Man-machine collaborative public opinion monitoring method and system based on reinforcement learning

Abstract

The invention discloses a man-machine collaborative public opinion monitoring method and system based on reinforcement learning. The system comprises a allusion, television and harmonic reference analysis module, a public opinion data processing module, a model recognition module, a public opinion risk assessment module, a public opinion propagation analysis module, a learning optimization module and a deployment configuration module. The public opinion data processing module is used for collecting, cleaning, sorting, storing and preprocessing public opinion related data; the public opinion data storage module is used for managing and storing a large amount of public opinion data; through a reinforcement learning algorithm, the model can adapt to different public opinion environments, through repeated annotation and feedback of public opinions identification experts, the expert ability is solidified, dependence on manual annotation of the experts is reduced, through continuous iterative optimization, the model performance can be continuously improved, the model adapts to changes of the public opinion environments, and the method is suitable for popularization and application. Modern technical means such as deep pre-training, reinforcement learning and reward models are adopted, and innovation and breakthrough of the public opinion monitoring technology are achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and specifically to a human-machine collaborative public opinion monitoring method and system based on reinforcement learning. Background Art

[0002] In the field of public opinion monitoring, how to achieve accurate judgment of public opinion has always been the core content. There are currently two main ways to judge public opinion. One is the traditional one, which uses regular expressions to set keywords to filter public opinion texts, and the filtered public opinion is further judged manually; the other is the latest one, which is based on NLP technology and uses manual labeling of texts to train public opinion judgment models.

[0003] The existing public opinion monitoring system has the following shortcomings: 1. Insufficient accuracy: Traditional machine learning models may not be able to accurately capture the deep semantics of the text, resulting in insufficient accuracy in public opinion analysis.

[0004] 2. Poor adaptability: Fixed thresholds and rules are difficult to adapt to the changing public opinion environment, resulting in poor adaptability of the model.

[0005] 3. Reliance on expert experience: Existing public opinion monitoring systems rely heavily on expert experience and manual annotation, which is not only inefficient but also difficult to scale. To this end, we propose a human-machine collaborative public opinion monitoring method and system based on reinforcement learning. Summary of the invention

[0006] The purpose of the present invention is to provide a human-machine collaborative public opinion monitoring method and system based on reinforcement learning to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a human-machine collaborative public opinion monitoring system based on reinforcement learning, comprising an allusion, television and homophonic reference analysis module, a public opinion data processing module, a model recognition module, a public opinion risk assessment module, a public opinion propagation analysis module, a learning optimization module and a deployment configuration module; The public opinion data processing module is used to collect, clean, organize, store and pre-process public opinion related data; The public opinion data storage module is used to manage and store a large amount of public opinion data; The model identification module is responsible for extracting useful sentiment information, hot topics and potential risks from public opinion data; The public opinion risk assessment module is used to predict and warn of risks in public opinion data and assess the potential influence of public opinion; The public opinion dissemination analysis module is used to analyze the dissemination path, dissemination speed and audience range of public opinion information; The learning optimization module is responsible for dynamically adjusting the algorithm model and parameters in the public opinion monitoring system through user feedback and the input of new data; The deployment configuration module is used to provide configuration, scheduling management and monitoring functions of the deployment environment.

[0008] Preferably, the public opinion data processing module includes a data acquisition unit and a data processing unit; The data acquisition unit is used to acquire multi-dimensional public opinion data from the public opinion crawling system and extract the image and video content into text; The data processing unit uses natural language processing technology to extract and clean multidimensional data.

[0009] Preferably, the public opinion data storage module includes a data storage unit and a data backup and recovery unit; The data storage unit is used to adopt a distributed database management system to ensure rapid access to massive data; The data backup and recovery unit is used to regularly back up public opinion data to ensure rapid recovery in the event of data loss or system failure.

[0010] Preferably, the model identification module includes a model identification unit and an expert discrimination unit; The model recognition unit is used to use the trained large model to recognize and mark the public opinion data, and push the marked results to the public opinion analysis experts; The expert judgment unit is used by public opinion experts to judge the results of model labeling, mark whether the model labeling is correct, and realize human-machine collaborative judgment operations.

