Social network public opinion monitoring method and system based on artificial intelligence

Abstract

The application discloses a kind of social network public opinion monitoring method and system based on artificial intelligence, comprising the following steps: obtaining public opinion data from social network platform and carrying out data preprocessing. The automatic extraction of hot events is carried out on public opinion data, and the density peak clustering algorithm is used to identify high-density areas and determine the heat of hot events. The BERT model is used for text sentiment classification, combined with bidirectional LSTM to extract context features, and the Softmax classifier is used to output sentiment tendency. For specific target events, use multi-head attention mechanism and convolutional neural network for specific target sentiment classification. The time series data is denoised and decomposed, and a plurality of basic prediction models are constructed and combined to generate a combined prediction model, to monitor and predict the development trend of public opinion in real time. The application provides accurate public opinion analysis and decision support for government and enterprises, which helps to reduce public opinion risk, maintain social stability and promote social construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a social network public opinion monitoring method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of the Internet and social media, social networks have become an important channel for people to access news, express opinions and participate in social affairs. According to the data of China Internet Network Information Center (CNNIC), the proportion of social media users accessing news information is as high as 90.7%. However, the openness and easy spread of social networks have made the spread of false information more common. Only 25.7% of users will verify the authenticity of the information before forwarding the news, and 60.3% of netizens will directly forward the false information without verification. This phenomenon not only promotes the spread of false news, but also to some extent affects the public opinion trend and social stability.

[0003] Social networks have become an important source of hot events. With the formation of the influence of communication, social media not only guides the follow-up reports of news websites and traditional media, but also promotes a larger wave of public opinion. To cope with this challenge, governments and enterprises need to establish effective public opinion monitoring and analysis mechanisms to identify and guide public sentiment in a timely manner and maintain social stability.

[0004] In this context, it is particularly important to use emerging technologies such as artificial intelligence, machine learning and big data to carry out sentiment analysis and trend prediction of social network hot events. Through in-depth analysis of public opinion data, it is possible to identify hot events of public concern, judge their emotional tendencies, and analyze the development context and transmission path of social emergencies.

[0005] Existing public opinion monitoring methods mainly rely on machine learning and deep learning technologies. Although these methods have improved the accuracy and efficiency of public opinion analysis to some extent, there are still deficiencies in data preprocessing, topic model coverage and automation. In addition, current sentiment analysis methods are often limited to text data, lacking comprehensive analysis of multi-modal data such as pictures and videos. SUMMARY

[0006] The present application provides a social network public opinion monitoring method and system based on artificial intelligence, which realizes the automatic extraction, sentiment classification and trend prediction of social network hot events by combining density peak clustering, BERT pre-training model, bidirectional LSTM, multi-attention CNN and other advanced algorithms.

[0007] In order to achieve the above application purposes, the technical solutions adopted by the present application are as follows:

[0008] A social network public opinion monitoring method based on artificial intelligence, comprising the following steps:

[0009] Step 1, data preprocessing: obtaining public opinion data from social network platforms, including text, pictures, comments and forwarding information; Chinese word segmentation of text data, removing stop words and invalid information; extracting key information, including characters, places, times and events; using BERT model for word vector encoding of text data.

[0010] Step 2, automatic extraction of hot events: clustering analysis of public opinion data, identifying the cluster center of high density area; determining the hot events in the social network by calculating the heat of each cluster; sorting the importance of hot events according to the heat index.

[0011] Step 3, sentiment classification: using BERT model to encode the preprocessed text into word vectors, and combining with the text theme features for feature fusion; using bidirectional LSTM model to extract the context features of the text and capture the information before and after the text; using Softmax classifier to classify the sentiment of the features output by LSTM, outputting positive, negative or neutral sentiment.

[0012] Step 4, specific target sentiment classification: obtaining the word vector of the text based on the BERT model; using multi-head attention mechanism to capture multiple important sentiment features in the text; inputting the attention mechanism results into the CNN model for local feature extraction; classifying the sentiment polarity of the specific target through the Softmax classifier.

