Method and system for extracting health epidemiology classification information of disease outbreak

Abstract

The present application relates to a kind of disease outbreak health epidemiology classification information extraction method and system, the method includes: data acquisition stage, utilizes ETL tool from hospital information system, disease control center database, social media and medical literature database extract multi-source disease related data;Data preprocessing stage, to the data collected is washed, duplicate removal, missing value completion and format standardization;Natural language processing stage, adopts named entity recognition and dependency syntax analysis technology to extract the key information such as symptom, pathogen, time, place in case;Machine learning classification stage, in combination with logistic regression, decision tree and neural network model determines, transmission risk classification and disease trend prediction to case;Knowledge graph construction stage, case, symptom, place and transmission relationship are organized into graph database, form structured disease knowledge graph;Information application stage.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and more specifically, to a method and system for extracting health epidemiological classification information on disease outbreaks. Background Technology

[0002] Traditional disease epidemiological information extraction relies primarily on manual processing and some basic information technology tools. Manual processing typically involves professional disease control personnel or medical staff reviewing, screening, and recording various disease-related data document by document, such as flipping through paper case reports and manually recording basic patient information, symptoms, and contact history. Regarding information technology applications, some use simple data retrieval tools, searching electronic document databases using keywords to obtain relevant textual data; others utilize basic database query statements to extract specific fields from structured case databases, such as querying the number of confirmed cases and age distribution from databases storing case information. Additionally, some use simple statistical software to perform preliminary statistical analysis of the collected data, such as calculating basic indicators like morbidity and mortality rates.

[0003] Manual processing is extremely inefficient. During large-scale disease outbreaks, the sheer volume of case reports, medical literature, and various monitoring data necessitates significant time and manpower for manual processing, leading to severe delays in information acquisition and hindering timely support for disease control decisions. Furthermore, manual processing is prone to human error. Due to the large volume and complexity of information, staff may overlook or misjudge details during prolonged processing, affecting the accuracy and completeness of the information. Simple information technology tools also have limitations when dealing with complex disease information. For example, keyword searches alone cannot accurately understand the semantics of text, making it difficult to accurately extract information with similar meanings but different expressions, easily leading to information omissions. Basic database queries are limited to structured data, failing to effectively extract information from the vast amounts of unstructured text data, such as case descriptions and medical research reports. Simple data statistical software can only perform routine statistical analysis, struggling to uncover deeper relationships and potential patterns between data, thus failing to meet the needs of disease control for precise analysis and risk assessment. These shortcomings can lead to an inability to grasp the spread of the disease in a timely and accurate manner, and an inability to formulate precise prevention and control strategies, thereby affecting the effectiveness of disease prevention and control and potentially causing further spread of the disease and a waste of prevention and control resources.

[0004] Therefore, there is an urgent need to design a method for extracting information on the health epidemiology classification of disease outbreaks in order to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for extracting health epidemiological classification information of disease outbreaks, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for extracting health epidemiological classification information of disease outbreaks includes the following steps:

[0008] S1, multi-source data acquisition, builds a data fusion platform, and uses ETL tools, distributed acquisition frameworks or message queue middleware to obtain case information, disease monitoring data, public feedback information and research results from medical institution information systems, disease control center monitoring databases, community health service platforms, social media and medical literature databases, ensuring the comprehensiveness and diversity of data sources;

[0009] S2. Preprocess the collected data, including: removing noisy and duplicate data using rule-driven or machine learning data cleaning algorithms; handling missing values ​​through mean imputation, K-nearest neighbor imputation, or labeling for verification; and standardizing data from different sources, including standardizing date formats, stemming symptom texts, and part-of-speech tagging, to ensure consistent data formats and semantic comparability.

[0010] S3 utilizes natural language processing technology to process unstructured text, including named entity recognition, dependency parsing, semantic understanding, and semantic role labeling. It identifies key information such as case symptoms, pathogens, locations, and transmission behaviors from case reports, social media texts, and medical literature, and transforms them into structured data.