[0011] Preferably, the public opinion risk assessment module includes a risk index calculation unit and an anomaly detection unit; The risk index calculation unit is used to assign a risk index to each piece of public opinion data, and to measure the influence and dissemination potential of public opinion by obtaining the reading volume, forwarding volume, number of comments and the average number of the above indicators within the release time. The formula is: Influence = w1*number of readings + w2*number of reposts + w3*number of comments; Communication potential = w4*number of readings / posting time + w5*number of reposts / posting time + w6*number of comments / posting time; Risk index = influence + communication potential; weights can be customized and adjusted, and average weights are generally used; The anomaly detection unit is used to detect abnormal fluctuations or emergencies in public opinion data and issue early warnings in a timely manner.

[0012] Preferably, the public opinion propagation analysis module includes a propagation path analysis unit and a propagation speed prediction unit; The propagation path analysis unit is used to track the propagation path of public opinion information, identify the main propagation nodes and key figures, extract the common paths of public opinion propagation based on the historical data of information flow, and identify the most active users and platforms in the information propagation process; The propagation speed prediction unit is used to predict the propagation speed and development trend of public opinion information in the social network.

[0013] Preferably, the learning optimization module includes a learning unit and an optimization unit; The learning unit is used for a model trained for the first time using a large model fine-tuning technique based on reinforcement learning; The optimization unit is used to collect expert annotation results, perform automatic optimization based on the results, and calculate the recognition accuracy of the optimized model.

[0014] Preferably, the deployment configuration module includes a model deployment unit and a configuration learning unit; The model deployment unit is used for optimized rapid deployment and new types of configuration learning; The configuration learning unit is used to manually configure learning strategies for different public opinion types to improve the adaptability and accuracy of the model.

[0015] As a preferred method, the large model fine-tuning technology based on reinforcement learning is a process of further fine-tuning the already trained large models GPT and BERT using reinforcement learning algorithms. Unlike traditional supervised learning methods, reinforcement learning learns strategies through interaction with the environment, and the ultimate goal is to optimize model behavior through reward signals.

[0016] A monitoring method of a human-machine collaborative public opinion monitoring system based on reinforcement learning comprises the following steps: Step 1: Data acquisition: Use web crawler technology to crawl public opinion data in various formats from the Internet in real time, and perform preliminary extraction on the crawled data to identify the text content in pictures and videos; Step 2: Data processing: cleaning the acquired multi-dimensional data, analyzing the text content extracted from the pictures and videos, and identifying key topics and emotional tendencies; Step 3: Model identification: Use the pre-trained large model to identify and preliminarily classify the data, and push the model identification results to public opinion analysis experts for further manual review and analysis; Step 4: Expert judgment and feedback: Public opinion experts manually judge the results of the model's automatic labeling to determine whether they are accurate. Experts can make fine adjustments and corrections based on the actual context of the data and provide feedback to the model. Through the expert feedback mechanism, experts' opinions and corrections on the model's labeling results are collected; Step 5: Learning and optimization: Use reinforcement learning algorithms to fine-tune the model based on public opinion data and expert feedback. By establishing an automated optimization system, input the results of expert annotations into the system to automatically update and adjust model parameters. After the optimization is completed, perform model verification to check its performance on different public opinion data sets. Step 6: Deployment and configuration: build a model management platform to achieve rapid deployment and version control of the model, adaptively learn the model according to new public opinion data types or trends, deploy a monitoring system, and conduct real-time analysis and early warning of public opinion data; Step 7: System feedback and continuous improvement. After deployment, the model will continue to optimize itself based on new data and expert feedback to maintain the efficiency and accuracy of the model, regularly evaluate the monitoring effect of the public opinion monitoring system, collect feedback from users and experts, and make improvements and adjustments.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Improve accuracy: The present invention uses a large model to more accurately understand and analyze text data, improve the accuracy of public opinion monitoring, and combine the allusion, TV citation and homophonic analysis modules to provide a powerful technical support for public opinion monitoring and processing. It can not only capture public opinion dynamics in real time, but also deeply explore the culture, emotions and communication trends in the information.

[0018] Enhanced adaptability: The present invention uses a reinforcement learning algorithm to enable the model to adapt to different public opinion environments, thereby enhancing the adaptability of the model.