[0013] Step 5, hot event trend prediction: denoising and decomposing time series data; building multiple basic prediction models to predict the decomposed modes; combining the results of multiple basic prediction models through weighted average method to form a combined prediction model;

[0014] Step 6, designing an online prediction module and combining the combined prediction model to monitor hot event data in real time and predict future trends.

[0015] Further, in the data preprocessing step, the named entity recognition (NER) technology is used to extract key entity information such as characters, places, times and events from the text.

[0016] Further, in the automatic extraction of hot events step, the public opinion data is clustered based on the density peak clustering algorithm.

[0017] Further, in the automatic extraction of hot events step, the hot events are determined by the microblog heat index.

[0018] Further, in the hot event trend prediction step, a variational mode decomposition algorithm is used to denoise and decompose the time series data, and a plurality of basic prediction models are constructed for the decomposed modes.

[0019] Further, in the hot event trend prediction step, a whale optimization algorithm based on dynamic back learning and golden sine is used for parameter optimization of the prediction model.

[0020] The application further discloses a social network public opinion monitoring system based on artificial intelligence, which can be used to implement the above-mentioned social network public opinion monitoring method based on artificial intelligence, and specifically comprises:

[0021] The data preprocessing module is used to obtain public opinion data from a social network platform, including text, pictures, comments and forwarding information. Chinese word segmentation is performed on the text data, and stop words and invalid information are removed. Key information is extracted, and word vector encoding is performed using a BERT model.

[0022] The hot event automatic extraction module performs cluster analysis on the public opinion data, identifies the cluster center of the high-density area, calculates the heat of each cluster, determines the hot event in the social network, and sorts the importance according to the heat index.

[0023] The sentiment classification module uses a BERT model to perform word vector encoding on the text, and combines text theme features for feature fusion. A bidirectional LSTM model is used to extract the context features of the text, a Softmax classifier is used for sentiment classification, and positive, negative or neutral sentiment tendencies are output.

[0024] The specific target sentiment classification module obtains the word vector of the text based on the BERT model, uses a multi-head attention mechanism to capture multiple important sentiment features in the text. The attention mechanism result is input into a CNN model for local feature extraction, and a Softmax classifier is used to classify the sentiment polarity of the specific target.

[0025] The hot event trend prediction module denoises and decomposes the time series data, constructs a plurality of basic prediction models to predict the decomposed modes, and combines the results of the plurality of basic prediction models by weighted average to form a combined prediction model.

[0026] The online prediction module combines the combined prediction model, monitors the hot event data in real time, and predicts the future trend according to the model to provide real-time support for decision-making.

[0027] The application further discloses a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned social network public opinion monitoring method based on artificial intelligence when executing the program.

[0028] The application also discloses a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the social network public opinion monitoring method based on artificial intelligence.

[0029] Compared with the prior art, the application has the following advantages:

[0030] 1. The method based on density peak clustering is used to realize automatic extraction of social network hot events, and the extraction efficiency and accuracy of public opinion information are significantly improved.

[0031] 2. The sentiment classification model based on BERT and bidirectional LSTM is constructed, which can accurately identify user sentiment tendency and improve the accuracy of sentiment classification.

[0032] 3. The specific target sentiment classification model based on BERT and multi-attention CNN is established, which improves the training efficiency and classification effect of the model and enhances the understanding ability of complex sentiment.

[0033] 4. The online combination optimization prediction system of hot event trends is designed, which realizes real-time monitoring and prediction of public opinion development trends and improves the prediction accuracy and real-time performance.

[0034] 5. The decision support for public opinion analysis and response is provided for governments and enterprises, which helps to reduce public opinion risks, maintain social stability and promote social harmony and development. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a flowchart of the social network public opinion monitoring method based on artificial intelligence according to an embodiment of the application. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below according to the drawings and examples.

[0037] As shown in Figure 1 The application provides a social network public opinion monitoring method based on artificial intelligence, which comprises the following steps:

[0038] Step 1: data preprocessing

[0039] Data collection: a large amount of public opinion data is collected from mainstream social network platforms (such as Weibo, WeChat, forums, etc.), including text, pictures, comments, forwarding and other information.

