Knowledge graph fused health degree radar map design method and system

Abstract

The invention relates to the technical field of medical data analysis, in particular to a health degree radar map design method and system fused with a knowledge graph. The system comprises the following steps: acquiring medical health information data of a medical system; establishing a health index-influence factor knowledge graph based on the medical health information data; acquiring real-time health data of the user; performing user individualized health state analysis processing on the real-time health data of the user through the health index-influence factor knowledge graph to generate user individualized health state data; establishing a health dimension logic dependence tree model based on the health index-influence factor knowledge graph; transmitting the individualized health state data of the user to a health dimension logic dependence tree model to carry out evaluation processing on each health dimension, and generating individualized health dimension evaluation data of the user; and designing a user health degree radar map based on the user individualized health dimension evaluation data. According to the invention, the design of the visual radar map reflecting the accurate health condition of the user is realized.

Description

A Method and System for Designing Health Radar Charts Integrating Knowledge Graphs Technical Field

[0001] This invention relates to the field of medical data analysis technology, and in particular to a method and system for designing a health radar chart that integrates knowledge graphs. Background Technology

[0002] With the rapid development of medical informatization and health management technologies, health assessment systems have become important tools in chronic disease management and daily health monitoring. Radar charts, due to their ability to intuitively display the characteristics and differences of multi-dimensional data, are widely used for visualizing health status. However, existing health radar chart design systems mostly rely on basic indicator statistics and simple visualization algorithms. In the correlation analysis between indicators and influencing factors, they do not incorporate knowledge graphs, but rely solely on static analysis based on preset rules. This makes it impossible to dynamically capture the complex causal coupling relationship between the two, and also difficult to quantify the co-occurrence effect of influencing factors. As a result, the assessment results lack causal traceability. Furthermore, at the individualized analysis level, they mostly focus on real-time vital sign data, neglecting the integration of behavioral records and medical history data. They also fail to conduct differentiated indicator node analysis based on individual user characteristics and adopt uniform assessment standards, making it difficult to build accurate health profiles. Radar charts are designed only based on basic assessment data, without incorporating data volatility, skewness, and trend characteristics to optimize their form, and without labeling health risk factors. They can only display static data distribution and cannot reflect risk trends and key influencing factors. Summary of the Invention

[0003] Based on this, the present invention provides a method and system for designing a health radar chart that integrates knowledge graphs, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a health radar chart design system integrating knowledge graphs is provided, comprising the following modules:

[0005] The knowledge graph design module is used to acquire medical and health information data from the medical system; based on the medical and health information data, a knowledge graph of health indicators and influencing factors is established, generating a health indicator-influencing factor knowledge graph;

[0006] The health status analysis module is used to acquire users' real-time health data; it performs personalized health status analysis on users' real-time health data through a health indicator-influencing factor knowledge graph, and generates personalized health status data for users.

[0007] The health dimension assessment module is used to establish a mathematical model of the logical dependency tree of health dimensions based on the knowledge graph of health indicators and influencing factors, and generate a logical dependency tree model of health dimensions; it transmits the user's individual health status data to the logical dependency tree model of health dimensions for assessment of each health dimension, and generates user-individual health dimension assessment data;

[0008] The Health Radar Chart Design Module is used to design and generate user health radar charts based on individualized health dimension assessment data.

[0009] Furthermore, the knowledge graph design module includes the following functions:

[0010] To acquire medical and health information data from the healthcare system;

[0011] Perform structured preprocessing on medical and health information data to generate structured health information data;

[0012] Entity extraction is performed on the structured health information data to obtain structured health entity data.

[0013] The structured health entity data is divided into health indicator entities and health influencing factor entities to obtain health indicator entity data and health influencing factor entity data respectively.

[0014] Based on entity data of health indicators and entity data of health influencing factors, causal characteristic analysis of health indicators and influencing factors is performed to generate causal characteristic data of health indicators and influencing factors.

[0015] The entity data of health indicators is processed to classify health indicator types, generating health indicator type data. Then, a preliminary knowledge graph of health indicator types is constructed based on the causal feature data of health indicators and influencing factors, generating a preliminary knowledge graph of health indicators and influencing factors.

[0016] Based on the preliminary health indicator-influencing factor knowledge graph, a graph analysis of the co-occurrence coupling mapping of influencing factors is performed to generate a health indicator-influencing factor knowledge graph.

[0017] Furthermore, causal characteristic analysis of health indicators and influencing factors based on entity data of health indicators and entity data of health influencing factors includes:

[0018] Based on the entity data of health indicators and entity data of health influencing factors, a correlation analysis of health indicators and health influencing factors is performed to generate correlation data between health indicators and influencing factors. Then, causal feature analysis is performed on the correlation data between health indicators and influencing factors to generate causal feature data between health indicators and influencing factors.

[0019] Furthermore, the real-time health data of the user in the health status analysis module includes real-time user vital sign monitoring data, real-time user behavior record data, and user health history data.

[0020] Furthermore, the health status analysis module includes the following functions:

[0021] Obtain users' real-time health data;

[0022] The user's real-time health data is transmitted to the health indicator-influencing factor knowledge graph for user health node mapping processing, generating user health graph mapping data.

[0023] Perform health indicator association path analysis on user health graph mapping data to obtain user health indicator association path depth data;

[0024] Based on the deep data of the association path of user health indicators, the influence characteristics of indicator nodes in the knowledge graph are analyzed to generate influence characteristic data of user health indicator nodes.

[0025] Health domain dimension analysis and processing are performed using a health indicator-influencing factor knowledge graph to generate health domain dimension data.

[0026] Based on health-related data and user health indicator node impact feature data, we perform personalized health status analysis and processing to generate personalized user health status data.

[0027] Furthermore, the analysis and processing of individualized user health status based on the influence feature data of user health indicator nodes includes:

[0028] Based on the impact feature data of user health indicator nodes, perform user health indicator specificity analysis to generate user health indicator specific data;

[0029] Based on the health domain dimension data, analyze the user health indicator specific data to obtain the user health dimension related candidate indicator data;

[0030] Based on the candidate indicator data related to user health dimensions, we perform individualized health status analysis and processing to generate individualized user health status data.

[0031] Furthermore, the health dimension assessment module includes the following functions:

[0032] Based on the knowledge graph of health indicators and influencing factors, we conducted analysis on the dependence characteristics of health indicators and the contributing factors of health indicators, and generated data on the dependence characteristics of health indicators and the contributing factors of health indicators, respectively.

[0033] Based on health domain dimension data and health indicator dependency feature data, a model architecture for a health dimension logical dependency tree is established, generating a health dimension logical dependency tree model architecture.

[0034] The model health assessment weights are designed and generated by using health indicator dependency feature data and health indicator contribution factor data.

[0035] The model parameter weights of the health dimension logical dependency tree model architecture are adjusted by using model health assessment weights to perform health assessment, thereby generating a health dimension logical dependency tree model.

[0036] The user's individual health status data is transmitted to the health dimension logical dependency tree model for evaluation of each health dimension, generating user-individual health dimension evaluation data.

[0037] Furthermore, the health indicator dependency feature data includes the dependency features between subsets of health indicator entity data and the dependency features between subsets of health indicator entity data and subsets of health influencing factor entity data.

[0038] Furthermore, the health radar chart design module includes the following functions:

[0039] We analyze the data volatility, skewness, and trend characteristics of each health dimension in the user's individualized health dimension assessment data, and generate user health volatility data, user health skewness data, and user health trend characteristic data respectively.