[0011] S4. Based on machine learning algorithms, feature extraction and classification model training are performed on structured data. The model may include logistic regression, decision tree, neural network, etc. The input features cover basic case information, symptom characteristics, transmission path, contact history and activity trajectory. The output classification results include confirmed cases, suspected cases, close contacts and transmission risk level, realizing automated disease information classification.

[0012] S5 uses entity recognition and relation extraction technology to construct a disease knowledge graph. It uses cases, locations, symptoms, and transmission events as nodes, and transmission relationships, activity trajectories, and symptom associations as edges. The attributes of the edges include contact time, symptom onset time, etc. It uses a graph database for storage and management and realizes visualization display.

[0013] S6, The knowledge graph and classification results are combined and applied to disease prevention and control decision support, including transmission chain tracing and at-risk population location, disease trend prediction and risk area delineation, prevention and control strategy generation and dynamic optimization, as well as supply and demand matching and scheduling of medical resources, isolation resources, materials and manpower.

[0014] As a preferred embodiment of the present invention, the multi-source data acquisition further includes:

[0015] a) For medical institution systems, configure database connection addresses, user permissions, and extraction rules, and periodically extract case diagnosis and treatment records;

[0016] b) For social media data, use web crawlers to set keyword strategies and time windows to obtain disease-related posts and comments;

[0017] c) Automatically retrieve and download research findings related to the current disease from literature databases to ensure the timely inclusion of research progress information.

[0018] 3. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that the data preprocessing further includes:

[0019] a) Delete invalid records in the input data caused by input errors, such as negative age or abnormal date data;

[0020] b) Use a deduplication algorithm to identify and remove duplicate data to avoid the same case being counted repeatedly;

[0021] c) Standardize and unify the technical terms and colloquial expressions in the text data to ensure semantic consistency.

[0022] As a preferred technical solution of the present invention, the natural language processing technology adopts a named entity recognition method based on conditional random field (CRF) or deep learning BERT model, combined with dependency parsing, to identify the semantic relationships between cases, symptoms, pathogens, geographical locations and transmission events, and form logically complete structured information based on semantic role labeling.

[0023] As a preferred technical solution of the present invention, the machine learning classification model is implemented in the following manner:

[0024] a) Using a logistic regression model, combined with basic characteristics such as age, gender, and symptom severity, cases are classified as either confirmed or suspected cases;

[0025] b) Using a decision tree model, the transmission risk level of cases is classified based on characteristics such as case contact history and activity trajectory;

[0026] c) Using neural network models, model and predict complex symptom combinations and potential nonlinear relationships, and output the case progression trend.

[0027] As a preferred technical solution of the present invention, the classification model is trained using cross-validation to evaluate accuracy, and the model performance is optimized by adjusting parameters such as regularization coefficient, tree depth, number of network layers and learning rate, thereby achieving a classification accuracy of over 95%.

[0028] As a preferred embodiment of the present invention, the construction of the disease knowledge graph includes:

[0029] a) Use case, symptoms, location, and transmission relationship as nodes;

[0030] b) Treat the relationships between cases and symptoms, cases and locations, and cases and transmission events as edges;

[0031] c) Add attributes such as contact duration, contact frequency, and symptom onset time to the edge;

[0032] d) The Neo4j graph database is used for storage, and the creation, querying and updating of nodes and edges are implemented using the Cypher language.

[0033] As a preferred technical solution of the present invention, the disease prevention and control decision support includes:

[0034] a) Visual tracing of transmission chains and generation of risk population lists based on knowledge graphs;

[0035] b) Prediction of new disease cases and risk area classification based on time series models (ARIMA or LSTM);

[0036] c) Develop regional prevention and control strategies based on case distribution, transmission characteristics, and population risk characteristics;

[0037] d) Optimize resource scheduling based on real-time supply and demand gap calculations to achieve precise allocation of medical supplies, isolation beds, and protective personnel.

