time, the self-media that reported the last public opinion event are mined and their attitudes towards the reappearance of the last public opinion event are predicted. According to the results of the predicted attitude, the self-media that may cause a public opinion crisis are warned, registered and certified, so as to maintain a good public opinion influence, punish bad speech, and maintain the green and healthy network platform.

Brief Description of the Drawings

[0034] Figure 1 A flowchart of an analysis method for public opinion information content mining and dissemination monitoring of the present invention;

[0035] Figure 2 It is a structural schematic diagram of a public opinion-related event analysis system of the present invention. [Specific implementation method]

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Figure 1 FIG. 1 is a flow chart of an analysis method for public opinion information content mining and dissemination monitoring of the present invention. Figure 1 As shown, the analysis method of public opinion information content mining and dissemination monitoring in this embodiment may specifically include:

[0038] Step 101, obtaining the last public opinion information.

[0039] Obtaining the last public opinion information based on Internet public opinion information includes: obtaining preset conditions; the preset conditions include at least one of preset time, preset field, preset place, and preset things; based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information. Based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information includes: crawling at least one first information that is the same as the preset condition from an Internet database through a crawler according to the preset conditions; when at least one of the first information is one, determining the first information as the last public opinion information; or, when at least one of the first information is multiple, crawling the page views of multiple first information, and filtering out at least one second information with the highest page views from multiple first information; when at least one of the second information is one, determining the second information as the last public opinion information, or, when at least one of the second information is multiple, crawling the interaction volume of multiple second information, and determining the second information with the highest interaction volume among multiple second information as the last public opinion information; the interaction volume refers to the sum of the like volume and comment reply volume contained in the information.

[0040] Step 102, based on the last public opinion information, identify the positive and negative aspects of the last public opinion event.

[0041] The identification of the positive and negative of the last public opinion event refers to classifying the positive and negative polarity of the last public opinion information, and the method is as follows: first, feature extraction is performed on the last public opinion information to allow the computer system to distinguish the text content that expresses real subjective information; the feature extraction refers to the computer detecting the information text and consulting the standard vocabulary to screen out the paragraphs and sentences containing subjective information; the subjective information refers to words with emotional colors and positive and negative tendencies; then, based on the paragraphs and sentences containing subjective information, the opinions expressed in the paragraphs and sentences are further extracted, and one party's views on a specific topic are selected, including: topic extraction, opinion holder identification and statement selection; the topic extraction refers to extracting the specific aspects of the topic with comments and the expressed topic; the opinion holder identification refers to determining the person who holds the commentary opinion; the statement selection refers to identifying the opinions released by the opinion holder and removing the statements of other people; then, taking all historical public opinion events as samples, a single-classification SVM model is constructed based on the impact factor; the impact factor includes the extracted topic and the statement of the opinion holder; then, the single-classification SVM is activated using the activation function. The value interval of the distance from the sample of the intermediate output of the model to the center of the hypersphere is mapped to [0,1]; if the mapping result is equal to the first preset threshold, the polarity of the last public opinion event is neutral; if the mapping result is less than the first preset threshold, the polarity of the last public opinion event is opposite to the polarity of the historical public opinion event; otherwise, the polarity of the last public opinion event is the same as the polarity of the historical public opinion event.

[0042] Step 103, analyzing the factors that caused the last public opinion event.

[0043] According to the positive and negative aspects of the last public opinion event identified, the factors that caused the last public opinion event are analyzed; the factors include associated events and associated time; the associated event refers to the event that caused the last public opinion event; the associated time refers to the time period that caused the last public opinion event; through the association function of the last public opinion event, the correlation degree of the last public opinion event is calculated to analyze the factors that caused it. For example, enter "March 21, 2022" in the administrator portal interface, and obtain the three pieces of information "Tencent invests in China Literature Group", "Ali invests in Cha Yan Yue Se", and "Quanzhou Wanda Plaza is officially put into use" through the public opinion calculation and analysis method; the views are 10w, 10w, and 2w respectively; then obtain the interaction volume of the two pieces of information "Tencent invests in China Literature Group" and "Ali invests in Cha Yan Yue Se", and find that the interaction volume is 1w and 6k respectively; finally, recommend and display the results on the administrator portal interface, and obtain the last public opinion information "Tencent invests in China Literature Group".

[0044] Analysis of related events related to the last public opinion event.

