Data monitoring and early warning method and system based on AI technology

Abstract

The invention discloses a data monitoring and early warning method and system based on an AI technology. According to the method, firstly, multi-source data including texts, images and videos are collected, then deep analysis is carried out on the data, emotion analysis is carried out on the text data through a semantic emotion recognition technology, and dynamic anomaly detection is carried out on the image and video data through a self-supervised learning framework. An analysis result generates a preliminary monitoring result through a grading early warning mechanism, and a targeted coping suggestion is generated in combination with historical data. And finally, processing the preliminary monitoring result by using a target model prejudice detection algorithm to ensure objectivity and reliability of an early warning result. Through the multi-modal data enhancement and multi-objective optimization technology, the generalization ability and adaptability of the system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data visualization, and particularly relates to a data monitoring and early warning method and system based on AI technology. Background Art

[0002] In today's digital age, the rapid generation and dissemination of data have brought unprecedented opportunities and challenges to various industries. Especially in the fields of medical treatment, public opinion monitoring, etc., it is particularly important to monitor and early warn massive multi-modal data in a timely and accurate manner. However, there are many limitations in the existing data monitoring and early warning technologies.

[0003] In terms of data collection and preprocessing: The existing technologies mainly obtain multi-source data (text, images, videos) through interfaces such as social media and news platforms, but these data are often processed independently, lacking cross-modal association. For example, the analysis of text data cannot be effectively verified with image or video content, resulting in the problem of data islands. In addition, traditional data cleaning relies on manual rules, with low efficiency and prone to misjudgment.

[0004] In terms of single-modal analysis models: Taking text sentiment analysis as an example, existing technologies such as the negative information detection method based on large models (patent number CN119005198A) can label the sentiment polarity of text, but have insufficient recognition ability for complex expressions (such as sarcastic statements). At the same time, most of these models are static and rely on a large amount of labeled data for updating, making it difficult to adapt to emergencies (such as sudden changes in public opinion during the outbreak period).

[0005] In terms of early warning and response mechanisms: Traditional early warning systems usually set fixed thresholds (such as negative keyword frequency > 10 times / minute to trigger an early warning), and rely on manual review to push early warning information to the management end. This method has the problem of lagging response, with an average delay exceeding 30 minutes, and cannot dynamically adjust the threshold to capture potential risks in data fluctuations.

[0006] In terms of security and reliability: In existing technologies, algorithm decision biases may affect the accuracy of early warning results, and in the process of processing medical data, privacy protection and data security issues need to be urgently solved.

[0007] In summary, the existing technologies have obvious deficiencies in multi-modal data fusion, dynamic early warning, security and reliability, etc., and it is difficult to meet the increasingly complex data monitoring and early warning requirements. Summary of the Invention

[0008] Based on this, the embodiments of the present application provide a data monitoring and early warning method and system based on AI technology, which can effectively solve the above problems and improve the accuracy and efficiency of monitoring and early warning.

[0009] In a first aspect, a data monitoring and early warning method based on AI technology is provided, and the method includes:

[0010] Collect multi-source data; wherein, the multi-source data includes text, images and videos;

[0011] Analyze the multi-source data, including: perform semantic emotion recognition on text data, fuse emotion prediction and sarcasm content recognition algorithms based on semantic vector technology to obtain text sentiment analysis results; perform dynamic anomaly detection on image and video data, adopt a self-supervised learning framework, use unlabeled data to train a model, and analyze the trend of time-series data through an LSTM network to obtain anomaly detection results;

[0012] According to the text sentiment analysis results and the anomaly detection results, adopt a hierarchical early warning mechanism, divide the risks into three levels: low, medium, and high, and generate preliminary monitoring results in combination with historical data;

[0013] Use the target model bias detection algorithm to process the preliminary monitoring results to obtain data monitoring and early warning results; wherein, the target model bias detection algorithm is used to monitor the implicit bias in the sentiment analysis results in real time and trigger model retraining to reduce the bias rate of the early warning results.

[0014] Optionally, the collecting of multi-source data includes:

[0015] Obtain text, image and video data in social media, news platforms and medical databases through network interfaces, and obtain real-time monitoring image and video data through Internet of Things devices.