[0040] By using a preset health radar morphology decision, user health fluctuation data, user health skewed data, and user health trend feature data are processed to transform the radar morphology features of user health, generating user health radar morphology feature data.

[0041] Based on user health volatility data, user health skewness data, and user health trend characteristic data, we conduct user health risk correlation factor analysis to generate user health risk correlation factor data.

[0042] The user health radar chart is designed and generated by using user health risk correlation factor data and user health radar morphology feature data.

[0043] This specification provides a method for designing a health radar chart based on a fused knowledge graph, characterized in that it is executed based on the health radar chart design system based on a fused knowledge graph as described in claim 1, and the method includes the following steps:

[0044] Acquire medical and health information data from the medical system; build a knowledge graph of health indicators and influencing factors based on the medical and health information data, and generate a health indicator-influencing factor knowledge graph.

[0045] Acquire users' real-time health data; perform personalized health status analysis and processing on users' real-time health data through a health indicator-influencing factor knowledge graph to generate personalized health status data for users;

[0046] A mathematical model of the logical dependency tree of health dimensions is established based on the knowledge graph of health indicators and influencing factors, and a logical dependency tree model of health dimensions is generated; the user's individual health status data is transmitted to the logical dependency tree model of health dimensions for evaluation processing of each health dimension, and user individual health dimension evaluation data is generated.

[0047] A user health radar chart is designed and generated based on user-individualized health dimension assessment data.

[0048] The beneficial effects of this application are as follows: the knowledge graph design module of this invention constructs a health indicator-influencing factor knowledge graph through a systematic process, effectively solving the problems of multi-source heterogeneity and ambiguous relationships in medical and health data. First, it performs structured preprocessing and entity extraction on medical and health information data, accurately classifying it into two core entities: health indicators and influencing factors, laying a clear data foundation for subsequent correlation analysis. Then, through correlation analysis and causal feature mining, it transforms scattered data into logically related structured knowledge, overcoming the limitation of only establishing superficial relationships. Through indicator type classification and co-occurrence coupling mapping optimization of influencing factors, the resulting knowledge graph achieves visualization and traceability of the causal relationship between health indicators and influencing factors, ensuring the standardization of the knowledge system while capturing the complex coupling effects between influencing factors, providing a reliable and reasonable data foundation for subsequent health status analysis. The health status analysis module achieves accurate transformation from multi-source data to individualized health profiles, significantly improving the targeting and accuracy of health assessments. It integrates real-time vital signs, behavioral records, and health history data from multiple dimensions, avoiding the one-sidedness of single-data-dimensional analysis. Through structured processing and knowledge graph node mapping, it efficiently links personalized user data with a standardized knowledge system, solving the problem of the disconnect between individual data and general health knowledge. By leveraging in-depth analysis of association paths and mining the influence characteristics of indicator nodes, combined with specificity analysis and candidate indicator screening, it can accurately capture the unique health indicator influence paths and key characteristics of each user. The generated personalized health status data not only matches the user's actual health condition but also provides accurate and detailed input for subsequent health dimension assessments. The health dimension assessment module addresses the limitations of traditional health dimension assessments, such as strong subjectivity and ambiguous logical hierarchy, by constructing a standardized and quantifiable assessment model. Relying on a health indicator-influencing factor knowledge graph, it accurately mines the dependency characteristics between indicator subsets and between indicators and influencing factor subsets, while quantifying the contribution factors of health indicators. This provides data support for model architecture design and weight setting, solving the problem of weights relying on empirical values ​​in assessments. By constructing and adjusting the parameters of a logical dependency tree model for health dimensions, scattered health indicators are transformed into a hierarchical assessment system. This system can achieve multi-dimensional and accurate assessments by combining individualized user health status data. It ensures the standardization of assessment criteria while adapting to individual differences through factor analysis and weight optimization. The generated individualized health dimension assessment data provides a logically clear and reliable core basis for subsequent visualization. The health radar chart design module effectively overcomes the limitation of radar charts that only present basic assessment data. By analyzing the volatility, skewness, and trend characteristics of each health dimension, it captures the dynamic changes in health status. At the same time, it combines data features to mine health risk-related factors, allowing the radar chart to not only display the current distribution of health dimensions but also reflect potential risks and changing trends.By leveraging preset radar morphology decisions, the system achieves precise conversion of data features into graphical features, enabling the dynamic attributes and risk information of health data to be presented intuitively through visual language. This solves the pain point of traditional visualization in failing to showcase the linkage of multiple variables and risk correlations, providing more intuitive and comprehensive visualization support for health intervention decisions.

[0049] Therefore, the health radar chart design system integrating knowledge graphs of this invention addresses the problems of existing systems lacking knowledge graph support and struggling to capture the dynamic relationships between indicators and influencing factors. The knowledge graph design module constructs a health indicator-influencing factor knowledge graph containing dynamic causal relationships through structured preprocessing, entity partitioning, causal feature analysis, and co-occurrence coupling mapping of influencing factors. This not only visualizes the complex relationships between indicators and influencing factors but also quantifies the co-occurrence effects of influencing factors, giving health assessment results clear causal traceability and overcoming the limitations of traditional static analysis based on preset rules. Secondly, addressing the pain point of insufficient individualized analysis, the health status analysis module integrates multi-dimensional data from real-time vital signs, behavioral records, and health history. Through knowledge graph node mapping, in-depth analysis of association paths, and indicator-specific screening, it achieves a shift from general assessment standards to individual characteristic adaptation, accurately constructing a user-specific health profile and solving the problem of profile distortion caused by a single data dimension and uniform assessment standards. Furthermore, the health dimension assessment module, based on the indicator dependency characteristics and contribution factors mined from knowledge graphs, constructs a logical dependency tree model for health dimensions and optimizes the assessment weights. This transforms subjective, experience-based assessments into quantitative, hierarchical assessments, providing a reliable data foundation for subsequent visualization. Finally, the health radar chart design module overcomes the limitations of traditional static displays. By analyzing data volatility, skewness, and trend characteristics, it dynamically optimizes the chart's form and integrates annotations of health risk-related factors. This allows the radar chart to not only intuitively present the current distribution of health dimensions but also clearly reflect potential risks and changing trends, significantly enhancing the decision-support value of the visualization results. Attached Figure Description

[0050] Figure 1 is a schematic diagram of the module flow of a health radar chart design system that integrates knowledge graphs according to the present invention.

[0051] Figure 2 is a detailed functional flowchart of the health dimension assessment module in Figure 1;

[0052] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0053] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0054] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. Functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods. The term "and / or" as used herein includes any and all combinations of one or more of the associated items listed.