[0040] Chinese word segmentation: the collected social network text is subjected to Chinese word segmentation processing. Common tools such as Jieba word segmentation or THULAC are used to segment sentences and remove invalid information such as stop words and punctuation marks.

[0041] Named Entity Recognition (NER): Utilize NER techniques to identify key entities in the text, extracting information such as characters, locations, times, and events, providing support for subsequent hot event analysis.

[0042] Time information extraction: Extract time information from the text and associate it with events, ensuring that hot events can be sorted and analyzed according to the timeline, laying the foundation for trend prediction.

[0043] Word vectorization: Use the BERT pre-training model to encode the processed text data into word vectors. The BERT model captures the context information of the text through bidirectional Transformer, generating high-quality word vector representations for subsequent model feature learning.

[0044] Step 2: Automatic extraction of hot events

[0045] Density Peak Clustering Algorithm (DPC): Use the density peak clustering algorithm to cluster text data in social networks. This algorithm calculates the density of each data point, identifies density peak areas, selects high-density areas as clustering centers, and forms different clustering clusters, including:

[0046] For each data point, calculate the local density of the point. Local density can be determined by counting the number of points within a certain cutoff distance from the point, or using a Gaussian kernel function to smooth the density estimate. The cutoff distance is an important parameter that affects the granularity of the clustering results.

[0047] For each data point, identify the nearest point in the local density higher than it, and calculate the distance. This distance is used to measure the separation of the point from the higher density area. For the points with the highest density, their distance is defined as the maximum distance from all other points, ensuring that these points are identified as potential clustering centers.

[0048] According to the local density and distance, select those points with high local density and large distance from other high-density points as clustering centers. These points usually form obvious "peaks" in the two-dimensional space of density and distance, representing potential clustering centers.

[0049] Assign each data point to the cluster to which the nearest clustering center belongs. In this way, multiple clustering clusters are formed, each cluster usually corresponding to a potential hot event or topic.

[0050] Hot event determination: Analyze each clustering cluster and determine the corresponding hot event based on microblog heat (such as the number of retweets, comments, and likes) and other indicators, including:

[0051] Extract multiple heat indicators for each micro-blog post, such as the number of retweets, comments, and likes. Each indicator represents a dimension of the heat of the micro-blog, and weights can be set to comprehensively calculate the overall heat of each micro-blog.

[0052] Using the weighted summation method, multiply the value of each heat indicator by its corresponding weight, then add the results to calculate the comprehensive heat value of each micro-blog.

[0053] Summarize the heat of micro-blog posts in each cluster, and calculate the overall heat of the cluster. The average or total of all micro-blog heat in the cluster can be calculated, depending on the analysis goal and data characteristics. The average is suitable for comparing the average popularity of different clusters, and the total is suitable for comparing the total influence of each cluster.

[0054] Event importance ranking: Rank the identified hot events according to heat, spread, and other indicators to determine the influence of the event. Importance ranking can help quickly discover and analyze major public opinion events.

[0055] Step 3: Sentiment classification

[0056] BERT-based word vector generation: Pre-processed text data is encoded into word vectors through the BERT model, generating context-dependent high-dimensional vector representations.

[0057] Feature fusion: Fuse the word vectors generated by BERT with the topic features extracted from the text (such as entities recognized by NER, keywords, etc.) to form more informative feature representations, making it easier for the sentiment classification model to understand the text semantics.

[0058] Bidirectional LSTM model: Input the fused feature vectors into the bidirectional LSTM (BiLSTM) network, which captures the information before and after the text to enhance the understanding of complex emotional expressions.

[0059] Sentiment classification: Use the Softmax classifier to classify the features output by the LSTM network to obtain the sentiment of the text, with classification categories being positive, negative, or neutral.

[0060] Step 4: Specific target sentiment classification

[0061] BERT word vector acquisition: For sentiment classification of specific target events, first use the BERT model to encode the text and generate word vectors.