[0016] Optionally, the dynamic anomaly detection step includes:

[0017] Extract key frames and perform object detection on image and video data, generate adversarial samples using unlabeled data based on a self-supervised learning framework to optimize the model's generalization ability for abnormal patterns, analyze the trend of time-series data through an LSTM network to trigger dynamic threshold adjustment, and identify potential risks in real time.

[0018] Optionally, adopting a hierarchical early warning mechanism, dividing the risks into three levels: low, medium, and high, and generating preliminary monitoring results in combination with historical data includes:

[0019] Divide the risks into three levels: low, medium, and high according to the risk score, automatically generate soothing words for low risks, provide public opinion guidance strategies for medium risks, and trigger resource allocation instructions for high risks;

[0020] Embed medical resource data and industry rules into the knowledge graph, and retrieve associated nodes in the knowledge graph in real time when an early warning is triggered to generate preliminary monitoring results.

[0021] Optionally, the preliminary monitoring results are processed using a target model bias detection algorithm to obtain data monitoring warning results, and the implicit biases in the sentiment analysis results are monitored in real time, including:

[0022] Monitor the implicit biases in the sentiment analysis results in real time. When a bias is detected, trigger model retraining to reduce the bias rate of the warning results.

[0023] Optionally, the processing of the preliminary monitoring results using a target model bias detection algorithm to obtain data monitoring warning results further includes:

[0024] Generate synthetic data through a multi-modal adversarial generation network to enhance the model's robustness to rare cases or extreme scenarios;

[0025] Design a multi-task loss function to synchronously optimize sentiment analysis accuracy, anomaly detection response speed, and recommendation generation rationality.

[0026] In a second aspect, a data monitoring warning system based on AI technology is provided. The system includes:

[0027] An acquisition module for acquiring multi-source data; wherein, the multi-source data includes text, images, and videos;

[0028] An analysis module for analyzing the multi-source data, including: performing semantic emotion recognition on text data, fusing sentiment prediction and sarcasm content recognition algorithms based on semantic vector technology to obtain text sentiment analysis results; performing dynamic anomaly detection on image and video data, using an unsupervised learning framework, training a model with unlabeled data, and analyzing the trend of time-series data through an LSTM network to obtain anomaly detection results;

[0029] A diversity module for classifying risks into three levels: low, medium, and high according to the text sentiment analysis results and the anomaly detection results, and generating preliminary monitoring results in combination with historical data;

[0030] A processing module for processing the preliminary monitoring results using a target model bias detection algorithm to obtain data monitoring warning results; wherein, the target model bias detection algorithm is used to monitor the implicit biases in the sentiment analysis results in real time and trigger model retraining to reduce the bias rate of the warning results.

[0031] In a third aspect, an electronic device is provided, including a memory and a processor. When the processor executes the computer program stored in the memory, it implements the data monitoring warning method described in any one of the first aspects above.

[0032] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the data monitoring and early warning method described in any one of the above first aspects is implemented.

[0033] In a fifth aspect, a computer program product is provided, including a computer program / instructions. When the computer program / instructions are executed by a processor, the data monitoring and early warning method described in any one of the above first aspects is implemented.

[0034] The beneficial effects brought by the technical solutions provided in the embodiments of the present application at least include:

[0035] (1) The monitoring accuracy is significantly improved, and cross-modal verification reduces the false alarm rate: Through the fusion analysis of multi-modal data (text, image, video) and dynamic anomaly detection technology, the present application can effectively reduce the false alarm rate, and at the same time, monitor the deviation in the sentiment analysis results in real time and trigger model retraining to ensure the objectivity and accuracy of the early warning results.

[0036] (2) Optimize the early warning response speed and decision-making efficiency: By adopting a hierarchical early warning mechanism and a knowledge graph-driven recommendation generation module, targeted response suggestions can be quickly generated, and the response time can be shortened to the second level, significantly improving the decision-making efficiency, especially suitable for scenarios that require quick response, such as medical public opinion monitoring.

[0037] (3) Enhance the generalization ability and adaptability of the system: Through cross-modal data augmentation and multi-objective optimization technology, the present application can effectively solve the problem of data scarcity, improve the model's recognition ability for rare events or extreme scenarios, and at the same time adapt to the complex requirements in different scenarios, with broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, other implementation drawings can be obtained by extending according to the provided drawings without creative efforts.