[0055] To achieve the above objectives, please refer to Figures 1 and 2. This invention provides a method and system for designing a health radar chart that integrates knowledge graphs. In an embodiment of this invention, please refer to Figure 1, which is a schematic flowchart of the module flow of a health radar chart design system that integrates knowledge graphs. The health radar chart design system that integrates knowledge graphs includes the following modules:

[0056] S1: Knowledge graph design module, used to acquire medical and health information data from the medical system; based on the medical and health information data, to build a knowledge graph of health indicators and influencing factors, and generate a health indicator-influencing factor knowledge graph;

[0057] In this embodiment of the invention, medical data is integrated with electronic medical record systems, physical examination center databases, and public health monitoring platforms via a medical data interface to acquire medical and health information data such as clinical diagnostic records, physical examination indicator data, dietary nutrition records, and exercise monitoring data. A structured preprocessing mechanism based on the OMOP CDM standard is employed to perform format conversion, text cleaning, and standardization on the acquired data. Irrelevant information such as headers and footers is removed, OCR recognition errors are corrected, and unstructured medical record text is converted into UTF-8 encoded structured text. Then, the text is segmented according to paragraph and sentence boundaries to generate structured health information data. Entity extraction is performed using a BERT-based pre-trained language model framework to identify and classify health indicator entities and health influencing factor entities. Health indicator entities include laboratory indicators such as blood pressure, blood sugar, and blood lipids, while health influencing factor entities include dietary structure, exercise frequency, and genetic history, generating structured health entity data. Entity normalization technology is used to map the extracted entities to standard medical terminology databases such as SNOMED CT and LOINC, completing the classification of the two types of entities and obtaining health indicator entity data and health influencing factor entity data. Association analysis based on Pearson correlation coefficients generates associated data. Then, Granger causality tests are used to mine causal features from this associated data, generating causal feature data of health indicators and influencing factors. Health indicator entity data are categorized according to physiological systems such as the circulatory and digestive systems. Triples are constructed based on indicator types to form a preliminary knowledge graph. Co-occurrence network analysis is used to mine coupling relationships among influencing factor entities in the preliminary graph, calculating entity co-occurrence frequency and association strength, supplementing coupling edges between entities, and generating the health indicator-influencing factor knowledge graph.

[0058] S2: Health Status Analysis Module, used to acquire users' real-time health data; through the health indicator-influencing factor knowledge graph, it performs personalized health status analysis on users' real-time health data to generate personalized health status data for users;

[0059] In this embodiment of the invention, real-time user vital sign monitoring data such as heart rate, blood pressure, and body temperature are collected every 5 minutes through a vital sign sensor array. Real-time user behavior records such as daily steps and sleep duration are obtained through a smart device interface. User health history data, such as past diagnostic reports and surgical records, are retrieved through a medical data interface. A structured processing mechanism, identical to that used in the knowledge graph design module, is employed to perform field alignment, data cleaning, and normalization on the three types of data. Data of different formats is uniformly converted into a structured format conforming to the OMOP CDM standard, generating structured real-time user health data. Based on an entity matching algorithm, entities in the user's structured real-time health data are compared with nodes in the health indicator-influencing factor knowledge graph to establish a one-to-one mapping relationship, generating user health graph mapping data. A depth-first traversal algorithm is used to traverse the associated paths in the mapping data, calculating the shortest path length between indicator nodes to obtain user health indicator association path depth data. Node influence weights are calculated based on path depth; shorter paths have higher weights. Combined with node attributes, influence feature analysis is completed, generating user health indicator node influence feature data. By employing subgraph partitioning technology within a knowledge graph, the graph is divided into domains based on dimensions such as cardiovascular health and metabolic health, generating health-related domain data. Based on the influence feature data of indicator nodes, an indicator specificity index is calculated to filter out specific indicators that deviate from the group mean. These are then combined with the health-related domain data to match candidate indicators for the corresponding dimensions. Finally, individualized health status analysis is completed through weighted calculation of indicator weights, generating personalized health status data for each user.

[0060] S3: Health Dimension Assessment Module, used to establish a mathematical model of the logical dependency tree of health dimensions based on the knowledge graph of health indicators and influencing factors, and generate a logical dependency tree model of health dimensions; transmit the user's individual health status data to the logical dependency tree model of health dimensions for assessment processing of each health dimension, and generate user's individual health dimension assessment data;

[0061] In this embodiment of the invention, relying on a health indicator-influencing factor knowledge graph, the analytic hierarchy process (AHP) is used to analyze the dependencies between subsets of health indicator entities. Simultaneously, the association strength between subsets of indicator entities and subsets of influencing factor entities is calculated using mutual information entropy, generating health indicator dependency feature data composed of two types of dependency features. The entropy weighting method is used to quantify the contribution of each health indicator in its corresponding dimension, and the quantification results are corrected based on expert consensus, generating health indicator contribution factor data. Based on the health domain dimension data, the top-level node of the model is determined. Second-level and third-level nodes are constructed according to the hierarchical relationship in the indicator dependency feature data. The top-level node represents health dimensions such as cardiovascular and respiratory health; the second-level nodes represent the core indicator groups under this dimension; and the third-level nodes represent specific indicators and influencing factors, forming a health indicator logical dependency tree model architecture. The association strength in the health indicator dependency feature data and the contribution factor data are weighted and fused to calculate the evaluation weight of each node, generating the model health evaluation weight. The weight values ​​are assigned to the corresponding nodes, and the node parameters are adjusted through iterative calculation until the deviation between the model output result and the standard sample is lower than a preset threshold, completing the model parameter weight adjustment and generating the health indicator logical dependency tree model. The indicator values ​​from the user's individual health status data are input into the model, and the values ​​are accumulated and calculated through the weight transmission of each level node to obtain the comprehensive evaluation value of each health dimension, thus generating the user's individual health dimension evaluation data.

[0062] S4: Health Radar Chart Design Module, used to design and generate user health radar charts based on individualized health dimension assessment data.

[0063] In this embodiment of the invention, the standard deviation is used to calculate the volatility of the health dimension assessment data, the skewness coefficient is used to analyze the skewness characteristics of the data distribution, and the sliding window algorithm is used to fit the trend of the assessment data over a continuous time period, generating user health volatility data, skewness data, and trend feature data respectively. A radar morphology decision rule library is preset, where volatility data corresponds to the thickness of the radar chart lines, skewness data corresponds to the depth of the graphic fill color, and trend feature data corresponds to the solidity or voidness of the lines. The three types of feature data are substituted into the rule library for matching and transformation to generate health radar morphology feature data containing line style and fill color parameters. The correlation relationship between volatility, skewness, and trend feature data and known health risks is analyzed through association rule mining algorithm, the risk correlation degree is calculated, and user health risk correlation factor data is generated, where high correlation factors are marked as core risk points. Canvas drawing technology is used to construct the radar chart drawing environment, the coordinates of the canvas center point and the maximum radius are set, the number of radiation axes is determined according to the number of health dimension data, and the angle between each axis is 360 degrees divided by the total number of dimensions. Based on the user's individualized health dimension assessment data, the radius of the data points on each axis is calculated. Combined with radar morphology feature data, polygonal lines and filled areas are drawn. The risk factor names and correlation degrees are marked at the axis positions corresponding to the core risk points to generate a user health radar chart.

[0064] Furthermore, the knowledge graph design module includes the following functions:

[0065] To acquire medical and health information data from the healthcare system;

[0066] In this embodiment of the invention, data is integrated with the electronic medical record system of a tertiary hospital, the core database of a regional health checkup center, and a public health monitoring platform, relying on a standard-based medical data interface. The interface uses HTTPS for encrypted transmission and employs OAuth 2.0 authentication to ensure the security and compliance of the data acquisition process. The integrated medical and health information data covers multiple dimensions: clinical diagnostic records from the electronic medical record system, including chief complaint, present illness, past medical history, and laboratory test results; annual health checkup data from the health checkup center database, covering 23 basic indicators such as height, weight, blood pressure, blood glucose, and blood lipids; and health management records for chronic disease patients from the public health monitoring platform, including daily dietary types and intake, and weekly exercise duration and intensity. Data transmission uses an incremental synchronization mechanism, triggering a data retrieval operation every hour, acquiring only newly added and changed data entries. After synchronization, a data integrity verification mechanism checks the number of fields and data format to ensure the acquired data is complete and conforms to preset specifications.