[0045] The associated events of the last public opinion event are analyzed through the public opinion associated event analysis system; the public opinion associated event analysis system includes a preprocessing module, a topology module, and an expert module; the preprocessing module is used to preprocess the text of the last public opinion event; the text preprocessing refers to dividing the keywords of the text as sub-information, including place, thing, and status; the topology module is used to classify the sub-information of the preprocessing module into different layers, and then perform topology on the sub-information of the same layer and the sub-information of different layers, and generate multiple types of sub-information based on the information of the last public opinion event; the expert module crawls texts containing the same sub-information in the Internet big database according to the multiple types of sub-information generated by the topology module, takes the content corresponding to the text whose sum of views and interactions exceeds a second preset threshold as an associated event, and generates an analysis report. For example, in historical public opinion events, the public opinion about the bankruptcy of listed companies was always negative; in an article, the computer detected the paragraph sentences corresponding to the two words "helpless" and "difficult"; further detection showed that the theme was the bankruptcy and reorganization of listed companies; the blogger's statement was that the financial and capital market sectors were not doing well, with a shortage of funds and insufficient supply; through the SVM model, the first preset threshold of the mapping result was 0.5; the positive and negative polarity of the last public opinion event was analyzed, and the mapping result was 0.9; it can be obtained that the last public opinion event was negative.

[0046] Analysis of the correlation time of the last public opinion event.

[0047] The correlation time analysis is to analyze the time period; first, the data granularity is preset; the data granularity refers to the degree of data refinement of the last public opinion event; the time interval is divided according to the data granularity, and the time interval distribution diagram is drawn. The horizontal axis is time, and the vertical axis is the occurrence index of the associated event. If the associated event occurs within the time interval, the vertical axis value of the time interval distribution diagram is 1, otherwise it is 0; if the vertical axis value appears in the preset time interval in the time interval distribution diagram within the preset time period, then the last public opinion event is associated with the time period, and the horizontal axis value corresponding to the vertical axis value in the time interval distribution diagram is the associated time. For example, the associated events of "car owners hotly discuss the increase in oil prices" are "increased demand for purchasing new energy vehicles" and "car owners choose to add compressed natural gas to cars as fuel"; there is no associated time; then the occurrence correlation of "car owners hotly discuss the increase in oil prices" is calculated to be 1.2 through the association function of the last public opinion event; it can be analyzed that the event occurred due to the associated event, and is not associated with time. The associated factors are "new energy" and "compressed natural gas".

[0048] According to the correlation function of the last public opinion event, calculate the correlation degree of the last public opinion event.

[0049] Establish the correlation function of the last public opinion event, that is, R=Sr*W1+Tr*W2; where R represents the correlation degree of the last public opinion event; Sr represents the number of related events; Tr represents the number of related time; W1 and W2 represent the corresponding weights respectively, and W1+W2=1. For example, a positive public opinion event has been obtained, and the content is "National Theatre of China creates the first 5G smart market"; through the public opinion related event analysis system, the four keywords "National Theatre of China", "first", "5G" and "smart market" are divided into four layers; the first layer topological relationship is "National Theatre of China" -> "China Unicom" -> "Huawei Technology"; the second layer topological relationship is "first" -> "integration of theaters and venues"; the third layer topological relationship is "5G" -> "smart theater"; the fourth layer topological relationship is "smart market" -> "rich experience" -> "widespread cultural dissemination"; then crawl the texts with more than 10,000 views and interactions to generate the final related event analysis report; there are two results, namely "National Theatre of China creates a 5G smart market" and "the rise of cloud-based cultural communication".

[0050] Step 104, based on the factors causing the last public opinion event, analyze the degree of spread of the recurrence of the last public opinion event.

[0051] The degree of spread is measured by the heat of the last public opinion event and the focus of the last public opinion event. The heat refers to the degree of popularity of the last public opinion event, reflecting the public's attention to the last public opinion event. The focus refers to the focus of public debate when the last public opinion event reappears. Based on the analyzed heat of the last public opinion event and the focus of the last public opinion event, a spread evaluation system for the reappearance of public opinion events is established, that is, SPR = (