[0039] Figure 1 It is a step flowchart of a data monitoring and early warning method based on AI technology provided by an embodiment of the present application;

[0040] Figure 2 It is a block diagram of a data monitoring and early warning system based on AI technology provided by an embodiment of the present application;

[0041] Figure 3 It is a schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] In order to make the objectives, technical solutions, and advantages of this application clearer and more understandable, the following further details this application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0043] In the description of the present invention, the terms "including", "having", and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units that are clearly listed, but may also include other steps or units that are inherent to these processes, methods, products, or devices although not clearly listed, or steps or units added based on further optimization solutions conceived from the present invention.

[0044] In today's digital age, the rapid generation and dissemination of data have brought unprecedented opportunities and challenges to various industries. Especially in fields such as healthcare and public opinion monitoring, it is particularly important to monitor and give early warnings to massive amounts of multimodal data in a timely and accurate manner. However, the existing data monitoring and early warning technologies have many limitations.

[0045] The following is an explanation of the existing technical solutions:

[0046] 1. Data collection and preprocessing specifically include:

[0047] Technical process:

[0048] ① Obtain multi-source data (text, images, videos) from interfaces such as social media and news platforms.

[0049] ② Independently process different modality data: For text: word segmentation, denoising, sentiment polarity annotation; for images / videos: key frame extraction, object recognition (such as medical device status monitoring).

[0050] Limitations:

[0051] Data silo problem: Text, image, and video data are analyzed independently, lacking cross-modal association (such as being unable to verify the consistency between the text description of "a shortage of a certain drug" and the inventory monitoring image).

[0052] Inefficient cleaning: Rely on manual rules to filter noise (such as misjudging medical terms as negative words).

[0053] 2. The single-modal analysis model specifically includes:

[0054] Technical process (taking the patent of CN119005198A: A method for detecting negative information based on a large model as an example):

[0055] ① Construct a multi-modal sentiment vector: Extract semantic features from the text and map them to the sentiment space (such as positive / negative).

[0056] ②Train a classification model by combining historical data and output sentiment labels.

[0057] Limitations:

[0058] Insufficient semantic understanding: Unable to recognize complex expressions (such as the ironic statement "The efficiency of this hospital is really 'high', I waited for three hours and didn't see the doctor").

[0059] Dependence on static models: A large amount of labeled data is required to update the model, making it difficult to adapt to emergencies (such as sudden changes in public opinion during the outbreak period).

[0060] 3. The early warning and response mechanism specifically includes:

[0061] Technical process:

[0062] ①Set a fixed threshold (e.g., when the frequency of negative keywords > 10 times / minute, an early warning is triggered).

[0063] ②After manual review, push the early warning information to the management terminal.

[0064] Limitations:

[0065] Lag in response: The manual review process causes delays (average > 30 minutes).

[0066] Lack of dynamic adjustment: Static thresholds cannot capture potential risks in data fluctuations (such as slowly fermenting doctor-patient conflicts).

[0067] For the defects of the above existing technologies, this application introduces multi-modal correlation analysis (such as text + image cross-verification), adopts self-supervised learning to dynamically adjust the early warning threshold, and embeds an automated bias detection module to replace manual review.

[0068] Please refer to Figure 1 , which shows the flowchart of a data monitoring and early warning method provided by an embodiment of this application, and may include the following steps:

[0069] S1, Collect multi-source data.

[0070] Among them, the multi-source data includes text, images, and videos. In this step, by integrating network interface and Internet of Things device data, multi-modal data including text, images, and videos is collected. Among them, the text data comes from channels such as social media and news platforms; the image and video data are obtained through Internet of Things devices or monitoring systems. During the collection process, the system supports the access of multiple data sources to ensure the diversity and comprehensiveness of the data, providing a rich data basis for subsequent analysis.

[0071] S2. Analyze multi-source data, including: perform semantic sentiment recognition on text data, fuse sentiment prediction and sarcasm content recognition algorithms based on semantic vector technology to obtain text sentiment analysis results; perform dynamic anomaly detection on image and video data, adopt a self-supervised learning framework, use unlabeled data to train the model, and analyze the trend of time-series data through the LSTM network to obtain anomaly detection results.