[0067] Perform structured preprocessing on medical and health information data to generate structured health information data;

[0068] In this embodiment of the invention, a structured preprocessing workflow based on the OMOP CDM standard is used to systematically process the acquired multi-source medical and health information data. First, text cleaning is performed, using regular expressions to match and remove irrelevant information such as headers, footers, and doctor signatures from electronic medical records. An edit distance-based text correction algorithm is then used to correct numerical errors and spelling errors in terms caused by OCR recognition. Next, format standardization is performed, converting PDF-formatted physical examination reports and XML-formatted monitoring data into a UTF-8 encoded structured text format. Then, text segmentation is performed, using a rule-based and statistically-based word segmentation tool to segment the text into independent sentences according to Chinese sentence boundary features, and then further segmenting the sentences at the word level based on a medical terminology dictionary. Finally, part-of-speech tagging is performed, using a conditional random field algorithm to tag the segmented words with nouns, verbs, adjectives, etc., with a focus on tagging proper nouns corresponding to medical terms, generating structured health information data containing text content, part-of-speech tags, and standardized fields.

[0069] Entity extraction is performed on the structured health information data to obtain structured health entity data.

[0070] In this embodiment of the invention, a pre-trained language model framework based on ClinicalBERT is used to perform entity extraction. The model configuration parameters are set as follows: hidden layer dimension 768, Transformer layers 8, attention heads 12, maximum sequence length 512, dropout coefficient 0.2, adapted to the dense technical terminology of medical texts. The model takes sentences from structured health information data as input units and allocates computing resources according to sentence length through a dynamic batch processing mechanism. Short texts such as outpatient records are processed with a sequence length of 256, and long texts such as inpatient medical records are processed with a sequence length of 512. The training process adopts a linear warm-up + cosine decay learning rate scheduling strategy. The learning rate for sequence labeling tasks is set to 5e-6, and the entity class imbalance problem is solved by a cross-entropy loss function weighted by class weights. The model output layer accurately identifies health indicator entities and health influencing factor entities through entity boundary localization and type classification. The health indicator entities include test indicators such as systolic blood pressure, diastolic blood pressure, fasting blood glucose, and total cholesterol, while the health influencing factor entities include related factors such as high-salt diet, weekly exercise frequency, and family history of diabetes. The model generates structured health entity data containing entity name, entity type, and text location.

[0071] The structured health entity data is divided into health indicator entities and health influencing factor entities to obtain health indicator entity data and health influencing factor entity data respectively.

[0072] In this embodiment of the invention, structured health entity data is classified and segmented. First, a mapping dictionary containing two standard medical terminology databases, SNOMEDCT and LOINC, is constructed. The SNOMEDCT dictionary contains 150,000 clinical terms, and the LOINC dictionary contains 80,000 laboratory indicator terms. The dictionaries are regularly updated to include the latest medical terminology. Through string matching and semantic similarity calculation, the extracted entities are compared with the standard terms in the mapping dictionary. Semantic similarity is calculated using the cosine similarity algorithm, with a threshold of 0.8 to determine the matching result. For successfully matched entities, the classification is completed based on the classification attributes of the standard terms: entities categorized as "laboratory indicators" or "physiological measurements" in the LOINC dictionary are classified as health indicator entity data, including indicators such as blood pressure, blood sugar, and blood lipids; entities categorized as "lifestyle," "environmental factors," or "genetic characteristics" in the SNOMEDCT dictionary are classified as health influencing factor entity data, including factors such as dietary structure, exercise frequency, and family history. For entities that fail to match, they are submitted to the medical expert database for manual classification via a manual annotation interface. The classification results are then added to the mapping dictionary, and the classification process is repeated to finally generate two clearly defined categories of entity data.

[0073] Based on entity data of health indicators and entity data of health influencing factors, causal characteristic analysis of health indicators and influencing factors is performed to generate causal characteristic data of health indicators and influencing factors.

[0074] In this embodiment of the invention, Pearson correlation coefficient is used to perform association analysis on health indicator entity data and health influencing factor entity data. Using time series as the dimension, the correlation coefficient is calculated for each pair of indicator-factor data, and the correlation of their numerical changes over a continuous 6-month period is statistically analyzed, generating health indicator-influencing factor association data containing entity pairs, correlation coefficient values, and association periods. Based on this, Granger causality tests are used to mine causal features. Health indicator data is set as the dependent variable, and influencing factor data as the independent variable. A time series regression model is constructed, and the F-test is used to determine whether the independent variable is a Granger cause of the dependent variable. The testing process uses monthly time units, selecting sample data from 12 time points for model training. The model fit is calculated using the sum of squared residuals, and statistically significant causal relationships are selected. For entity pairs with causal relationships, the causal direction and period of action are further labeled. For example, "high-salt diet → increased systolic blood pressure" is labeled as a positive causal relationship with a period of 3 months. Finally, health indicator-influencing factor causal feature data containing entity pairs, causal direction, period of action, and significance level are generated.

[0075] The entity data of health indicators is processed to classify health indicator types, generating health indicator type data. Then, a preliminary knowledge graph of health indicator types is constructed based on the causal feature data of health indicators and influencing factors, generating a preliminary knowledge graph of health indicators and influencing factors.

[0076] In this embodiment of the invention, health indicator entity data is categorized into four main types based on the classification standards of human physiological systems: circulatory system, digestive system, respiratory system, and endocrine system. The circulatory system includes indicators such as blood pressure and heart rate, while the digestive system includes indicators such as liver function and gastrointestinal function, generating health indicator type data. A preliminary knowledge graph is constructed using an RDF triplet model, with health indicator entities and health influencing factor entities as the subject and object, respectively, and causal relationships in the causal feature data as the predicate, forming a triplet structure of "indicator entity - causal relationship - influencing factor entity". Examples include triples such as "systolic blood pressure - positive influence - high-salt diet" and "fasting blood glucose - negative influence - weekly exercise duration". A graph database storage architecture is adopted, with entities as nodes, whose attributes include entity name, type, and standard terminology encoding; and causal relationships as edges, whose attributes include causal direction, period of action, and significance level. Triples are grouped and stored according to the type of health indicator. Triples corresponding to indicators of the same physiological system form subgraphs. Each subgraph is initially associated by sharing influencing factor entities, generating a preliminary health indicator-influencing factor knowledge graph.

[0077] Based on the preliminary health indicator-influencing factor knowledge graph, a graph analysis of the co-occurrence coupling mapping of influencing factors is performed to generate a health indicator-influencing factor knowledge graph.

[0078] In this embodiment of the invention, a co-occurrence network analysis method is used to optimize the preliminary knowledge graph. All influencing factor entities in the preliminary graph are traversed, and the co-occurrence frequency of any two influencing factor entities in the same user's health record is counted using a sliding window mechanism. The window time span is set to 1 month, and the statistical period is 6 months, generating an influencing factor co-occurrence matrix. The association strength between entities is calculated based on the co-occurrence frequency: association strength = co-occurrence frequency / total number of records × 100%. Significantly co-occurring pairs of influencing factor entities are selected using a 15% threshold. For the selected entity pairs, coupling edges are added to the preliminary graph, with edge attributes including co-occurrence frequency, association strength, and co-occurrence period. For example, "high-salt diet" and "alcohol consumption" frequently co-occur in records of hypertensive patients; therefore, a coupling edge is added between them, and relevant attributes are labeled. Simultaneously, a community detection algorithm is used to perform cluster analysis on the graph after adding coupling edges, identifying coupled communities of influencing factor entities, such as the "diet-exercise" community and the "genetic-environment" community. Community nodes are marked with different colors in the graph. By supplementing coupling relationships and community annotations, the association dimensions of the graph are improved, and a health indicator-influencing factor knowledge graph is generated.