[0052] HOT+FOC) / 2; wherein, SPR represents the evaluation index of the degree of propagation of the last public opinion event; HOT represents the heat evaluation index of the last public opinion event; if the heat of the last public opinion event exceeds the third preset threshold, HOT=1, otherwise HOT=0; FOC represents the focus evaluation index of the last public opinion event; if the probability of the keywords of the focus of the last public opinion event appearing in the crawled big data vocabulary is greater than the fourth preset threshold, FOC=1, otherwise, FOC=0; the degree of propagation of the last public opinion event is analyzed according to the BP neural network; the training sample and the test sample are determined according to the public opinion event heat function and the focus prediction model, the training sample is used for neural network training, and the test sample is used to detect the relative error between the actual value and the predicted value. For example, in the data granularity selection of the public opinion event "A citizen in Guangdong Province uses up 1.5L of peanut oil every week for cooking", there are two options: "category" and "brand"; and the citizen is less dependent on the brand of peanut oil; if the granularity of a single brand is relatively large, the citizen may buy other types of peanut oil between two purchases of one type of peanut oil, and it is often impossible to mine effective associations; but if at the granularity of the category, although the citizen has changed the brand, the demand for peanut oil products remains constant every week; the time interval is divided into one week, and by drawing a time interval distribution graph, it can be seen that the vertical axis value of every Saturday is 1; so the associated time is every Saturday.

[0053] Based on the number of related events, analyze the popularity of the recurrence of the previous public opinion event.

[0054] Based on the number of related events, the heat of the last public opinion event is analyzed; if there are more related events, the higher the total attention to the related events and the total attention to the last public opinion event, the higher the heat of the last public opinion event, and the higher the heat of the last public opinion event reappearance; the heat of the last public opinion event calculated according to the total attention to related events and the heat function of the public opinion event is used as the original data, and the ARIMA regression model is established: the change curve of the total attention to related events and the heat of the last public opinion event is plotted in chronological order; if the curve is not stable, the sequence corresponding to the curve is differentiated, and the difference line graph is drawn to determine the order d; then the autocorrelation function graph and the partial autocorrelation function graph of the difference line graph are continued to be drawn, and the order p and q of the model are determined according to the shape of the graph; finally, according to the ARIMA (p, d, q) output curve, the predicted value after the time is moved back is read, which is the heat of the last public opinion event reappearance. For example, the related events of "car owners hotly discuss the increase in oil prices" include "increased demand for purchasing new energy vehicles" and "car owners choose to use compressed natural gas as fuel for their cars"; there is no related time; so Sr = 2, Tr = 0; the weights of the corresponding correlation indexes are set to 0.6 and 0.4 respectively; then the correlation function of the last public opinion event is used to calculate that the occurrence correlation degree of "car owners hotly discuss the increase in oil prices" is 1.2.

[0055] According to the public opinion event heat function, calculate the heat of the previous public opinion event.

[0056] Establish a public opinion event heat function, that is, Hpr = Sr * Ca1 + Ca2; where Hpr represents the heat of the last public opinion event; Sr represents the number of related events; Ca1 represents the total attention to the related event; the total attention to the related event refers to the sum of the number of views, likes, comments, and reposts of the public on the related event in the statistical data of the event publishing system background. Ca2 represents the total attention to the last public opinion event; the total attention to the last public opinion event refers to the sum of the number of views, likes, comments, and reposts of the public on the last public opinion event in the statistical data of the event publishing system background. For example, the evaluation system of the degree of spread of the reappearance of public opinion events is a large range (0.70-1), a general range (0.40-0.69), and a small range (0-0.39); the input layer of the BP neural network includes two indicators: the heat of the last public opinion event reappearance, and the probability of the keywords of the focus of the last public opinion event reappearance in the crawled big data vocabulary, that is, n=2, and the output layer is the evaluation result of the degree of spread of the last public opinion event reappearance, that is, m=1, and the hidden layer is 2 layers. After the data is trained by the neural network, the values ​​of the degree of spread of the reappearance of the last public opinion events A, B, and C are 0.60, 0.82, and 0.25 respectively, which means that the last public opinion event A reappeared in a general range, the last public opinion event B reappeared in a large range, and the last public opinion event C reappeared only in a small range.

[0057] Predict the focus of the last public opinion event.