[0072] In this step, analyze the collected multi-source data, which specifically includes two parts:

[0073] Semantic sentiment recognition of text data: Based on semantic vector technology, fuse sentiment prediction and sarcasm content recognition algorithms to process text data. First, perform preprocessing operations such as word segmentation and denoising on the text, then extract the semantic features of the text and map them to the sentiment space to obtain sentiment labels. By combining historical data to train a classification model, further optimize the accuracy of sentiment prediction, so as to obtain text sentiment analysis results.

[0074] Dynamic anomaly detection of image and video data: Adopt a self-supervised learning framework, use unlabeled data to train the model, analyze the trend of time-series data through the LSTM network, and real-time identify potential risks in data fluctuations. In specific implementation, extract key frames and perform object detection on image and video data, generate adversarial samples based on self-supervised learning, optimize the generalization ability of the model for abnormal patterns, and finally obtain anomaly detection results. This step provides an accurate analysis basis for subsequent early warning decisions through in-depth analysis of multi-modal data.

[0075] S3. According to the text sentiment analysis results and anomaly detection results, adopt a hierarchical early warning mechanism, divide the risks into three levels: low, medium, and high, and generate preliminary monitoring results in combination with historical data.

[0076] In this step, according to the text sentiment analysis results and the anomaly detection results of images / videos, adopt a hierarchical early warning mechanism to divide the risks into three levels: low, medium, and high. In specific implementation, the system generates preliminary monitoring results in combination with historical data and determines the early warning level according to the risk score. For low risks, automatically generate soothing words; for medium risks, provide public opinion guidance strategies; for high risks, trigger resource allocation instructions. In addition, through a knowledge graph-driven advice generation module, embed medical resource data and industry rules into the knowledge graph, and when an early warning is triggered, retrieve associated nodes in real time to generate targeted coping suggestions.

[0077] S4. Use the target model bias detection algorithm to process the preliminary monitoring results to obtain data monitoring early warning results.

[0078] Among them, the target model bias detection algorithm is used to monitor the implicit bias in the sentiment analysis results in real time and trigger model retraining to reduce the bias rate of the early warning results.

[0079] In this step, the embedded target model bias detection algorithm is used to process the preliminary monitoring results to obtain the final data monitoring and early warning results. In specific implementation, the bias detection algorithm monitors the implicit biases in the sentiment analysis results in real time. For example, it detects whether there are misjudgments of certain sentiments or events (such as misjudging "first aid delay" as low risk). Once a bias is detected, the system triggers model retraining, and adjusts the model parameters to reduce the bias rate of the early warning results, thereby ensuring the objectivity and reliability of the early warning results.

[0080] It can be seen that the platform corresponding to this application is divided into four major modules:

[0081] ① Multi-source data acquisition layer: Integrate network interfaces (social media, news, medical databases), and data from Internet of Things devices, supporting multi-modal input of text, images, and videos.

[0082] ② AI analysis engine:

[0083] Semantic emotion recognition module: Based on the semantic vector technology of Zhongke Wenge, integrating emotion prediction and sarcasm content recognition algorithms to improve the accuracy of negative public opinion detection.

[0084] Dynamic anomaly detection module: Adopt a self-supervised learning framework, use unlabeled data to train the model, and identify potential risks in data fluctuations in real time.

[0085] ③ Intelligent early warning center:

[0086] Hierarchical early warning mechanism (low / medium / high risk), generating response suggestions (such as resource allocation plans, public opinion guidance strategies) in combination with historical data.

[0087] ④ Security and feedback system:

[0088] Embed target model bias detection to avoid the influence of algorithm decision bias on early warning results.

[0089] In summary, the advantages and effects of the present invention compared with the prior art include:

[0090] 1. Significantly improved monitoring accuracy: Cross-modal verification reduces the false alarm rate.

[0091] Technical solution association: Dynamic fusion of multi-modal data and cross-attention verification mechanism.

[0092] Effect deduction:

[0093] Defects of the prior art: Traditional single-modal analysis (such as only relying on text sentiment analysis) is vulnerable to false information interference (for example, a user posts a text "drug shortage in a certain hospital", but the actual inventory image shows sufficient).

[0094] Improvements of the present invention: Logical verification: Through cross-modal attention mechanism, calculate the semantic similarity between text description and image / video content (such as the matching degree between the text of "drug shortage" and the object detection results of the inventory shelf image). If the similarity is lower than the threshold (such as <60%), it is determined as a suspected false public opinion.