[0079] Furthermore, causal characteristic analysis of health indicators and influencing factors based on entity data of health indicators and entity data of health influencing factors includes:

[0080] Based on the entity data of health indicators and entity data of health influencing factors, a correlation analysis of health indicators and health influencing factors is performed to generate correlation data between health indicators and influencing factors. Then, causal feature analysis is performed on the correlation data between health indicators and influencing factors to generate causal feature data between health indicators and influencing factors.

[0081] In this embodiment of the invention, the correlation analysis of health indicators and health influencing factors takes the divided entity data of health indicators and entity data of health influencing factors as the core input, and relies on time series data alignment and multi-dimensional quantitative calculation mechanisms. First, data preprocessing is performed. Using the natural month as the time unit, the entity data of health indicators (such as the monthly average of systolic blood pressure and fasting blood glucose) and entity data of health influencing factors (such as the monthly statistics of daily salt intake and weekly exercise duration) of the same user are matched in time dimension to form one-to-one corresponding sample data pairs. For the matched data, linear interpolation is used to fill in a small number of missing values, and outliers are removed using the interquartile range method—the upper quartile (Q3) and lower quartile (Q1) of each indicator and factor data are calculated, and data exceeding the range [Q1-1.5×(Q3-Q1), Q3+1.5×(Q3-Q1)] are marked as outliers and removed to ensure data quality. The Pearson correlation coefficient is used as the core method for correlation quantification. Since both types of data are continuous numerical time series data, this method can effectively quantify the degree of linear correlation. The data is grouped and calculated based on health indicator types (circulatory system, endocrine system, etc.) and influencing factor types (diet, exercise, genetics, etc.). For example, circulatory system indicators (blood pressure, heart rate) and dietary factors (salt intake, fat intake) are grouped together. Within each group, the correlation coefficient is calculated for each pair of sample data from all users, and then the arithmetic mean is used to summarize the correlation at the group level. A t-test is performed simultaneously to verify significance. A formula for calculating the t-statistic is constructed, using sample size and correlation coefficient values ​​as inputs to calculate the test results. A significance level of 0.05 is set, and only correlation results with a p-value less than 0.05 are retained. The final generated health indicator-influencing factor correlation data includes structured fields such as entity pair name, group category, correlation coefficient value, significance p-value, data time span (default 12 months), and effective sample size, providing a precise screening basis for subsequent causal feature analysis. Based on the causal characteristic analysis of the correlation data between health indicators and influencing factors, a data screening process is first performed, setting a threshold of 0.3 for the absolute value of the correlation coefficient. Only entity pairs meeting this condition are retained for the causal testing process, reducing unnecessary computational costs. The Granger causality test is used as the core analytical method. This method identifies causal relationships by determining whether the independent variable can significantly improve the prediction accuracy of the dependent variable, and is suitable for causal inference of time series data. Before the test, data stationarity processing is performed. The ADF unit root test is used to determine the stationarity of the time series. The test model includes a constant term and a trend term, and the lag order is automatically determined by the Schwarz information criterion. If the data is non-stationary (p-value ≥ 0.05), first-order or second-order differencing is performed until the ADF test p-value is less than 0.05, satisfying the stationarity requirement.A Granger causal regression model was constructed, with entity data of health influencing factors set as independent variables and entity data of health indicators set as dependent variables. The model parameters were configured as follows: lag period range of 1-3 months (set according to the lag characteristics of health impacts). The regression equation was solved using the least squares method, and the residual terms needed to satisfy normality and homoscedasticity. The significance of the causal relationship was determined by the F-test. The sum of squared residuals of the unconstrained model and the constrained model (with lagged terms of independent variables removed) was calculated to obtain the F-statistic. A significance level of 0.05 was set. A p-value of less than 0.05 corresponding to the F-statistic was considered to indicate the existence of a Granger causal relationship. For entity pairs with causal relationships, further characteristic attributes were analyzed: the causal direction was determined by the sign of the regression coefficients (positive correlation indicates positive causation, negative correlation indicates negative causation), the optimal lag period was determined based on the absolute value of the coefficients at different lag periods, and the strength of the causal association was quantified by combining the regression goodness of fit (R²). The final generated health indicator - causal feature data of influencing factors includes fields such as entity pairs, causal direction, optimal lag period, correlation strength (R² value), and significance p-value, providing core attribute support for the construction of causal relationship edges in the knowledge graph.

[0082] Furthermore, the real-time health data of the user in the health status analysis module includes real-time user vital sign monitoring data, real-time user behavior record data, and user health history data.

[0083] Furthermore, the health status analysis module includes the following functions:

[0084] Obtain users' real-time health data;

[0085] In this embodiment of the invention, real-time user health data is acquired through a multi-source data acquisition mechanism. A medical-grade vital sign sensor array collects user physiological indicator data, including a photoelectric heart rate sensor, a piezoelectric blood pressure sensor, and a temperature sensor. The sampling frequency is set to collect heart rate and blood oxygen saturation data every 3 minutes, and systolic blood pressure, diastolic blood pressure, and body temperature data every 15 minutes. The collected data is transmitted in digital signal form after analog-to-digital conversion. User behavior record data is acquired through a smart wearable device interface, including daily steps, exercise duration, and exercise intensity data collected by a triaxial accelerometer, and sleep onset time, sleep duration, and sleep stage data collected by a sleep monitoring sensor. Behavioral data is statistically summarized hourly. A medical data interaction interface connects to the hospital's electronic medical record system and health record database to retrieve the user's past diagnostic records, surgical history, chronic disease history, medication records, and other health history data. The medical history data uses an incremental acquisition mechanism, synchronizing only the latest updated diagnostic and treatment information. Preprocessing was performed on the three types of collected data. Outliers in the vital signs data were removed using the interquartile range method. The upper and lower quartiles of each indicator were calculated, and data exceeding 1.5 times the interquartile range were marked as outliers and removed. Time alignment was performed on the behavioral data, and the timestamps collected by different devices were uniformly converted to UTC time. Text standardization was performed on the medical history data, and unstructured diagnostic descriptions were converted into standardized terms corresponding to ICD-10 disease codes. This resulted in real-time health data for users that included vital signs monitoring, behavioral records, and medical history information. The data fields covered indicator names, collection time, values, units, data sources, and standardized codes.

[0086] The user's real-time health data is transmitted to the health indicator-influencing factor knowledge graph for user health node mapping processing, generating user health graph mapping data.