[0062] Multi-attention mechanism: Use a multi-head attention mechanism to capture multiple important sentiment features in the text. The multi-head attention mechanism allows the model to focus on multiple aspects of the text simultaneously, improving the accuracy of sentiment classification for specific targets.

[0063] Multi-Attention CNN Model: The results of multiple attention mechanisms are input into a Convolutional Neural Network (CNN), which captures local sentiment features in the text through local perception mechanisms and performs classification.

[0064] Sentiment Polarity Output: The output of the CNN model is processed using a Softmax classifier to obtain the sentiment polarity (positive, neutral, or negative) of a specific target event.

[0065] Step 5: Hotspot Event Trend Prediction

[0066] Time Series Data Acquisition: Collect time series data for each hotspot event, such as the spread speed, heat change, and forwarding volume of the event.

[0067] Variational Modal Decomposition (VMD): Use the Variational Modal Decomposition algorithm to denoise and decompose the time series data, decomposing the original time series into different frequency modes and eliminating noise interference.

[0068] Basic Prediction Model Construction: For the decomposed modes, construct multiple basic prediction models, including Support Vector Regression (SVR), Long Short-Term Memory Network (LSTM), and Gated Recurrent Unit (GRU).

[0069] Improved Whale Optimization Algorithm (WOA): Use the dynamic backpropagation learning and golden sine improved whale optimization algorithm to optimize the parameters of the basic prediction model, find the optimal parameter combination, and improve the prediction accuracy of the model.

[0070] Combined Prediction Model: Combine the results of multiple basic prediction models through weighted averaging or other methods to form a combined prediction model. The combined model can integrate the advantages of multiple models to improve overall prediction accuracy.

[0071] Online Prediction and Monitoring: Build an online prediction module to monitor the heat change and development trend of hotspot events in real time, and use the combined prediction model to predict future trends, providing real-time public opinion response support for governments and enterprises.

[0072] The structure of the online prediction module is as follows:

[0073] Data Collection Layer

[0074] Real-time Data Acquisition: Use API interfaces, web crawlers, or stream data processing tools to continuously collect time series data from social networks. These data include the spread speed, heat change, and forwarding volume of the event.

[0075] Data Storage and Management: Store the collected data in an efficient database (such as NoSQL database MongoDB or time series database InfluxDB) to support fast querying and processing.

[0076] Data preprocessing layer

[0077] Data cleaning: Remove missing values, outliers, and noise to ensure data quality and consistency.

[0078] Variational modal decomposition (VMD): Apply VMD algorithm to denoise and decompose time series data, decompose the original sequence into different frequency modes, and eliminate noise interference.

[0079] Model construction layer

[0080] Basic prediction model: For each decomposed mode, build multiple basic prediction models such as support vector regression (SVR), long short-term memory network (LSTM), and gated recurrent unit (GRU).

[0081] Model parameter optimization: Use dynamic backpropagation and golden sine improved whale optimization algorithm (WOA) to optimize the parameters of the basic prediction model, find the optimal parameter combination, and improve the prediction accuracy.

[0082] Model integration layer

[0083] Combined prediction model: Combine the prediction results of basic prediction models through weighted averaging or other integration methods to form a comprehensive prediction model. The combined model can take advantage of multiple models to improve overall prediction accuracy and robustness.

[0084] Prediction and monitoring layer

[0085] Real-time prediction: Use the combined prediction model to predict the future trend of hot events in real time. Prediction results can include short-term and long-term predictions to meet different application needs.

[0086] Monitoring and alarm: Set monitoring indicators and alarm thresholds, and trigger alarms to notify decision-makers or relevant personnel when the heat of hot events changes beyond the set range.

[0087] Visualization and user interaction layer

[0088] Data visualization: Dynamically display time series data, prediction results, and trend changes of hot events through charts, dashboards, and other forms to facilitate user understanding and analysis.

[0089] User interaction: Provide a user interface that allows users to customize monitoring indicators, prediction periods, and alarm thresholds to meet the needs of different users.

[0090] Construction ideas

[0091] Modular design: The online prediction module is divided into multiple independent functional layers, and data transmission and interaction between layers are carried out through interfaces. This modular design facilitates system expansion and maintenance.