[0038] As a preferred technical solution of the present invention, the disease prevention and control strategy generation has a dynamic optimization mechanism. By extracting indicators such as the rate of change in the number of new cases, the proportion of cases in risk areas, the infection rate of key populations, and the implementation rate of measures in real time, the strategy implementation effect is evaluated, and the risk level and prevention and control intensity are automatically adjusted when the case growth rate exceeds the threshold or when there are no new cases for a period of time.

[0039] The Disease Outbreak Health Epidemiology Classification Information Extraction System includes a multi-source data acquisition module, which is used to obtain original disease-related data such as case diagnosis and treatment records, epidemiological survey data, public opinion texts, and population activity trajectories from hospital information systems, disease control center databases, social media platforms, environmental and meteorological monitoring devices, etc.

[0040] The data preprocessing module is used to clean, desensitize, denoise, standardize the format, fill in missing data and align the time and space of the raw data, and output the normalized data to the downstream module.

[0041] The natural language processing module is used to receive the normalized data, perform word segmentation, named entity recognition, relation extraction and event extraction on the text information therein, and output the entity, relation and feature information of the disease elements;

[0042] The machine learning classification module is used to receive the disease element entity, relationship and feature information, classify and determine the case category, transmission scenario type, risk level and cluster event based on the pre-trained model, and generate classification labels;

[0043] The knowledge graph construction module is used to receive the classification tags and disease element entities and relationships, align entities and mapping relationships, and perform reasoning verification to form a disease knowledge graph.

[0044] The information application module is used to provide disease monitoring and early warning, source tracing analysis, risk stratification and resource scheduling based on the disease knowledge graph, and to feed back manual correction, early warning hit status and false alarms to the machine learning classification module, natural language processing module and data preprocessing module to realize model update, rule revision and knowledge graph ontology evolution.

[0045] This invention also proposes a system for extracting information on the classification of public health epidemiology of disease outbreaks, including a multi-source data acquisition module, a data preprocessing module, a natural language processing module, a machine learning classification module, a knowledge graph construction module, and an information application module.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. Efficiency Improvement: In actual disease prevention and control cases, before adopting this invention, a certain region required five professionals to spend two days manually processing 1,000 new case reports and related data daily to complete the initial information extraction and classification. After adopting this invention, through automated data collection and processing, information extraction and preliminary analysis can be completed within 10 minutes, improving efficiency by hundreds of times. This enables timely provision of the latest data to support disease prevention and control decisions, allowing control measures to respond more quickly to changes in the disease.

[0048] 2. Improved Accuracy: Through testing and verification on historical disease data, the classification accuracy of traditional methods for disease information classification is approximately 70% due to human factors and technical limitations. However, by employing the machine learning classification model and natural language processing technology of this invention, the classification accuracy can reach over 95%. For example, in the task of classifying confirmed and suspected cases, it can more accurately identify the true status of cases, reduce misjudgments, and provide more precise information for timely treatment and isolation control of patients. 3. Optimized Decision Support: Taking disease transmission risk assessment as an example, the constructed disease knowledge graph can intuitively display the key paths and potential risk nodes of disease transmission. In a certain disease transmission event, knowledge graph analysis quickly identified a key location for cluster transmission, enabling timely lockdown and personnel screening measures, effectively curbing the further spread of the disease. Compared to traditional data analysis methods, knowledge graphs provide more comprehensive and intuitive information, helping decision-makers quickly understand the disease situation, formulate more scientific and reasonable prevention and control strategies, rationally allocate medical resources, and improve the targeting and effectiveness of disease prevention and control. Attached Figure Description

[0049] Figure 1 This is a flowchart of a method for extracting health epidemiological classification information of disease outbreaks according to the present invention. Detailed Implementation

[0050] The following will refer to the appendices in the embodiments of the present invention. Figure 1 The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the embodiments described below are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. In order to provide a clearer explanation and description of the technical solutions and implementation methods of the present invention, specific examples of preferred implementations of the technical solutions of the present invention are introduced below.