[0087] In this embodiment of the invention, the user health node mapping process is implemented based on a two-layer entity matching algorithm. First, string exact matching is performed, comparing the indicator names and disease names in the user's real-time health data with the entity names in the health indicator-influencing factor knowledge graph at the character level. Entities that match successfully are directly mapped, and the unique node identifier of the entity in the knowledge graph and the matching type are recorded as "exact match". For entities that do not match string exact, semantic similarity matching is performed, using a semantic computing model based on medical word vectors. The model is constructed based on medical word vectors pre-trained in the PubMed corpus, converting the entity names into 768-dimensional vector representations. The semantic similarity between the user data entity and the knowledge graph entity is calculated using a cosine similarity algorithm, with a similarity threshold of 0.85. Entities with similarity higher than the threshold are mapped, and the matching type is marked as "semantic match". For entities that still do not find a corresponding node after semantic matching, a manual annotation process is performed, submitting the entity information to a medical expert review platform, where professional physicians complete the association annotation between the entity and the knowledge graph node, and the annotation results are added to the matching rule base. After the mapping process is completed, user health graph mapping data is generated. The data structure includes fields such as user data entity name, knowledge graph node ID, node type (health indicator node / influencing factor node), matching method, similarity value, and mapping timestamp. It also records the list of entities that failed to match and the reasons, providing data support for subsequent iterative optimization of the knowledge graph.

[0088] Perform health indicator association path analysis on user health graph mapping data to obtain user health indicator association path depth data;

[0089] In this embodiment of the invention, a depth-first traversal algorithm is used to traverse the knowledge graph subgraph corresponding to the user health graph mapping data. The traversal process starts from the core health indicator node mapped by the user and extends to adjacent nodes according to the edge relationships (causal relationships, association relationships) between nodes in the knowledge graph. The initial traversal depth is set to 5 layers to ensure coverage of the nodes of direct and indirect influencing factors of the indicators. During the traversal, the node sequence and edge attributes of each path are recorded. The path node sequence includes the identifiers and names of the starting node, intermediate nodes, and ending node. The edge attributes include information such as relationship type, association strength, and period of action. The path depth is calculated based on the number of hops between nodes. The number of hops from the starting node to an adjacent node is 1, and the number of hops increases by 1 for each intermediate node. The final path depth is the difference between the number of hops between the ending node and the starting node. A filtering operation is performed on all the paths obtained by traversal to remove weak association paths with an association strength of less than 0.3 and retain statistically significant effective paths. The filtered paths are grouped by the starting node, and the number of paths, average path depth, and longest path depth of each core indicator node are counted. At the same time, the key influencing factor nodes in each path are recorded. The final generated user health metric association path depth data includes fields such as core metric node ID, path sequence, path depth value, association strength, influencing factor nodes, and path validity markers, fully presenting the association path characteristics of user health metrics in the knowledge graph.

[0090] Based on the deep data of the association path of user health indicators, the influence characteristics of indicator nodes in the knowledge graph are analyzed to generate influence characteristic data of user health indicator nodes.

[0091] In this embodiment of the invention, the basic influence weight of nodes is calculated based on path depth using an inverse distance weighting method. The smaller the path depth value, the greater the node's influence weight. The weight calculation formula is: Basic Weight = 1 / Path Depth Value, ensuring that directly related nodes receive higher initial weights. The basic weight is then corrected by considering the association strength of edges in the knowledge graph. The corrected weight = Basic Weight × Association Strength Value. The association strength value is taken from the absolute value of the correlation coefficient in the causal characteristic data of health indicators and influencing factors, ensuring that nodes with higher association strength receive greater corrected weights. The weights of multiple influencing nodes for the same indicator node are normalized, and the proportion of each node's weight to the total weight is calculated to obtain the node's relative influence weight. Based on this, the node's influence characteristics are analyzed, including the degree of influence (relative influence weight value), the direction of influence (positive / negative influence based on causal relationships), the scope of influence (the number of indicator nodes associated with the node), and the timeliness of influence (short-term / medium-term / long-term influence based on the action cycle). The impact characteristics are quantified and graded. Nodes with a relative impact weight ≥ 0.2 are marked as high-impact nodes, nodes with a weight between 0.1 and 0.2 are marked as medium-impact nodes, and nodes with a weight < 0.1 are marked as low-impact nodes. The promoting / inhibiting effect of nodes is marked according to the direction of impact. The duration of impact is divided into short-term (≤ 1 month), medium-term (1-6 months), and long-term (> 6 months) according to the period of effect. The final generated user health indicator node impact characteristic data includes structured fields such as indicator node ID, influencing factor node ID, basic weight, modified weight, relative impact weight, impact direction, impact scope, impact duration, and impact level.

[0092] Health domain dimension analysis and processing are performed using a health indicator-influencing factor knowledge graph to generate health domain dimension data.

[0093] In this embodiment of the invention, a community detection algorithm is used to partition the health indicator-influencing factor knowledge graph. The algorithm is designed based on the principle of maximizing modularity. By iteratively calculating the community affiliation of nodes in the graph, the edge density of nodes within the same community is higher than the edge density between communities. Initially, the algorithm treats each node as an independent community. Then, it calculates the modularity gain after a node is added to an adjacent community. The modularity gain calculation formula is ΔQ=[(Σin+2Σtot) / 2m] - [(Σtot / m)²], where Σin is the sum of edge weights within the community, Σtot is the sum of edge weights of all nodes in the community, and m is the total number of edges in the graph. The community that maximizes the modularity gain is selected as the new affiliation of the node. This iteration is repeated until the modularity value no longer increases. The modularity value threshold is set to 0.6 to ensure the effectiveness of the community partitioning. Based on the community partitioning results, combined with the classification standards of human physiological systems and the domain partitioning specifications of health management, the communities are named as health domain dimensions such as cardiovascular health, metabolic health, respiratory health, digestive health, and neurological health. Each dimension corresponds to a sub-community in the knowledge graph, containing the core health indicator nodes, influencing factor nodes, and the relationships between nodes in that domain. Feature extraction is performed on each health domain dimension, and the number of nodes, edges, core nodes (nodes ranking in the top 10% by degree centrality), and key influencing factors (nodes ranking in the top 10% by betweenness centrality) are counted for each dimension. Simultaneously, the correlation strength between dimensions is calculated, quantifying the degree of dimensional correlation by the number of shared nodes and edge weights across domains. The final generated health domain dimension data includes information such as dimension name, dimension identifier, list of core nodes, list of key influencing factors, number of nodes, number of edges, and dimensional correlation strength, providing a basis for dimensional division in subsequent personalized health status analysis.

[0094] Based on health-related data and user health indicator node impact feature data, we perform personalized health status analysis and processing to generate personalized user health status data.

[0095] In this embodiment of the invention, multi-dimensional feature fusion calculation is implemented. First, based on the influence feature data of user health indicator nodes, the individualized influence index of each indicator is calculated. The influence index = relative influence weight × influence direction coefficient (positive influence is 1, negative influence is -1) × influence timeliness coefficient (short-term is 1.2, medium-term is 1.0, long-term is 0.8). The influence index quantifies the comprehensive effect of each factor on user health indicators. Combining health domain dimension data, user health indicators are grouped by dimension. Scoring calculation is performed on indicators within each dimension. The entropy weight method is used to determine the weight of each indicator. First, the indicator values ​​are standardized to eliminate the influence of dimensions. Then, the information entropy of each indicator is calculated. The indicator weight is calculated based on the information entropy value. The smaller the entropy value, the greater the weight, ensuring that indicators with large data differences receive a higher weight ratio. The comprehensive score of each health domain dimension is calculated by weighted summation. Dimension score = Σ(indicator standardized value × indicator weight × influence index). The score range is mapped to 0-100 points. The higher the score, the better the health status of that dimension. Anomaly analysis was performed on the dimensional scores, comparing each user's scores to the reference score range for healthy individuals. The reference score range was determined based on the 95% confidence interval of a large sample of health data. Dimensions exceeding the range were marked as anomalous, and the degree of anomalousness was recorded (mild / moderate / severe). Simultaneously, key influencing factors for each dimension were extracted, and the top three factors with the highest absolute values ​​of their influencing indices were selected as the core influencing factors for each dimension. Combining user medical history and behavioral data, the causal relationship between anomalous dimensions and core influencing factors was analyzed. This ultimately generated personalized health status data for each user, encompassing structured information such as scores for each health dimension, anomalous dimension markings, anomalous degree, core influencing factors, influencing mechanism analysis, and health recommendations. This data provides crucial support for subsequent health dimension assessments and radar chart design.