[0095] 2. Optimization of early warning response speed: Dynamic self-supervised learning shortens the delay.

[0096] Technical solution association: Self-supervised adversarial training and dynamic threshold adjustment.

[0097] Effect deduction:

[0098] Defects of the prior art: Traditional fixed-threshold early warning cannot adapt to sudden fluctuations (such as a sharp increase in public opinion during the outbreak period), and relies on manual review (average delay > 30 minutes).

[0099] Improvements of the present invention: Dynamic modeling: Analyze the trend of time-series data (such as the hourly growth rate of patient complaints) through an LSTM network. If the slope exceeds 3 times the standard deviation of the historical baseline, trigger dynamic threshold adjustment (such as automatically reducing the threshold to 5 times per minute).

[0100] 3. Improvement of the coverage rate of response suggestions: Knowledge graph-driven automated decision-making.

[0101] Technical solution association: Hierarchical early warning and knowledge graph-based suggestion generation.

[0102] Effect deduction:

[0103] Defects of the prior art: Manually generated suggestions are inefficient (such as hospitals need to manually retrieve inventory data), and the covered scenarios are limited (only common cases can be processed).

[0104] Improvements of the present invention: Knowledge graph integration: Integrate medical resource data (drug inventory, doctor scheduling), industry rules into the graph. When an early warning is triggered, retrieve associated nodes in real time (such as "drug shortage in a certain area" is automatically associated with the inventory distribution of nearby hospitals).

[0105] 4. Enhancement of security and compliance: Bias correction.

[0106] Technical solution association: Model bias detection.

[0107] Effect deduction:

[0108] Defects of the prior art: Traditional models have implicit biases (such as misjudging "emergency treatment delay" as low risk), and centralized processing of medical data is prone to leakage of patient privacy.

[0109] Improvements of the present invention: Bias correction logic: By detecting statistical biases in sentiment analysis results (such as the proportion of negative reviews of a certain hospital being abnormally higher than that of similar institutions), trigger retraining of the model, reducing the deviation rate of early warning results by 35%.

[0110] 5. Improvement of model generalization ability: Data augmentation and multi-objective optimization.

[0111] Association of technical solutions: Cross-modal data augmentation and joint optimization.

[0112] Effect deduction:

[0113] Defects of the prior art: The scarcity of medical data leads to overfitting of the model (such as insufficient samples of rare disease public opinions), and a single optimization objective (such as only pursuing accuracy) sacrifices other performances.

[0114] Improvements of the present invention: ① Effect of data augmentation: Generate synthetic data through MAGAN (such as simulating text and image consultation records of rare disease patients), improving the recognition rate of the model for long-tail scenarios to 78% (originally 52%). ② Multi-objective balance: Jointly optimize the loss function (sentiment analysis loss + response time loss + suggestion rationality loss), achieving balanced improvements in F1-score (85%), response time (<5 seconds), and suggestion adoption rate (82%) in the test set.

[0115] In an alternative embodiment of the present application, the present application may further include:

[0116] 1. Multi-modal data dynamic fusion and cross-modal verification technology.

[0117] Technical means:

[0118] Cross-modal association modeling: Through a joint learning framework, dynamically fuse multi-modal data such as text (patient feedback, public opinion text), images (medical device monitoring images), and videos (remote diagnosis and treatment video streams) at the feature level, adopting an early-middle-late fusion strategy (early fusion for feature alignment, middle fusion for semantic association, and late fusion for decision complementarity).

[0119] Innovation:

[0120] Solve the data island problem of traditional single-modal analysis, and reduce the false alarm rate through cross-modal verification (such as filtering false public opinion information).

[0121] The fusion strategy adapts to the characteristics of the telemedicine scenario, such as the association modeling between medical terms and image pathological features.

[0122] 2. Dynamic anomaly detection technology based on self-supervised learning.

[0123] Technical means:

[0124] Self-supervised Adversarial Training: Generate adversarial samples using unlabeled data (such as simulating sudden public opinion fluctuations and abnormal signals of medical equipment), and optimize the generalization ability of the model for abnormal patterns through contrastive learning.

[0125] Dynamic Threshold Adjustment: Based on the fluctuation trend of time-series data (such as the frequency of patient complaints and equipment monitoring indicators), predict the risk threshold through an LSTM network to replace the traditional fixed threshold for triggering early warnings.