[0058] By crawling the Internet big data to obtain all the historical focuses, the focus prediction model is used to predict the focus of the last public opinion event. The content of the focus prediction model is as follows: take all the historical focuses as sample data, analyze them through the K-means clustering algorithm, and select multiple centroids according to the keywords of all the historical focuses; the keywords refer to words that appear in the big data with a frequency greater than the fifth preset threshold; then calculate the Euclidean distance from each sample to the centroid, and classify each sample to the nearest centroid; then take the average of the sample data of each category to find the new centroid of each category; finally, iterate the sample classification and find the new centroid until the clustering converges and the new and old centroids no longer change. The last public opinion event and the associated event are input as parameters, and the keyword type of the focus that is most similar to it is classified by the clustering algorithm as the algorithm output; finally, according to the keywords of the focus and the last public opinion event, analyze and obtain the focus of the last public opinion event. For example, "Tencent invests in China Literature Group" has three related events, and the total attention of the related events is 10w; so the popularity of the last public opinion event is 35w; through ARIMA regression model analysis, the original data curve shows an upward trend and is not smooth, so the sequence is processed by first-order difference; after processing, a first-order difference line graph is obtained and there is no obvious upward or downward trend, so d=1; then continue to draw the autocorrelation analysis graph and partial autocorrelation function graph of the first-order difference line graph to determine the model order; the result shows that the autocorrelation function sequence presents a sine wave shape, which is the characteristic of the AR(2) model, that is, p=2; only one of the partial autocorrelation function sequences is significantly non-zero, so it is determined that the time series is suitable for the second-order moving average model MA(1), that is, q=1; so the output result is ARIMA(2,1,1); draw a function graph based on the model parameter value, and find the predicted value of the future time point based on the function graph is 40w, that is, the predicted popularity of the last public opinion event is 40w.

[0059] Step 105, based on the degree of dissemination of the last public opinion event, mining the self-media related to the report of the last public opinion event.

[0060] Based on the degree of spread of the last public opinion event, the types and quantities of related self-media reporting on the last public opinion event are mined; the related self-media include: graphic self-media, video self-media, audio self-media and live self-media; the more related self-media reporting on the last public opinion event, the higher the popularity of the last public opinion event, and the greater the degree of spread of the last public opinion event; the association rule mining Apriori algorithm is used to mine the correlation between the degree of spread of the related self-media corresponding to the last public opinion event; the association rule algorithm process is as follows: through iterative input, all frequent item sets in the two factors of type and quantity of related self-media in the network big data platform are retrieved, that is, the item sets whose support is not less than the set threshold; the support is the percentage of the number of times each factor item set appears in the big data platform to the total number of each item set; then the frequent item sets are used to construct rules that meet the confidence level; the confidence level is the percentage of the total item set of each factor to all item sets in the database. By comparing the support and confidence, the strength of the correlation with the degree of spread of the last public opinion event is determined, and a correlation network diagram of the degree of spread of the last public opinion event is drawn according to the correlation rules; the correlation network diagram of the degree of spread of the last public opinion event indicates the related self-media reporting on the last public opinion event. For example, "Tencent invests in China Literature Group" has 3 related events. The data from the background of the publishing system shows that the total number of views, likes, comments, and reposts is 10w; that is, the total attention of the related events is 10w; and the total attention of the last public opinion event itself is 5w; calculated by the heat function of the public opinion event, the heat of the last public opinion event is 35w.

[0061] Step 106, predicting the attitude of the relevant self-media towards the reappearance of the previous public opinion event.

[0062] Based on the deep learning method, a relevant self-media attitude prediction model is established to predict the attitude of relevant self-media towards the reappearance of the last public opinion event. First, a large amount of public opinion influence of relevant self-media is collected, including self-media popularity, self-media platform activity, self-media influence and the impact of the last public opinion event; the self-media popularity refers to the number of fans of the self-media, and the more fans, the higher the self-media popularity; the self-media platform activity includes the frequency and total number of releases of all public opinion events on the self-media platform; the self-media influence includes the number of likes, reposts and comments after the self-media reports all public opinion events; the impact of the last public opinion event refers to the positive or negative impact of the self-media itself on the last public opinion event; the positive impact is measured by the number of likes after the last public opinion event is reported, and the more likes, the greater the positive impact; the negative impact is measured by the number of reports after the last public opinion event is reported, and the more reports, the greater the negative impact; crawl the content data of the public opinion influence of the relevant self-media on the Internet big data platform through crawlers, Including the number of fans of the self-media, the frequency and total number of all public opinion events posted on the self-media platform, the number of likes, reposts and comments after the self-media reported all public opinion events, and the number of likes and reports after the last public opinion event was reported; the content data of the public opinion influence of the relevant self-media is divided into a training set and a test set; then the number of likes and reports after the last public opinion event was reported are pre-processed, and if the number of likes is greater than the number of reports and the value exceeds the sixth preset threshold, it is determined to be a positive influence as the pre-processing result; if the number of reports is greater than the number of likes and the value exceeds the seventh preset threshold, it is determined to be a negative influence as the pre-processing result; the self-media popularity, self-media platform activity, and self-media influence are used as features, and the pre-processing results are used as labels to train the relevant self-media attitude prediction neural network model, judge the attitude of the relevant self-media towards the recurrence of the last public opinion event, and record the judgment result in a table file in a preset format and return it to the system; finally, the data of the test set is used to continuously adjust the parameters of the model to improve the accuracy of the relevant self-media attitude prediction model. For example, the related events of the last public opinion event "Tencent invested in China Literature Group" include "China Literature Group develops a new reading APP" and "novel reading becomes a habit of primary school students"; these three events are input into the focus prediction model, and the key word of focus is obtained through K-means clustering algorithm analysis as "reading"; finally, combined with the last public opinion event, it is predicted that the focus of the last public opinion event will be "the successful development of WeChat conversational novel extraction and reading software."