[0096] Furthermore, the analysis and processing of individualized user health status based on the influence feature data of user health indicator nodes includes:

[0097] Based on the impact feature data of user health indicator nodes, perform user health indicator specificity analysis to generate user health indicator specific data;

[0098] Based on the health domain dimension data, analyze the user health indicator specific data to obtain the user health dimension related candidate indicator data;

[0099] Based on the candidate indicator data related to user health dimensions, we perform individualized health status analysis and processing to generate individualized user health status data.

[0100] In this embodiment of the invention, the specificity of user health indicators is analyzed through a population benchmark database and a weighted calculation mechanism based on influencing features. The population benchmark database is constructed based on physical examination data and clinical records of tens of millions of healthy individuals, and is divided into subgroups according to age (every 5 years), gender, and region. It stores the mean, standard deviation, and 95% reference interval of health indicators for each subgroup, and the data is updated quarterly to include new samples. The analysis process takes the user health indicator node influencing feature data as the core input. First, the indicator value is extracted and compared with the corresponding indicator in the same subgroup in the population benchmark data. The Z-score calculation formula is used to quantify the deviation of the value: Z-score = (user indicator value - subgroup mean) / subgroup standard deviation. The direction of deviation (positive deviation / negative deviation) and the degree of deviation (absolute value of Z-score) are recorded. The specificity coefficient is calculated by combining the relative influence weights and the rarity of influencing factors in the influencing feature data. The rarity of influencing factors is determined based on the frequency of occurrence of the factor in the population data: a rarity coefficient of 2.0 is set for factors with a frequency below 5%, 1.5 for 5%-20%, and 1.0 for above 20%. The specificity coefficient = absolute Z-score × relative influence weight × rarity coefficient. Specificity level classification standards are set: a specificity coefficient ≥ 1.8 is extremely high specificity, 1.2 ≤ specificity coefficient < 1.8 is high specificity, 0.6 ≤ specificity coefficient < 1.2 is moderate specificity, and a specificity coefficient < 0.6 is low specificity. The uniqueness of the indicator's influence characteristics is analyzed simultaneously. If an indicator is affected by three or more low-rarity factors, it is additionally marked as "compound influence type specificity," generating specific data for user health indicators. This process analyzes the core node list and relationships within the health domain dimension data. Each health dimension (e.g., cardiovascular health, metabolic health) has a pre-defined set of core indicator nodes. Core nodes are labeled from the top 10% of nodes in the knowledge graph based on their degree centrality, and a threshold for the association strength between nodes (set to 0.4) is stored. The first layer of matching is performed: the indicator nodes in the user's specific health indicator data are precisely compared with the core node list of each dimension. Successfully matched indicators are directly included in the candidate indicator pool of the corresponding dimension and marked as "core candidate indicators," with the dimension identifier, matching type, and core node association strength recorded. For unmatched indicators, a second layer of association filtering is performed. Based on the association path data of the knowledge graph, the shortest path association strength between the indicator node and the core nodes of each dimension is calculated. The path association strength is the product of the association strengths of all edges in the path. Only indicators with a path association strength ≥ 0.4 are retained and included in the candidate indicator pool of the corresponding dimension, marked as "association candidate indicators."The third layer of semantic verification is performed using a medical word vector model (768-dimensional vectors, cosine similarity calculation) pre-trained on the PubMed corpus to verify the semantic consistency between candidate indicators and dimension topics. Indicators with a similarity ≥ 0.8 are retained, while those below the threshold are removed. Feature statistics are performed on the candidate indicator pools for each dimension, recording the number of indicators, the proportion of core / related candidate indicators, the average specificity coefficient, and the highest specificity level, generating candidate indicator data related to the user's health dimension. A multi-feature weighted fusion model is adopted, with the following configuration parameters: the feature input layer includes three types of features: standardized indicator values, specificity coefficients, and influence indices; the weight allocation layer uses the entropy weight method to dynamically calculate feature weights; the output layer is the dimension health score and anomaly label; and the model iteration terminates when the validation set error is below 0.03. The analysis process first performs standardization processing on the candidate indicator values. Positive indicators are standardized by maximization, while negative indicators (such as blood pressure and blood sugar) are standardized by minimization, uniformly mapping the values ​​to the 0-1 range. The weights of each candidate indicator are calculated using the entropy weighting method. First, the information entropy of the indicator is calculated: Entropy = -(1 / lnm) × Σ(pi × lnpi) (where m is the number of samples and pi is the proportion of the standardized indicator value). Then, the indicator weight is determined by weight = (1 - entropy value) / Σ(1 - entropy value). Indicators with a specificity coefficient ≥ 1.2 are multiplied by an enhancement coefficient of 1.5. Combining the influence index from the user health indicator node influence feature data, the weighted score of each indicator is calculated: Indicator weighted score = Standardized value × Indicator weight × Influence index. The comprehensive score for each health dimension is calculated using the weighted summation method: Dimension score = Σ Indicator weighted score × 100, with the score range mapped to 0-100 points. The dimensional scores were compared with the 95% reference range for the same subgroup of healthy individuals. Scores within 10% of the lower limit of the range were considered mildly abnormal, 10%-20% below were considered moderately abnormal, and more than 20% below were considered severely abnormal. Scores more than 10% above the upper limit of the range were marked as abnormally high indicators. The top three candidate indicators with the highest absolute values ​​of their impact indices within each dimension were extracted as core influencing factors. Combined with user medical history and behavioral data, the causal relationship between these factors and abnormal scores was analyzed (e.g., high-salt diet → elevated systolic blood pressure → decreased cardiovascular dimension score). The generated personalized health status data for users included fields such as scores for each dimension, abnormal dimension markings, degree of abnormality, a list of core influencing factors, factor-indicator-dimensional causal chain, and health risk level.

[0101] Furthermore, as an embodiment of the present invention, referring to FIG2, which is a detailed functional flowchart of the health dimension assessment module in FIG1, the health dimension assessment module in this embodiment includes the following functions:

[0102] S31: Based on the health indicator-influencing factor knowledge graph, analyze the health indicator dependency characteristics and health indicator contribution factors, and generate health indicator dependency characteristic data and health indicator contribution factor data respectively.