[0051] The specific implementation of this invention is based on a complete process of "data acquisition → data preprocessing → natural language processing → machine learning classification → knowledge graph construction and application → information application". All links are interconnected to form a closed-loop disease information processing system.

[0052] 1. Data Collection Phase: ETL tools were used to extract diagnostic information and treatment records of cases from the information systems of major hospitals in the city; disease surveillance data, such as the daily number of new cases and the distribution area of ​​the disease, was obtained from the CDC's monitoring database; web crawlers were used to collect disease-related information posted by residents from local social media platforms, such as feedback on disease prevention and control measures and local disease information; and the latest research results related to this disease were retrieved and downloaded from medical literature databases. During the collection process, the connection information and extraction rules of each data source were configured to ensure the integrity and timeliness of the data.

[0053] 2. Data Preprocessing Stage: The collected data is cleaned to remove invalid records from the hospital information system due to data entry errors, such as case records with negative ages. Data deduplication algorithms are used to remove duplicate posts from social media data. For missing values, if a case has missing symptom descriptions, symptom information is extracted from the relevant examination reports to supplement the missing values; otherwise, it is marked as requiring further verification. Data in different formats is standardized, converting date formats from the hospital system to a standard format and standardizing the professional terminology used in case symptom descriptions.

[0054] 3. Natural Language Processing Stage: The text data from the case reports is input into a CRF-based named entity recognition model to identify entities such as symptoms, pathogens, and locations. For example, from a case report stating "The patient developed cough and fatigue symptoms on October 5, 2024, and recently visited a farmers' market," "October 5, 2024" is identified as a time entity, "cough" and "fatigue" as symptom entities, and "farmers' market" as a location entity. Then, dependency parsing is used to determine the relationships between entities, clarifying the associations between symptoms and the patient, time, and location.

[0055] 4. Machine Learning Classification Stage: Preprocessed structured data, including basic case information, symptom data, and transmission path data, are used as features and input into a classification model composed of logistic regression, decision trees, and neural networks for training and classification. For example, a logistic regression model determines whether a case is a confirmed case, a decision tree model classifies the transmission risk level of cases, and a neural network model predicts the disease progression trend of cases. During training, model parameters are continuously adjusted to improve the model's accuracy and generalization ability. In the classification of influenza A cases, the final model may use the "age weight" of logistic regression, the "contact history rule" of decision trees, and the "symptom combination pattern" of neural networks to make a joint judgment, ensuring accuracy while explaining "why it was judged as a suspected case" (e.g., "age 60+ (logistic regression) + history of contact with confirmed cases (decision tree) + high feature matching degree of fever + dry cough (neural network)").

[0056] 5. Knowledge Graph Construction and Application Stage: The categorized disease information is used to construct a disease knowledge graph through entity recognition and relation extraction techniques. Cases, locations, symptoms, and transmission relationships are stored and managed as nodes and edges in the Neo4j graph database. For example, an edge is created between a case node and a symptom node to represent the symptoms of the case; an edge is created between a case node and a location node to represent the case's activity trajectory. The knowledge graph visually displays the network structure of disease transmission, providing a basis for disease prevention and control decisions, such as analyzing disease transmission hotspots based on the knowledge graph and rationally allocating medical resources for key prevention and control measures.

[0057] 6. Information Application Stage: This part is the final output stage of the entire information extraction process. It takes over the structured information generated in the "Knowledge Graph Construction" stage (including case classification data, transmission chain relationships, risk area marking, and intervention measure associations), focuses on the four core scenarios of emergency response, risk management, resource allocation, and policy formulation in disease control work, and provides accurate data support for disease control departments at all levels, medical institutions, and government prevention and control command agencies.