[0063] Step 107, based on the self-media attitude prediction results, the self-media that may cause a public opinion crisis is registered and authenticated.

[0064] The public opinion crisis refers to the negative impact on society caused by the release of bad remarks by self-media; the self-media filing and certification includes a data processing module, a neural network model construction module and a self-media early warning generation module; according to the attitude prediction results of the self-media, the self-media that may cause the public opinion crisis is filed and certified, including: judging whether it is necessary to conduct bad speech certification based on the number of reports after the self-media reported the last public opinion event; if the number of reports is higher than the eighth preset threshold, it is determined that bad speech certification is required, and the last public opinion information reported by the self-media is certified as the bad speech; the content data of the last public opinion event is obtained through the data processing module, and the deep learning RNN algorithm is used to analyze the number of reports and the attitude prediction results of the relevant self-media for the recurrence of the public opinion event, find out the bad speech, and construct a bad speech vector; the neural network model construction module uses the bad speech vector to construct and train a convolutional neural network model; the self-media early warning generation module uses the trained convolutional neural model to issue an early warning, so as to realize the filing and certification of the self-media that may cause the public opinion crisis. For example, the spread index value of "Tencent invests in China Literature Group" calculated by the spread evaluation system of the reappearance of public opinion events is 0.80, which means the spread range is relatively large. The Apriori algorithm is used to mine the correlation between the spread degrees of all relevant self-media corresponding to the reappearance of the previous public opinion event, among which the first association rule is "large range (

[0065] The second association rule is "larger range (0.70-0.80)=>WeChat official account, support=0.769, confidence=0.953"; the second association rule is "larger range (0.70-0.80)=>WeChat official account, support=0.746, confidence=0.921"; the third association rule is "larger range (0.70-0.80)=>TikTok, support=0.824, confidence=0.896", indicating that the last public opinion event with a large spread was likely to be spread on WeChat official account, Weibo and TikTok, that is, the spread degree of "larger range (0.70-0.80)" is associated with the three types of related self-media of "WeChat official account", "Weibo" and "TikTok". Summarizing multiple association rules and drawing a network diagram of the spread degree of the last public opinion event, it is shown that the types and number of related self-media on the network diagram of the spread degree of the last public opinion event are the related self-media that reported the last public opinion event "Tencent invested in China Literature Group".

[0066] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

[0067] The program for implementing the information control of the present invention can be written in one or more programming languages ​​or a combination thereof to form a computer program code for performing the operations of the present invention, wherein the programming language includes an object-oriented programming language such as Java, python, C++, and a conventional procedural programming language such as C or a similar programming language.

[0068] The program code may execute entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0069] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation.

[0070] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0071] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0072] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present invention.

[0073] The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes.