[0126] Innovation:

[0127] Reduce the dependence on manually labeled data and improve the adaptability of the model to emergencies (such as real-time response during the outbreak of public opinion).

[0128] Combine the dynamic characteristics of medical scenarios (such as periodic equipment maintenance data) to optimize the balance between the sensitivity and specificity of anomaly detection.

[0129] 3. Hierarchical Early Warning and Automated Response Recommendation Generation Technology.

[0130] Technical Means:

[0131] Multi-level Risk Classification Engine: Based on risk scores (such as the intensity of negative public opinion and the equipment anomaly index), divide low / medium / high risk levels, and each level is associated with different response strategies (low risk: automatically generate soothing words; high risk: trigger resource allocation instructions).

[0132] Knowledge Graph-driven Recommendation Generation: Integrate a medical knowledge graph and generate response recommendations in combination with real-time data.

[0133] Innovation:

[0134] Achieve full-link automation from early warning to response, and shorten the response time to the second level (compared with the delay of more than 30 minutes in traditional manual review).

[0135] 4. Model Bias Correction Technology.

[0136] Technical Means:

[0137] Bias Detection Module: Embed the target model bias detection algorithm (refer to the patent of Zhongke Wenge), and monitor the implicit bias in the sentiment analysis results in real time (such as misjudging "first aid delay" as low risk).

[0138] Innovation:

[0139] Improve the objectivity of early warning results and avoid misleading allocation of medical resources by algorithmic decisions.

[0140] 5. Cross-modal Data Augmentation and Joint Optimization Technology.

[0141] Technical Means:

[0142] Multi-modal Adversarial Generation Network (MAGAN): Generate synthetic data (such as simulated doctor-patient dialogue texts and corresponding diagnosis and treatment videos) to enhance the model's robustness to rare cases or extreme scenarios.

[0143] Joint optimization objective function: Design a multi-task loss function to synchronously optimize the accuracy of sentiment analysis, the response speed of anomaly detection, and the rationality of recommendation generation, avoiding overfitting of a single metric.

[0144] Innovation:

[0145] Solve the problem of scarcity of medical data (such as insufficient samples of public opinion on rare diseases) and improve the generalization ability of the model.

[0146] Balance performance metrics through multi-objective optimization to adapt to the complex requirements of the telemedicine scenario.

[0147] Such as Figure 2 , the embodiment of the present application also provides a block diagram of a data monitoring and warning system based on AI technology. The system may include:

[0148] An acquisition module for acquiring multi-source data; wherein, the multi-source data includes text, images, and videos;

[0149] An analysis module for analyzing the multi-source data, including: performing semantic emotion recognition on text data, fusing an emotion prediction and sarcasm content recognition algorithm based on semantic vector technology to obtain a text sentiment analysis result; performing dynamic anomaly detection on image and video data, adopting a self-supervised learning framework, training a model using unlabeled data, and analyzing the trend of time-series data through an LSTM network to obtain an anomaly detection result;

[0150] A diversity module for dividing risks into three levels of low, medium, and high according to the text sentiment analysis result and the anomaly detection result by adopting a hierarchical warning mechanism, and generating a preliminary monitoring result in combination with historical data;

[0151] A processing module for processing the preliminary monitoring result using a target model bias detection algorithm to obtain a data monitoring and warning result; wherein, the target model bias detection algorithm is used to monitor the implicit bias in the sentiment analysis result in real time and trigger model retraining to reduce the bias rate of the warning result.

[0152] For the specific limitations of the AI technology-based data monitoring and early warning system, reference can be made to the limitations of the AI technology-based data monitoring and early warning method in the above text, which will not be elaborated here. Each module in the above AI technology-based data monitoring and early warning system can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0153] In one embodiment, an electronic device is provided. The electronic device may be a computer, and its internal structure diagram may be as Figure 3 shown. The electronic device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the device is used to provide computing and control capabilities. The memory of the device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for data monitoring and early warning data based on AI technology. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a data monitoring and early warning method based on AI technology.

[0154] Those skilled in the art can understand that the structure shown in Figure 3 is only a block diagram of a part of the structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0155] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, covering all or part of the processes in the method of the above embodiment.

[0156] In one embodiment, a computer program product is also provided, including a computer program / instructions, covering all or part of the processes in the method of the above embodiment.