[0019] Solidifying expert capabilities: Through repeated labeling and feedback from public opinion recognition experts, the present invention realizes the solidification of expert capabilities and reduces the reliance on manual labeling by experts.

[0020] Continuous optimization: Through continuous iterative optimization, the present invention can continuously improve model performance and adapt to changes in the public opinion environment.

[0021] The present invention adopts technical means different from the existing technologies, such as deep pre-training, reinforcement learning, reward models, etc. These technical means work together to achieve innovation and breakthroughs in public opinion monitoring technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is the module principle diagram of the present invention; Figure 2 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] See also Figure 1 , the present invention provides a technical solution: a human-machine collaborative public opinion monitoring system based on reinforcement learning, including an allusion, television and homophonic reference analysis module, a public opinion data processing module, a model recognition module, a public opinion risk assessment module, a public opinion propagation analysis module, a learning optimization module and a deployment configuration module; The allusion, TV, and homophonic reference analysis modules help the public opinion monitoring system better identify potential public opinion crises. The system collects real-time data from multiple channels such as news websites, social platforms (such as Weibo, Zhihu, Douyin, etc.), forums, blogs, and TV news through crawlers and API technology. The captured data is deduplicated, denoised, and formatted to remove noise data and retain high-quality information. Use Transformer-based pre-trained models (such as GPT-4, BERT, etc.) for sentiment analysis to capture user emotions (such as anger, sadness, fear, etc.) and the intensity of emotions. Use clustering and classification techniques to identify current public opinion hotspots, analyze the public opinion trends of various topics, and provide trend forecasts for decision makers. The system builds a huge allusion library, including allusions such as idioms, historical stories, and literary works in traditional Chinese culture. Use a large language model to analyze allusions that appear in public opinion and judge their meaning based on the context. For example, a certain allusion may be used to describe the deviation from tradition of an event, or to express a certain metaphorical emotion, affecting the direction of public opinion. Identify the cultural meaning and historical background behind the allusion to provide a deeper interpretation for sentiment analysis. For example, the use of the allusion of "empty city plan" may imply a deliberate emptiness in a certain strategy, and may have some kind of derogatory or suggestive meaning. Through deep learning models, voice recognition, subtitle analysis and other processing are performed on multimedia content such as TV programs, news reports, and online videos. This module can capture high-frequency speeches, quotations and potential cultural backgrounds in TV programs in real time. Analyze the possible public opinion guidance in TV programs and identify the possible impact on public opinion. For example, discussions on certain TV programs may trigger emotional fluctuations or social emotions in a certain group, thereby affecting the direction of public opinion. When a TV program attracts widespread attention, the system can automatically detect and track the dissemination path of the content in the program and conduct public opinion guidance analysis. Combining phonetics, dialectology and Internet terms, the system needs to establish a database for homophonic words. Considering the language characteristics of Chinese, many public opinions are spoofed or punned through homophony during the dissemination process, which often has strong emotional colors and dissemination effects. Through deep learning-based language models (such as BERT, RoBERTa, etc.), homophonic phenomena in texts are detected and the implicit meanings behind words are identified. For example, some words may cause misunderstandings or produce humorous effects through homophony, or they may become carriers of offensive language. Track and analyze the public opinion trends triggered by homophones to determine whether they are diffusible, spoof-like, or socially sensitive, so as to assess the risk of their spread.

[0025] The public opinion data processing module is used to collect, clean, organize, store and pre-process public opinion related data. The public opinion data processing module includes a data acquisition unit and a data processing unit. The data acquisition unit is used to obtain multi-dimensional public opinion data from the public opinion crawling system, including text, pictures, videos and other formats, and extract the picture and video content into text, using web crawler technology to crawl public opinion data on the Internet in real time; the data processing unit is used to use natural language processing technology to extract and clean multi-dimensional data, such as text content in pictures and videos.

[0026] The public opinion data storage module is used to manage and store a large amount of public opinion data. The public opinion data storage module includes a data storage unit and a data backup and recovery unit. The data storage unit is used to use a distributed database management system to ensure rapid access to massive data; the data backup and recovery unit is used to regularly back up public opinion data to ensure rapid recovery in the event of data loss or system failure.