Claims

1. A method for analyzing public opinion information content mining and dissemination monitoring, characterized in that: The method comprises: Obtain the last public opinion information; identify the positive and negative aspects of the last public opinion event based on the last public opinion information; analyze the factors that caused the last public opinion event, the analysis of the factors that caused the last public opinion event specifically includes: analysis of related events of the last public opinion event, analysis of the related time of the last public opinion event, and calculation of the correlation of the last public opinion event based on the correlation function of the last public opinion event; based on the factors of the last public opinion event, analyze the degree of spread of the reappearance of the last public opinion event, the analysis of the degree of spread of the reappearance of the last public opinion event based on the factors of the last public opinion event specifically includes: analyzing the heat of the reappearance of the last public opinion event based on the number of related events, calculating the heat of the last public opinion event based on the heat function of the public opinion event, and predicting the focus of the reappearance of the last public opinion event; according to the degree of spread of the reappearance of the last public opinion event, mine the self-media related to the report of the last public opinion event; predict the attitude of the related self-media to the reappearance of the last public opinion event; according to the self-media attitude prediction results, register and certify the self-media that may cause a public opinion crisis; The factors for analyzing the occurrence of the last public opinion event include: According to the positive and negative aspects of the last public opinion event identified, the factors that caused the last public opinion event to occur are analyzed; the factors include associated events and associated time; the associated event refers to the event that caused the last public opinion event to occur; the associated time refers to the time period that caused the last public opinion event to occur; through the last public opinion event associated function, the last public opinion event associated degree is calculated, and the factors that caused it to occur are analyzed; including: the analysis of the associated events of the last public opinion event; the analysis of the associated time of the last public opinion event; the calculation of the last public opinion event associated degree according to the last public opinion event associated function; The analysis of related events of the previous public opinion event specifically includes: The associated events of the last public opinion event are analyzed by the public opinion associated event analysis system; the public opinion associated event analysis system includes a preprocessing module, a topology module and an expert module; the preprocessing module is used to preprocess the text of the last public opinion event; the text preprocessing refers to dividing the keywords of the text as sub-information, including place, thing and status; the topology module is used to classify the sub-information of the preprocessing module into different layers, and then perform topology on the sub-information of the same layer and the sub-information of different layers respectively, and generate multiple types of sub-information based on the information of the last public opinion event; the expert module crawls texts containing the same sub-information in the Internet big database according to the multiple types of sub-information generated by the topology module, takes the content corresponding to the text whose sum of page views and interaction volume exceeds the second preset threshold as an associated event, and generates an analysis report; The correlation time analysis of the last public opinion event specifically includes: The associated time analysis is to analyze the time period; first, the data granularity is preset; the data granularity refers to the degree of data refinement of the last public opinion event; the time interval is divided according to the data granularity, and a time interval distribution diagram is drawn, the horizontal axis is time, and the vertical axis is the associated event occurrence index. If the associated event occurs within the time interval, the vertical axis value of the time interval distribution diagram is 1, otherwise it is 0; if within the preset time period, the vertical axis value appears in the preset time interval in the time interval distribution diagram. 1, then the last public opinion event is associated with the time period, and the horizontal axis value corresponding to the vertical axis value in the time interval distribution diagram is the associated time; The calculation of the correlation degree of the last public opinion event based on the correlation function of the last public opinion event specifically includes: Establish the correlation function of the last public opinion event, that is, R = Sr*W1+Tr*W2; where R represents the correlation degree of the last public opinion event; Sr represents the number of related events; Tr represents the number of related time; W1 and W2 represent the corresponding weights respectively, and W1+W2=1; Wherein, the analysis of the degree of spread of the reappearance of the last public opinion event based on the factors causing the last public opinion event includes: The degree of dissemination is measured by the heat of the last reappearance of the public opinion event and the focus of the last reappearance of the public opinion event; the heat refers to the degree of heat of the last reappearance of the public opinion event, reflecting the public's attention to the last reappearance of the public opinion event; the focus refers to the focus of public debate when the last public opinion event reappeared; based on the analyzed heat of the last reappearance of the public opinion event and the focus of the last reappearance of the public opinion event, a dissemination degree evaluation system for the reappearance of the public opinion event is established, that is, SPR=(HOT+FOC) / 2; wherein SPR represents the evaluation index of the dissemination degree of the last reappearance of the public opinion event; HOT represents the heat evaluation index of the last reappearance of the public opinion event. If the heat of the last reappearance of the public opinion event exceeds the third preset threshold, then HOT=1 , otherwise HOT=0; FOC represents the focus evaluation index of the last public opinion event reappearance. If the probability of the keyword of the focus of the last public opinion event reappearance in the crawled big data vocabulary is greater than the fourth preset threshold, then FOC=1, otherwise, FOC=0; Analyze the spread degree of the last public opinion event reappearance according to the BP neural network; Determine the training sample and the test sample according to the public opinion event heat function and the focus prediction model, the training sample is used for neural network training, and the test sample is used to detect the relative error between the actual value and the predicted value; Including: Based on the number of related events, analyze the heat of the last public opinion event reappearance; According to the public opinion event heat function, calculate the heat of the last public opinion event; Predict the focus of the last public opinion event reappearance; The analysis of the popularity of the last public opinion event based on the number of related events specifically includes: Based on the number of related events, the heat of the last public opinion event reappearance is analyzed; if there are more related events, the higher the total attention of the related events and the total attention of the last public opinion event, the higher the heat of the last public opinion event, and the higher the heat of the last public opinion event reappearance; the heat of the last public opinion event calculated by the total