[0092] Real-time and high efficiency: Real-time data collection and processing are achieved using stream data processing technology and efficient algorithms, ensuring that the prediction module can quickly respond to data changes and provide timely prediction results.

[0093] Algorithm integration and optimization: Combine multiple prediction algorithms to improve prediction accuracy through model integration. At the same time, use improved optimization algorithms to adjust model parameters to obtain the best prediction performance.

[0094] Through the above ideas and design, the online prediction module can realize real-time monitoring and trend prediction of hot events in social networks. This not only helps decision-makers understand the development of events in a timely manner, but also provides strong support for public opinion management and crisis response. The design of the module should fully consider the real-time, flexibility and security of the system to ensure efficient and reliable performance in actual application.

[0095] In another embodiment of the present application, a social network public opinion monitoring system based on artificial intelligence is provided, which can be used to implement the above-mentioned social network public opinion monitoring method based on artificial intelligence. Specifically, it includes:

[0096] Data preprocessing module: used to obtain public opinion data from social network platforms, including text, pictures, comments and forwarding information. Chinese word segmentation is performed on text data, and stop words and invalid information are removed. Key information is extracted, and BERT model is used for word vector encoding.

[0097] Hot event automatic extraction module: cluster analysis is performed on public opinion data, the cluster center of high density area is identified, the heat of each cluster is calculated, the hot event in social network is determined, and the importance is sorted according to the heat index.

[0098] Sentiment classification module: BERT model is used to encode the word vector of the text, and the feature fusion is carried out combined with the text theme feature. The bidirectional LSTM model is used to extract the context feature of the text, and the Softmax classifier is used for sentiment classification, and the positive, negative or neutral sentiment tendency is output.

[0099] Specific target sentiment classification module: based on BERT model to obtain the word vector of the text, using multi-head attention mechanism to capture multiple important sentiment features in the text. The attention mechanism result is input into the CNN model for local feature extraction, and the Softmax classifier is used for classification of the sentiment polarity of the specific target.

[0100] Hotspot event trend prediction module: denoising and decomposition of time series data, building multiple basic prediction models to predict the decomposed modes. The results of multiple basic prediction models are combined by weighted average method to form a combined prediction model.

[0101] Online prediction module: combining the combined prediction model, real-time monitoring of hotspot event data, and predicting future trends according to the model to provide real-time support for decision-making.

[0102] In another embodiment of the present application, a terminal device is provided, which includes a processor and a memory, the memory being used to store a computer program, the computer program including program instructions, and the processor being used to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions; the processor in the embodiments of the present application can be used for the operation of the social network public opinion monitoring method based on artificial intelligence.

[0103] In another embodiment of the present application, a storage medium is provided, specifically a computer readable storage medium (Memory), which is a memory device in a terminal device, used to store programs and data. It can be understood that the computer readable storage medium here can include the built-in storage medium in the terminal device, and of course can also include the expansion storage medium supported by the terminal device. The computer readable storage medium provides a storage space, which stores the operating system of the terminal. Moreover, in the storage space, one or more instructions suitable for being loaded and executed by the processor are also stored, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium here can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory.

[0104] The one or more instructions stored in the computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the method for monitoring public opinion of a social network based on artificial intelligence in the above embodiments; and the one or more instructions stored in the computer-readable storage medium are loaded and executed by the processor.

[0105] Those skilled in the art will appreciate that embodiments of the present application can be supplied as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) having computer-usable program code embodied therein.

[0106] The present application is described in reference to the flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams.

[0107] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams.

[0108] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in the flowcharts and / or block diagrams.

[0109] Those skilled in the art will appreciate that the embodiments described herein are presented for purposes of illustration and understanding of the methods of practicing the present application and are not intended to limit the scope of the present application in any way. Various modifications and alterations of this application can be made by those skilled in the art without departing from the scope and spirit of this application, which is measured by the appended claims and their equivalents.