[0027] The model recognition module is responsible for extracting useful sentiment information, hot topics and potential risks from public opinion data. The model recognition module includes a model recognition unit and an expert discrimination unit. The model recognition unit is used to use the trained large model to identify and label the public opinion data, and push the labeled results to the public opinion analysis experts, deploy the pre-trained large model, perform preliminary classification and labeling on the data, and use machine learning technology to improve the accuracy of recognition; the expert discrimination unit is used for public opinion experts to judge the results of model labeling, mark whether the model labeling is correct, and establish an expert feedback mechanism. Experts can review and correct the model labeling results through the interface to improve the accuracy of the model.

[0028] The public opinion risk assessment module is used to predict and warn of public opinion data and assess the potential influence of public opinion. The public opinion risk assessment module includes a risk index calculation unit and an anomaly detection unit. The risk index calculation unit is used to assign a risk index to each piece of public opinion data, and to measure the influence and dissemination potential of public opinion by obtaining the number of readings, forwardings, and comments of public opinion and the average number of the above indicators within the release time. The formula is: Influence = w1*number of readings + w2*number of reposts + w3*number of comments; Communication potential = w4*number of readings / posting time + w5*number of reposts / posting time + w6*number of comments / posting time; Risk index = influence + communication potential; weights can be customized and adjusted, and average weights are generally used; the anomaly detection unit is used to detect abnormal fluctuations or emergencies in public opinion data and issue early warnings in a timely manner.

[0029] The public opinion propagation analysis module is used to analyze the propagation path, propagation speed and audience range of public opinion information. The public opinion propagation analysis module includes a propagation path analysis unit and a propagation speed prediction unit. The propagation path analysis unit is used to track the propagation path of public opinion information, identify the main propagation nodes and key figures, and use data mining technology to extract common paths of public opinion propagation based on historical data of information flow, and identify the most active users and platforms in the process of information propagation. The calculation of the propagation path depends on the graph traversal algorithm and path length calculation. Shortest path: If you need to find the shortest propagation path of information from the source node to the target node, use the Dijkstra algorithm. The shortest path length d(u, v) formula is: , in, represents all paths from node u to node v, It is the edge The weight of (may be propagation time, propagation distance, etc.); Identification of propagation paths: Using the depth-first search (DFS) of the graph, for a source node s of information propagation, the information propagation path Ps can be expressed as: , in, It is the node through which information propagates; The node centrality measurement is used to identify the key nodes in the propagation process. The degree centrality is used for centrality measurement. The formula is: , Among them, deg(v) is the degree of node v, which indicates the number of nodes connected to it; When calculating the influence of a node, the adjacency matrix A and the eigenvalue are used to measure: , Where C is the centrality vector of the node and λ is the maximum eigenvalue; The propagation speed prediction unit is used to predict the propagation speed and development trend of public opinion information in social networks.

[0030] The learning optimization module is responsible for dynamically adjusting the algorithm model and parameters in the public opinion monitoring system through user feedback and the input of new data. The learning optimization module includes a learning unit and an optimization unit. The learning unit is used for the first time to train the model using the large model fine-tuning technology based on reinforcement learning. It adopts the reinforcement learning algorithm to learn the optimal strategy through interaction with the environment and fine-tune the model to adapt to the specific public opinion monitoring task; the optimization unit is used to collect expert annotation results, automatically optimize according to the results, and calculate the recognition accuracy of the optimized model, develop an automated feedback processing system, and use the expert annotation results for further training and optimization of the model to achieve self-learning and evolution of the model. The large model fine-tuning technology based on reinforcement learning is the process of further fine-tuning the already trained large models GPT and BERT using the reinforcement learning algorithm. Unlike traditional supervised learning methods, reinforcement learning learns strategies through interaction with the environment, and the ultimate goal is to optimize model behavior through reward signals.

[0031] The deployment configuration module is used to provide configuration, scheduling management and monitoring functions for the deployment environment. The deployment configuration module includes a model deployment unit and a configuration learning unit. The model deployment unit is used for optimized rapid deployment and new types of configuration learning, building a model management platform to achieve rapid deployment and version management of the model, while supporting rapid learning and adaptation of new types of public opinion data; the configuration learning unit is used to manually configure learning strategies for different types of public opinion to improve the adaptability and accuracy of the model, and develop a flexible configuration interface that allows users to customize the model's learning strategies and parameters according to different public opinion scenarios.