attention of the related events and the heat function of the public opinion event is used as the original data, and the ARIMA regression model is established: the change curve of the total attention of the related events and the heat of the last public opinion event is drawn in chronological order; if the curve is not stable, the sequence corresponding to the curve is differentiated, and the difference line graph is drawn to determine the order d; then the autocorrelation function graph and the partial autocorrelation function graph of the difference line graph are continued to be drawn, and the order p and q of the model are determined according to the shape of the graph; finally, according to the ARIMA (p, d, q) output curve, the predicted value after the time is moved back is read, which is the heat of the last public opinion event reappearance; The calculation of the heat of the previous public opinion event according to the public opinion event heat function specifically includes: Establish a public opinion event heat function, that is, Hpr = Sr * Ca1 + Ca2; where Hpr represents the heat of the last public opinion event; Sr represents the number of related events; Ca1 represents the total attention to the related events; the total attention to the related events refers to the sum of the number of views, likes, comments and forwardings of the public on the related events in the statistical data of the event publishing system background; Ca2 represents the total attention to the last public opinion event; the total attention to the last public opinion event refers to the sum of the number of views, likes, comments and forwardings of the public on the last public opinion event in the statistical data of the event publishing system background; The focus of predicting the recurrence of the previous public opinion event specifically includes: By crawling the Internet big data to obtain all the historical focuses, the focus prediction model is used to predict the focus of the last public opinion event. The content of the focus prediction model is as follows: taking all the historical focuses as sample data, analyzing them through the K-means clustering algorithm, and selecting multiple centroids according to the keywords of all the historical focuses; the keywords refer to words that appear in the big data with a frequency greater than the fifth preset threshold; then calculating the Euclidean distance from each sample to the centroid, and classifying each sample to the nearest centroid; then taking the average of the sample data of each category to find the new centroid of each category; finally iteratively classifying samples and finding new centroids until the clustering converges and the new and old centroids no longer change; taking the last public opinion event and the related event as parameter inputs, and classifying them into the keyword type of the focus that is most similar to it through the clustering algorithm as the algorithm output; finally, according to the keywords of the focus and the last public opinion event, analyzing and obtaining the focus of the last public opinion event; Among them, the above-mentioned mining of self-media related to the reporting of the last public opinion event based on the degree of dissemination of the last public opinion event reappears includes: Based on the degree of spread of the last public opinion event, the types and quantity of related self-media reporting on the last public opinion event are mined; the related self-media include: graphic self-media, video self-media, audio self-media and live self-media; the more related self-media reporting on the last public opinion event, the higher the popularity of the last public opinion event, and the greater the degree of spread of the last public opinion event; the association rule mining Apriori algorithm is used to mine the correlation between the degree of spread of the related self-media corresponding to the last public opinion event; the association rule algorithm process is as follows: through iterative input, the types and quantity of related self-media in the network big data platform are retrieved. All frequent item sets in the two factors, that is, item sets whose support is not less than the set threshold; the support is the percentage of the number of times each factor item set appears in the big data platform to the total number of each item set; then the frequent item sets are used to construct rules that meet the confidence; the confidence is the percentage of the total item set of each factor to all item sets in the database; by comparing the support and confidence, the strength of the correlation with the degree of spread of the last public opinion event reappearance is judged, and a correlation network diagram of the degree of spread of the last public opinion event reappearance is drawn according to the association rules; the correlation network diagram of the degree of spread of the last public opinion event reappearance indicates the related self-media reporting on the last public opinion event; The prediction of the attitude of the relevant self-media towards the reappearance of the previous public opinion event includes: Based on the deep learning method, a relevant self-media attitude prediction model is established to predict the attitude of the relevant self-media towards the reappearance of the last public opinion event; first, a large amount of public opinion influence of the relevant self-media is collected, including self-media popularity, self-media platform activity, self-media influence and the impact of the last public opinion event; the self-media popularity refers to the number of fans of the self-media, and the more fans there are, the higher the self-media popularity; the self-media platform activity includes the frequency and total number of releases of all public opinion events on the self-media platform; the self-media influence includes the number of likes, reposts and comments after the self-media reports all public opinion events; the impact of the last public opinion event refers to the positive or negative impact of the self-media itself on the last public opinion event; the positive impact is measured by the number of likes after the last public opinion event is reported, and the more likes, the greater the positive impact; the negative impact is measured by the number of reports after the last public opinion event is reported, and the more reports, the greater the negative impact; crawl the content data of the public opinion influence of the relevant self-media on the Internet big data platform through crawlers , including the number of fans of the self-media, the frequency and total number of releases of all public opinion events on the self-media platform, the number of likes, reposts and comments after the self-media reported all public opinion events, and the number of likes and reports after the last public opinion event was reported; the content data of the public opinion influence of the relevant self-media is divided into a training set and a test set; then the number of likes and reports after the last public opinion event was reported are pre-processed, and if the number of likes is greater than the number of reports and the value exceeds the sixth preset threshold, it is determined to be a positive influence as the pre-processing result; if the number of reports is greater than the number of likes and the value exceeds the seventh preset threshold, it is determined to be a negative influence as the pre-processing result; the self-media popularity, the activity of the self-media platform, and the influence of the self-media are used as features, and the pre-processing results are used as labels to train the relevant self-media attitude prediction neural network model, judge the attitude of the relevant self-media towards the recurrence of the last public opinion event, and record the judgment result in a table file in a preset format and return it to the system; finally, the data of the test set is used to continuously adjust the parameters of the model to improve the accuracy of the relevant self-media attitude prediction model.