Claims

1. A method for monitoring public opinion on a social network based on artificial intelligence, characterized in that, The method comprises the following steps: Step 1, data preprocessing: obtaining public opinion data from a social network platform, including text, pictures, comments and forwarding information; performing Chinese word segmentation on the text data, removing stop words and invalid information; extracting key information, including characters, locations, times and events; using a BERT model to encode the text data into word vectors; Step 2, automatic extraction of hot events: clustering analysis is performed on the public opinion data to identify the clustering centers of high-density areas; the heat of each cluster is calculated to determine the hot events in the social network; and the importance of the hot events is ranked according to the heat index; Step 3, sentiment classification: using a BERT model to encode the preprocessed text into word vectors, and combining the text theme features for feature fusion; Using a bidirectional LSTM model to extract the context features of the text and capture the information before and after the text; using a Softmax classifier to classify the features output by the LSTM and output positive, negative or neutral sentiment; Step 4, specific target sentiment classification: obtaining the word vectors of the text based on the BERT model; using a multi-head attention mechanism to capture multiple important sentiment features in the text; Inputting the attention mechanism results into a CNN model for local feature extraction; and classifying the sentiment polarity of the specific target through a Softmax classifier; Step 5, hot event trend prediction: denoising and decomposing the time series data; constructing multiple basic prediction models to predict the decomposed modes; combining the results of the multiple basic prediction models through weighted averaging to form a combined prediction model; Step 6, designing an online prediction module and combining the combined prediction model to monitor the hot event data in real time and predict future trends. 2.The social network public opinion monitoring method of claim 1, wherein: In the data preprocessing step, the named entity recognition technology is used to extract key entity information such as characters, locations, times and events from the text. 3.The social network public opinion monitoring method of claim 1, wherein: In the automatic extraction of hot events step, the public opinion data is clustered based on the density peak clustering algorithm.

4. The social network public opinion monitoring method of claim 1, wherein: In the automatic extraction of hot events step, the hot events are determined by the microblog heat index.

5. The social network public opinion monitoring method of claim 1, wherein: In the hot event trend prediction step, a variational mode decomposition algorithm is used to denoise and decompose the time series data, and multiple basic prediction models are constructed for the decomposed modes.

6. The social network public opinion monitoring method of claim 1, wherein: In the hot event trend prediction step, the whale optimization algorithm based on dynamic back learning and golden sine is used for parameter optimization of the prediction model.

7. An artificial intelligence-based social network public opinion monitoring system, characterized in that: The system can be used to implement the social network public opinion monitoring method of any one of claims 1 to 6, specifically comprising: A data preprocessing module: used to obtain public opinion data from a social network platform, including text, pictures, comments and forwarding information; perform Chinese word segmentation on the text data, remove stop words and invalid information, extract key information, and use a BERT model to encode word vectors; A hot event automatic extraction module: performing clustering analysis on the public opinion data to identify the clustering centers of high-density areas, calculating the heat of each cluster to determine the hot events in the social network, and ranking the importance according to the heat index; The emotion classification module: the BERT model is used for word vector coding of the text, and the text theme features are combined for feature fusion; the bidirectional LSTM model is used for extracting the context features of the text, and the Softmax classifier is used for emotion classification to output positive, negative or neutral emotional tendencies; The specific target emotion classification module: the BERT model is used to obtain the word vector of the text, and the multi-head attention mechanism is used to capture multiple important emotional features in the text; the attention mechanism result is input into the CNN model for local feature extraction, and the Softmax classifier is used for classifying the emotional polarity of the specific target; The hot event trend prediction module: the time series data is denoised and decomposed, a plurality of basic prediction models are constructed to predict the decomposed modes; the results of the plurality of basic prediction models are combined by weighted average to form a combined prediction model; The online prediction module: combined with the combined prediction model, the hot event data is monitored in real time, and the future trend is predicted according to the model to provide real-time support for decision-making.

8. A computer device, comprising: A computer program product, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the social network public opinion monitoring method of any one of claims 1-6 when executing the program.

9. A computer-readable storage medium, characterized in that: A computer program product, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the social network public opinion monitoring method of any one of claims 1-6 when executing the program.

Images