[0157] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in M forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (SyMchliMk) DRAM (SLDRAM), memory bus (RaMbus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0158] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0159] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A data monitoring and early warning method based on AI technology, characterized in that, The method includes: Collecting multi-source data; wherein, the multi-source data includes text, images, and videos; Analyzing the multi-source data, including: performing semantic sentiment recognition on the text data, fusing an emotion prediction and sarcasm content recognition algorithm based on semantic vector technology to obtain a text sentiment analysis result; performing dynamic anomaly detection on the image and video data, adopting a self-supervised learning framework, training a model using unlabeled data, and analyzing the trend of time-series data through an LSTM network to obtain an anomaly detection result; According to the text sentiment analysis result and the anomaly detection result, adopting a hierarchical early warning mechanism, dividing the risks into three levels: low, medium, and high, and generating a preliminary monitoring result in combination with historical data; Processing the preliminary monitoring result using a target model bias detection algorithm to obtain a data monitoring early warning result; wherein, the target model bias detection algorithm is used to monitor the implicit bias in the sentiment analysis result in real time and trigger model retraining to reduce the bias rate of the early warning result.

2. The data monitoring and early warning method according to claim 1, wherein The collecting of multi-source data includes: Obtaining text, image, and video data from social media, news platforms, and medical databases through a network interface, and obtaining real-time monitoring image and video data through Internet of Things devices.

3. The data monitoring and early warning method according to claim 1, wherein The dynamic anomaly detection step includes: Performing key frame extraction and object detection on the image and video data, generating adversarial samples using unlabeled data based on a self-supervised learning framework to optimize the model's generalization ability for abnormal patterns, analyzing the trend of time-series data through an LSTM network to trigger dynamic threshold adjustment, and identifying potential risks in real time.

4. The data monitoring and early warning method according to claim 1, characterized in that Adopting a hierarchical early warning mechanism, dividing the risks into three levels: low, medium, and high, and generating a preliminary monitoring result in combination with historical data includes: Dividing the risks into three levels: low, medium, and high according to the risk score, automatically generating soothing words for low risks, providing public opinion guidance strategies for medium risks, and triggering resource allocation instructions for high risks; Embedding medical resource data and industry rules into a knowledge graph, and retrieving associated nodes in the knowledge graph in real time when an early warning is triggered to generate a preliminary monitoring result.

5. The data monitoring and early warning method according to claim 1, characterized in that Processing the preliminary monitoring result using a target model bias detection algorithm to obtain a data monitoring early warning result, and monitoring the implicit bias in the sentiment analysis result in real time, including: Monitoring the implicit bias in the sentiment analysis result in real time, and triggering model retraining when a bias is detected to reduce the bias rate of the early warning result.

6. The data monitoring and early warning method according to claim 1, characterized in that Processing the preliminary monitoring result using a target model bias detection algorithm to obtain a data monitoring early warning result further includes: Generating synthetic data through a multi-modal adversarial generation network to enhance the model's robustness to rare cases or extreme scenarios; Designing a multi-task loss function to synchronously optimize the sentiment analysis accuracy, anomaly detection response speed, and rationality of suggestion generation.

7. A data monitoring and warning system based on AI technology, characterized in that, The system includes: A collection module for collecting multi-source data; wherein, the multi-source data includes text, images, and videos; An analysis module for analyzing the multi-source data, including: performing semantic emotion recognition on text data, fusing an emotion prediction and sarcasm content recognition algorithm based on semantic vector technology to obtain a text emotion analysis result; performing dynamic anomaly detection on image and video data, adopting a self-supervised learning framework, training a model using unlabeled data, and analyzing the trend of time-series data through an LSTM network to obtain an anomaly detection result. A diversity module for, according to the text emotion analysis result and the anomaly detection result, adopting a hierarchical early warning mechanism, dividing risks into three levels: low, medium, and high, and generating a preliminary monitoring result in combination with historical data. A processing module for processing the preliminary monitoring result using a target model bias detection algorithm to obtain a data monitoring early warning result; wherein the target model bias detection algorithm is used to monitor the implicit bias in the emotion analysis result in real time and trigger model retraining to reduce the bias rate of the early warning result.

8. An electronic device, characterized in that, It includes a memory and a processor, and the memory stores a computer program, and when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images