[0032] See also Figure 2 , a monitoring method of a human-machine collaborative public opinion monitoring system based on reinforcement learning, comprising the following steps: Step 1: Data acquisition: Use web crawler technology to crawl public opinion data in various formats from the Internet in real time, and perform preliminary extraction on the crawled data to identify the text content in pictures and videos; Step 2: Data processing: cleaning the acquired multi-dimensional data, analyzing the text content extracted from the pictures and videos, and identifying key topics and emotional tendencies; Step 3: Model identification: Use the pre-trained large model to identify and preliminarily classify the data, and push the model identification results to public opinion analysis experts for further manual review and analysis; Step 4: Expert judgment and feedback: Public opinion experts manually judge the results of the model's automatic labeling to determine whether they are accurate. Experts can make fine adjustments and corrections based on the actual context of the data and provide feedback to the model. Through the expert feedback mechanism, experts' opinions and corrections on the model's labeling results are collected; Step 5: Learning and optimization: Use reinforcement learning algorithms to fine-tune the model based on public opinion data and expert feedback. By establishing an automated optimization system, input the results of expert annotations into the system to automatically update and adjust model parameters. After the optimization is completed, perform model verification to check its performance on different public opinion data sets. Step 6: Deployment and configuration: build a model management platform to achieve rapid deployment and version control of the model, adaptively learn the model according to new public opinion data types or trends, deploy a monitoring system, and conduct real-time analysis and early warning of public opinion data; Step 7: System feedback and continuous improvement. After deployment, the model will continue to optimize itself based on new data and expert feedback to maintain the efficiency and accuracy of the model, regularly evaluate the monitoring effect of the public opinion monitoring system, collect feedback from users and experts, and make improvements and adjustments.

[0033] To sum up: the present invention uses a large model and a reinforcement learning algorithm to enable the model to adapt to different public opinion environments. Through repeated labeling and feedback from public opinion identification experts, the present invention realizes the solidification of expert capabilities and reduces dependence on manual labeling by experts. Through continuous iterative optimization, it can continuously improve model performance and adapt to changes in the public opinion environment. It adopts modern technical means such as deep pre-training, reinforcement learning, and reward models. These technical means work together to achieve innovation and breakthroughs in public opinion monitoring technology.

[0034] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A human-machine collaborative public opinion monitoring system based on reinforcement learning, characterized in that: It includes the modules for analyzing allusions, TV and homophonic references, the modules for processing public opinion data, the modules for storing public opinion data, the modules for model identification, the modules for assessing public opinion risks, the modules for analyzing public opinion communication, the modules for learning optimization and the modules for deployment and configuration. The allusion, television and homophonic reference analysis module identifies and processes the details of various information dissemination through in-depth analysis and contextual understanding; The public opinion data processing module is used to collect, clean, organize, store and pre-process public opinion related data; The public opinion data storage module is used to manage and store a large amount of public opinion data; The model identification module is responsible for extracting useful sentiment information, hot topics and potential risks from public opinion data; The public opinion risk assessment module is used to predict and warn of risks in public opinion data and assess the potential influence of public opinion; The public opinion dissemination analysis module is used to analyze the dissemination path, dissemination speed and audience range of public opinion information; The learning optimization module is responsible for dynamically adjusting the algorithm model and parameters in the public opinion monitoring system through user feedback and the input of new data; The deployment configuration module is used to provide configuration, scheduling management and monitoring functions of the deployment environment.

2. According to claim 1, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The public opinion data processing module includes a data acquisition unit and a data processing unit; The data acquisition unit is used to acquire multi-dimensional public opinion data from the public opinion crawling system and extract the image and video content into text; The data processing unit uses natural language processing technology to extract and clean multidimensional data.

3. According to claim 2, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The public opinion data storage module includes a data storage unit and a data backup and recovery unit; The data storage unit is used to adopt a distributed database management system to ensure rapid access to massive data; The data backup and recovery unit is used to regularly back up public opinion data to ensure rapid recovery in the event of data loss or system failure.