[0058] (a) Application of transmission risk assessment

[0059] Transmission chain visualization and tracing, and at-risk population identification

[0060] Input requirement: Query the scope of close contacts and secondary contacts of a confirmed case.

[0061] Implementation process: Retrieve the "activity trajectory entity" and "contact relationship edge" of the case from the knowledge graph (such as contact with A for 1 hour in a restaurant, or taking the subway line 3 with B, etc.); Combine spatiotemporal dimension analysis (such as whether the contact time is within the incubation period of the virus, and whether the contact space is a closed environment), draw the transmission chain map through visualization tools (such as GIS map); mark high-risk groups (such as close contacts who have not been isolated) and medium-risk groups (such as spatiotemporal companions), and output the "List of Risk Groups and Control Recommendations".

[0062] Output formats: interactive transmission chain map, exportable Excel spreadsheet of at-risk populations, and control priority ranking (from high to low infection probability).

[0063] Disease trend prediction and risk area delineation

[0064] Input requirement: Predict the number of new disease cases and the range of high-risk areas in a certain region over the next 7 days.

[0065] Implementation process: Extract "historical case data" (such as daily new cases and case distribution), "regional attributes" (such as population density and number of transportation hubs), and "intervention measure records" (such as lockdown time and nucleic acid screening frequency) from the knowledge graph; import them into time series prediction models (such as ARIMA and LSTM), combine them with the impact coefficient of intervention measures (such as a 30% decrease in infection rate after lockdown), and generate a trend prediction curve; based on the prediction results, overlay regional case density thresholds (such as 5 new cases per 10,000 people being high risk), and delineate high, medium, and low risk areas on the GIS map.

[0066] Output formats: Disease trend prediction chart (including 95% confidence interval), risk area classification map, and prevention and control intensity recommendations (such as "stay at home" in high-risk areas and "people in medium-risk areas").

[0067] (II) Formulation and Application of Prevention and Control Strategies

[0068] Based on the structured information (case distribution, transmission chain, risk areas, population characteristics, etc.) output from the knowledge graph construction process, the system focuses on "tiered prevention and control and precise policy implementation" and generates scientific strategies in four steps, covering three dimensions: regional prevention and control, population management, and measure optimization.

[0069] Core input data includes basic disease data, transmission characteristic data, population risk data, and data on the effectiveness of existing measures.

[0070] Dynamic risk level determination: Establish a risk level assessment model, using case density, transmission speed, and infection rate of key populations as core indicators, and combine real-time data with virus mutation characteristics to calculate the risk level and output the "Regional Disease Risk Level Classification Report".

[0071] Regional prevention and control strategies are generated: high-risk areas implement "stay-at-home orders with door-to-door service", medium-risk areas implement "avoid unnecessary outings", and low-risk areas strengthen routine prevention and control and early warning.

[0072] Precise population control and dynamic optimization: Implement "one person, one file" management for key populations such as the elderly and patients with underlying diseases, implement health monitoring and green channels, and make dynamic adjustments based on indicators such as the number of new cases, infection rate, and implementation rate of measures.

[0073] (III) Resource allocation planning and application: With the goal of "supply and demand matching and efficient utilization", a resource scheduling model is constructed, covering four major categories: pathogen detection resources, isolation resources, medical supplies and human resources, to ensure that resources are accurately allocated to the areas and links that need them most.

[0074] Core input data includes resource stock, regional demand, spatiotemporal correlation, and disease trend data.

[0075] Resource demand forecasting and gap calculation: Generate a "Resource Supply and Demand Gap Table" through demand forecasting models and trend forecasts.

[0076] Resource allocation by type: including pathogen detection resource allocation (reagents, sampling personnel, testing vehicles), isolation resource allocation (hotels, beds, supplies), medical supply allocation (emergency and routine), and human resource allocation (medical staff, epidemiological investigators, volunteers).

[0077] Resource allocation effectiveness monitoring and optimization: Real-time tracking of delivery rate, utilization rate, and demand fulfillment rate, and optimization through measures such as adjusting transportation, activating backup isolation points, and increasing manpower.