2. The method according to claim 1, wherein: The obtaining of the last public opinion information includes: Obtaining the last public opinion information based on Internet public opinion information includes: obtaining preset conditions; the preset conditions include at least one of preset time, preset field, preset place and preset thing; based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information; based on the preset conditions, adopting the public opinion calculation and analysis method to obtain the last public opinion information, including: according to the preset conditions, crawling at least one first information that is the same as the preset condition from an Internet database through a crawler; when at least one of the first information is one, determining the first information as the last public opinion information; or, when at least one of the first information is multiple, crawling the page views of multiple first information, and filtering out at least one second information with the highest page views from multiple first information; when at least one of the second information is one, determining the second information as the last public opinion information, or, when at least one of the second information is multiple, crawling the interaction volume of multiple second information, and determining the second information with the highest interaction volume among multiple second information as the last public opinion information; the interaction volume refers to the sum of the like volume and comment reply volume contained in the information.

3. The method according to claim 1, wherein: The method of identifying the positive and negative aspects of the last public opinion event based on the last public opinion information includes: The identification of the positive and negative of the last public opinion event refers to classifying the positive and negative polarity of the last public opinion information, and the method is as follows: first, feature extraction is performed on the last public opinion information, so that the computer system can distinguish the text content that expresses the real subjective information; the feature extraction refers to the computer detecting the information text and consulting the standard word library to filter out the paragraphs and sentences containing subjective information; the subjective information refers to words with emotional colors and positive and negative tendencies; Then, based on the paragraphs and sentences containing subjective information, the opinions expressed in the paragraphs and sentences are further extracted, and one party's opinions on a specific topic are selected, including: topic extraction, opinion holder identification and statement selection; the topic extraction refers to extracting the specific aspects of the topic with comments and the opinions held; the opinion holder identification refers to determining the person holding the comments; the statement selection refers to identifying the opinions issued by the opinion holder and removing the statements of other people; then, taking all historical public opinion events as samples, a single-classification SVM model is constructed based on the impact factor; the impact factor includes the extracted topic and the statement of the opinion holder; then, using the activation function, the value interval of the distance from the intermediate output sample of the single-classification SVM model to the center of the hypersphere is mapped to [0,1]; if the mapping result is equal to the first preset threshold, the polarity of the last public opinion event is neutral; if the mapping result is less than the first preset threshold, the polarity of the last public opinion event is opposite to the polarity of the historical public opinion event; otherwise, the polarity of the last public opinion event is the same as the polarity of the historical public opinion event.

4. The method according to claim 1, wherein: According to the predicted results of the self-media attitude, the self-media that may cause a public opinion crisis will be registered and certified, including: The public opinion crisis refers to the negative impact on society caused by the release of bad remarks by self-media; the self-media filing and certification includes a data processing module, a neural network model construction module and a self-media early warning generation module; according to the attitude prediction results of the self-media, the self-media that may cause the public opinion crisis is filed and certified, including: judging whether it is necessary to conduct bad speech certification based on the number of reports after the self-media reported the last public opinion event; if the number of reports is higher than the eighth preset threshold, it is determined that bad speech certification is required, and the last public opinion information reported by the self-media is certified as the bad speech; the content data of the last public opinion event is obtained through the data processing module, and the deep learning RNN algorithm is used to analyze the number of reports and the attitude prediction results of the relevant self-media for the reappearance of the public opinion event, find out the bad speech, and construct a bad speech vector; the neural network model construction module uses the bad speech vector to construct and train a convolutional neural network model; the self-media early warning generation module uses the trained convolutional neural model to issue an early warning, so as to realize the filing and certification of the self-media that may cause the public opinion crisis.

Images