4. According to claim 3, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The model recognition module includes a model recognition unit and an expert discrimination unit; The model recognition unit is used to use the trained large model to recognize and mark the public opinion data, and push the marked results to the public opinion analysis experts; The expert judgment unit is used by public opinion experts to judge the results of model labeling, mark whether the model labeling is correct, and realize human-machine collaborative judgment operations.

5. According to claim 4, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The public opinion risk assessment module includes a risk index calculation unit and an anomaly detection unit; The risk index calculation unit is used to assign a risk index to each piece of public opinion data, and to measure the influence and dissemination potential of public opinion by obtaining the reading volume, forwarding volume, number of comments and the average number of the above indicators within the release time. The formula is: Influence = w1*number of readings + w2*number of reposts + w3*number of comments; Communication potential = w4*number of readings / posting time + w5*number of reposts / posting time + w6*number of comments / posting time; Risk index = influence + communication potential; weights can be customized and adjusted, and average weights are generally used; The anomaly detection unit is used to detect abnormal fluctuations or emergencies in public opinion data and issue early warnings in a timely manner.

6. According to claim 5, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The public opinion propagation analysis module includes a propagation path analysis unit and a propagation speed prediction unit; The propagation path analysis unit is used to track the propagation path of public opinion information, identify the main propagation nodes and key figures, extract the common paths of public opinion propagation based on the historical data of information flow, and identify the most active users and platforms in the information propagation process; The propagation speed prediction unit is used to predict the propagation speed and development trend of public opinion information in the social network.

7. According to claim 6, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The learning optimization module includes a learning unit and an optimization unit; The learning unit is used for a model trained for the first time using a large model fine-tuning technique based on reinforcement learning; The optimization unit is used to collect expert annotation results, perform automatic optimization based on the results, and calculate the recognition accuracy of the optimized model.

8. According to claim 7, a human-machine collaborative public opinion monitoring system based on reinforcement learning is characterized in that: The deployment configuration module includes a model deployment unit and a configuration learning unit; The model deployment unit is used for optimized rapid deployment and new types of configuration learning; The configuration learning unit is used to manually configure learning strategies for different public opinion types to improve the adaptability and accuracy of the model.

9. The human-machine collaborative public opinion monitoring system based on reinforcement learning according to claim 8 is characterized by: The large model fine-tuning technology based on reinforcement learning is the process of further fine-tuning the already trained large models GPT and BERT using the reinforcement learning algorithm. Unlike traditional supervised learning methods, reinforcement learning learns strategies through interaction with the environment, and the ultimate goal is to optimize model behavior through reward signals.

10. The monitoring method of a human-machine collaborative public opinion monitoring system based on reinforcement learning according to claim 9 is characterized in that: The steps include: Step 1: Data acquisition: Use web crawler technology to crawl public opinion data in various formats from the Internet in real time, and perform preliminary extraction on the crawled data to identify the text content in pictures and videos; Step 2: Data processing: cleaning the acquired multi-dimensional data, analyzing the text content extracted from the pictures and videos, and identifying key topics and emotional tendencies; Step 3: Model identification: Use the pre-trained large model to identify and preliminarily classify the data, and push the model identification results to public opinion analysis experts for further manual review and analysis; Step 4: Expert judgment and feedback: Public opinion experts manually judge the results of the model's automatic labeling to determine whether they are accurate. Experts can make fine adjustments and corrections based on the actual context of the data and provide feedback to the model. Through the expert feedback mechanism, experts' opinions and corrections on the model's labeling results are collected; Step 5: Learning and optimization: Use reinforcement learning algorithms to fine-tune the model based on public opinion data and expert feedback. By establishing an automated optimization system, input the results of expert annotations into the system to automatically update and adjust model parameters. After the optimization is completed, perform model verification to check its performance on different public opinion data sets. Step 6: Deployment and configuration: build a model management platform to achieve rapid deployment and version control of the model, adaptively learn the model according to new public opinion data types or trends, deploy a monitoring system, and conduct real-time analysis and early warning of public opinion data; Step 7: System feedback and continuous improvement. After deployment, the model will continue to optimize itself based on new data and expert feedback to maintain the efficiency and accuracy of the model, regularly evaluate the monitoring effect of the public opinion monitoring system, collect feedback from users and experts, and make improvements and adjustments.

Images