[0078] The Disease Outbreak Health Epidemiology Classification Information Extraction System includes a multi-source data acquisition module, which is used to obtain original disease-related data such as case diagnosis and treatment records, epidemiological survey data, public opinion texts, and population activity trajectories from hospital information systems, disease control center databases, social media platforms, environmental and meteorological monitoring devices, etc.

[0079] The data preprocessing module is used to clean, desensitize, denoise, standardize the format, fill in missing data and align the time and space of the raw data, and output the normalized data to the downstream module.

[0080] The natural language processing module is used to receive the normalized data, perform word segmentation, named entity recognition, relation extraction and event extraction on the text information therein, and output the entity, relation and feature information of the disease elements;

[0081] The machine learning classification module is used to receive the disease element entity, relationship and feature information, classify and determine the case category, transmission scenario type, risk level and cluster event based on the pre-trained model, and generate classification labels;

[0082] The knowledge graph construction module is used to receive the classification tags and disease element entities and relationships, align entities and mapping relationships, and perform reasoning verification to form a disease knowledge graph.

[0083] The information application module is used to provide disease monitoring and early warning, source tracing analysis, risk stratification and resource scheduling based on the disease knowledge graph, and to feed back manual correction, early warning hit status and false alarms to the machine learning classification module, natural language processing module and data preprocessing module to realize model update, rule revision and knowledge graph ontology evolution.

[0084] The contents not described in detail in this description are existing technologies known to those skilled in the art. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for extracting health epidemiological classification information of disease outbreaks, characterized in that, Includes the following steps: S1, multi-source data acquisition, builds a data fusion platform, and uses ETL tools, distributed acquisition frameworks or message queue middleware to obtain case information, disease monitoring data, public feedback information and research results from medical institution information systems, disease control center monitoring databases, community health service platforms, social media and medical literature databases, ensuring the comprehensiveness and diversity of data sources; S2. Preprocess the collected data, including: removing noisy and duplicate data using rule-driven or machine learning data cleaning algorithms; handling missing values ​​through mean imputation, K-nearest neighbor imputation, or labeling for verification; and standardizing data from different sources, including standardizing date formats, stemming symptom texts, and part-of-speech tagging, to ensure consistent data formats and semantic comparability. S3 utilizes natural language processing technology to process unstructured text, including named entity recognition, dependency parsing, semantic understanding, and semantic role labeling. It identifies key information such as case symptoms, pathogens, locations, and transmission behaviors from case reports, social media texts, and medical literature, and transforms them into structured data. S4. Based on machine learning algorithms, feature extraction and classification model training are performed on structured data. The model may include logistic regression, decision tree, neural network, etc. The input features cover basic case information, symptom characteristics, transmission path, contact history and activity trajectory. The output classification results include confirmed cases, suspected cases, close contacts and transmission risk level, realizing automated disease information classification. S5 uses entity recognition and relation extraction technology to construct a disease knowledge graph. It uses cases, locations, symptoms, and transmission events as nodes, and transmission relationships, activity trajectories, and symptom associations as edges. The attributes of the edges include contact time, symptom onset time, etc. It uses a graph database for storage and management and realizes visualization display. S6, The knowledge graph and classification results are combined and applied to disease prevention and control decision support, including transmission chain tracing and at-risk population location, disease trend prediction and risk area delineation, prevention and control strategy generation and dynamic optimization, as well as supply and demand matching and scheduling of medical resources, isolation resources, materials and manpower.

2. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The multi-source data acquisition further includes: a) For medical institution systems, configure database connection addresses, user permissions, and extraction rules, and periodically extract case diagnosis and treatment records; b) For social media data, use web crawlers to set keyword strategies and time windows to obtain disease-related posts and comments; c) Automatically retrieve and download research findings related to the current disease from literature databases to ensure the timely inclusion of research progress information.

3. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The data preprocessing further includes: a) Delete invalid records in the input data caused by input errors, such as negative age or abnormal date data; b) Use a deduplication algorithm to identify and remove duplicate data to avoid the same case being counted repeatedly; c) Standardize and unify the technical terms and colloquial expressions in the text data to ensure semantic consistency.

4. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The natural language processing technology employs named entity recognition methods based on conditional random fields or deep learning BERT models, combined with dependency parsing, to identify semantic relationships between cases, symptoms, pathogens, geographical locations, and transmission events, and to form logically complete structured information based on semantic role labeling.

5. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The machine learning classification model is implemented in the following way: a) Using a logistic regression model, combined with basic characteristics such as age, gender, and symptom severity, cases are classified as either confirmed or suspected cases; b) Using a decision tree model, the transmission risk level of cases is classified based on characteristics such as case contact history and activity trajectory; c) Using neural network models, model and predict complex symptom combinations and potential nonlinear relationships, and output the case progression trend.

6. The method for extracting epidemiological classification information of disease outbreaks according to claim 5, characterized in that, The classification model is trained using cross-validation to evaluate accuracy, and its performance is optimized by adjusting parameters such as regularization coefficient, tree depth, number of network layers, and learning rate, thereby achieving a classification accuracy of over 95%.

7. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The construction of the disease knowledge graph includes: a) Use case, symptoms, location, and transmission relationship as nodes; b) Treat the relationships between cases and symptoms, cases and locations, and cases and transmission events as edges; c) Add attributes such as contact duration, contact frequency, and symptom onset time to the edge; d) The Neo4j graph database is used for storage, and the creation, querying and updating of nodes and edges are implemented using the Cypher language.

8. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The disease prevention and control decision support includes: a) Visual tracing of transmission chains and generation of risk population lists based on knowledge graphs; b) Prediction of new disease cases and risk area classification based on time series models; c) Develop regional prevention and control strategies based on case distribution, transmission characteristics, and population risk characteristics; d) Optimize resource scheduling based on real-time supply and demand gap calculations to achieve precise allocation of medical supplies, isolation beds, and protective personnel.

9. The method for extracting epidemiological classification information of disease outbreaks according to claim 1, characterized in that, The disease prevention and control strategy is generated with a dynamic optimization mechanism. It evaluates the effectiveness of the strategy by extracting indicators such as the rate of change in the number of new cases, the proportion of cases in high-risk areas, the infection rate of key populations, and the implementation rate of measures in real time. It also automatically adjusts the risk level and the intensity of prevention and control when the case growth rate exceeds the threshold or when there are no new cases for a period of time.

10. A system for extracting health epidemiological classification information on disease outbreaks, characterized in that, It includes a multi-source data acquisition module, used to acquire raw disease-related data such as case records, epidemiological survey data, public opinion texts, and population activity trajectories from hospital information systems, disease control center databases, social media platforms, environmental and meteorological monitoring devices, etc. The data preprocessing module is used to clean, desensitize, denoise, standardize the format, fill in missing data and align the time and space of the raw data, and output the normalized data to the downstream module. The natural language processing module is used to receive the normalized data, perform word segmentation, named entity recognition, relation extraction and event extraction on the text information therein, and output the entity, relation and feature information of the disease elements; The machine learning classification module is used to receive the disease element entity, relationship and feature information, classify and determine the case category, transmission scenario type, risk level and cluster event based on the pre-trained model, and generate classification labels; The knowledge graph construction module is used to receive the classification tags and disease element entities and relationships, align entities and mapping relationships, and perform reasoning verification to form a disease knowledge graph. The information application module is used to provide disease monitoring and early warning, source tracing analysis, risk stratification and resource scheduling based on the disease knowledge graph, and to feed back manual correction, early warning hit status and false alarms to the machine learning classification module, natural language processing module and data preprocessing module to realize model update, rule revision and knowledge graph ontology evolution.

Images