[0103] In this embodiment of the invention, health indicator dependency features and health indicator contribution factors are analyzed based on a health indicator-influencing factor knowledge graph. The health indicator dependency feature analysis employs a breadth-first search algorithm to scan the entire graph, tracing the direct and indirect association paths between each health indicator node and other indicator nodes. A direct association path is defined as an edge connection between two nodes without intermediate nodes, while an indirect association path is defined as an edge connection containing 1-2 intermediate nodes. The dependency strength of each path is calculated; direct dependency strength is taken as the association strength value of the edge, and indirect dependency strength is taken as the product of the association strengths of all edges in the path. A dependency strength threshold of 0.3 is set, and only paths higher than the threshold are retained. Dependencies are categorized by type, labeled as causal dependency or association dependency based on the relationship attribute of the edge, and labeled as first-level dependency (1 hop) or second-level dependency (2 hops) based on path length. The number of dependent nodes for each indicator is simultaneously counted to form a dependency feature matrix. The final generated health indicator dependency feature data includes fields such as indicator node ID, list of dependent nodes, dependency type, dependency strength, dependency level, and number of dependent nodes. The health indicator contribution factor analysis focuses on the interaction between influencing factor nodes and indicator nodes in a knowledge graph. It extracts the relative influence weight, influence range, and influence timeliness coefficient from causal feature data. The contribution factor value is calculated through a weighted product: Contribution Factor Value = Relative Influence Weight × Influence Range Coefficient (1.0 for ≤3 related indicators, 1.5 for 4-6, and 2.0 for ≥7) × Timeliness Coefficient (1.2 for short-term, 1.0 for medium-term, and 0.8 for long-term). The contribution factor values ​​are divided into 5 levels at 0.2 intervals, marking the contribution level of each influencing factor. This generates health indicator contribution factor data containing indicator node ID, influencing factor node ID, contribution factor value, contribution level, and effect attribute.

[0104] S32: Based on health domain dimension data and health indicator dependency feature data, establish a model architecture for a health dimension logical dependency tree and generate a health dimension logical dependency tree model architecture.

[0105] In this embodiment of the invention, the health dimension logical dependency tree model architecture is established based on the dimensional division of health domain dimension data, combined with the dependency feature data of health indicators to construct a hierarchical structure. First, the health domain dimension data is parsed, with dimensions such as cardiovascular health and metabolic health serving as the root nodes of the dependency tree. Each root node is bound to a corresponding list of core indicator nodes (nodes with the highest degree centrality in the top 10%). Core indicator nodes are used as first-level child nodes of the dependency tree. Based on the direct dependencies in the dependency feature data, indicators with first-level causal dependencies on the core indicators are designated as second-level child nodes, and related dependent indicators are designated as third-level child nodes, ensuring that the hierarchical depth does not exceed three levels. Node sorting follows the "contribution factor descending order" principle, with nodes within the same level arranged from highest to lowest contribution factor value. The edge attributes of the tree are bound to the dependency strength and dependency type in the dependency feature data, and each edge is labeled with a corresponding numerical value and attribute label. The model architecture adopts a hybrid structure of binary and multi-branch trees. A maximum of 8 second-level child nodes are attached to a core indicator node, and a maximum of 4 third-level child nodes are attached to a second-level child node. Nodes exceeding this number are retained based on dependency strength. The architecture uses leaf node identifiers, marking only third-level child nodes and second-level child nodes without dependencies as leaf nodes for subsequent score calculations. The final generated health dimension logical dependency tree model architecture includes core components such as the dimension root node ID, hierarchical structure table, node attribute set, edge attribute set, leaf node list, and architecture constraint rules.

[0106] S33: Design model health assessment weights by using health indicator dependency feature data and health indicator contribution factor data;

[0107] In this embodiment of the invention, the model health assessment weight design employs a combined weighting method to fuse health indicator dependency features and contribution factor data. First, structural weights are calculated. Based on the dependency strength and dependency level in the dependency feature data, an inverse distance weighting method is used, with a first-level dependency weight coefficient of 1.0 and a second-level dependency weight of 0.5. The structural weight = dependency strength × level coefficient. Normalization is performed on all structural weights for the same indicator. Next, contribution weights are calculated. Based on the contribution factor values ​​in the contribution factor data, contribution weights are determined according to the indicator weight information in the relevant candidate indicators of the user health dimension. The structural weights and contribution weights are weighted and fused in a 4:6 ratio. The fused weight = structural weight × 0.4 + contribution weight × 0.6. Normalization is performed again to ensure the sum is 1. A correction coefficient is set separately for leaf nodes. The correction coefficient for leaf nodes corresponding to core indicators is 1.2, and the correction coefficient for leaf nodes corresponding to non-core indicators is 1.0. The final model health assessment weights are obtained after correction. The generated data includes fields such as dimension identifier, node ID, structural weight, contribution weight, fusion weight, correction coefficient, and final weight. The total weight of nodes within each dimension is 1.0.

[0108] S34: Use the model health assessment weights to adjust the model parameter weights of the health dimension logical dependency tree model architecture to generate a health dimension logical dependency tree model.

[0109] In this embodiment of the invention, a top-down weight allocation mechanism is adopted. First, the root node weight is fixed at 1.0. The weights of the first-level child nodes (core indicators) are allocated to the root node according to their final weights. For example, if the weights of three first-level child nodes in a certain dimension are 0.4, 0.3, and 0.3 respectively, then the allocated parameter weights are 0.4, 0.3, and 0.3. The parameter weight of a second-level child node is the product of its parent node's parameter weight and its own final weight. For example, if the parent node's weight is 0.4 and the child node's own weight is 0.25, then the child node's parameter weight is 0.4 × 0.25 = 0.1. The parameter weight of third-level child nodes is calculated using the same logic, i.e., parent node's parameter weight × its own final weight. Weight verification is performed during the adjustment process. After each level of adjustment is completed, the sum of the parameter weights of all nodes at that level is calculated and must equal the parameter weight of the corresponding parent node at the next higher level. If the deviation exceeds 0.01, the weights are redistributed. An iterative adjustment mechanism is set up. For nodes with cross-dependencies (depending on multiple parent nodes), the average weight of each parent node is taken as the final parameter weight. This process is repeated until the weights of all levels are balanced. The model configuration parameters are set as follows: maximum level depth of 3, weight adjustment deviation threshold of 0.01, cross-dependent node weights are calculated using an arithmetic mean, and the maximum number of adjustment iterations is 5. After adjustment, a healthy dimension logical dependency tree model is generated, including the parameter weights of each node, the distribution of weights at each level, and a weight verification report.

[0110] S35: Transmit the user's individual health status data to the health dimension logical dependency tree model for evaluation of each health dimension, and generate user's individual health dimension evaluation data.

[0111] In this embodiment of the invention, the health dimension assessment based on the dependency tree model takes individualized health status data of users as input and performs weighted calculations hierarchically. The standardized values ​​and anomaly markers of each indicator in the input data are parsed, and the standardized values ​​are mapped to the 0-1 range. Positive indicators are standardized using maximization, and negative indicators are standardized using minimization. The standardized value of a leaf node is multiplied by the parameter weight of the corresponding node in the model to obtain the leaf node score. For example, if the standardized value of a leaf node is 0.8 and the parameter weight is 0.1, the score is 0.08. The scores of second-level child nodes are calculated by summing the scores of all leaf nodes under them, the scores of first-level child nodes are summed by summing the scores of all second-level child nodes, and the root node (dimension) score is summed by summing the scores of all first-level child nodes, then multiplied by 100 to convert to a dimension assessment score of 0-100. Anomaly markers are penalized: 5 points for mild anomalies, 10 points for moderate anomalies, and 20 points for severe anomalies. The deductions are directly deducted from the corresponding node score. The dimensional assessment scores are compared with the 95% reference range for the same subgroup of healthy individuals. Scores below the lower limit of the range are considered dimensional abnormalities, and are categorized as mild (5%-10%), moderate (10%-20%), or severe (>20%) abnormalities based on the percentage difference. The influencing factors corresponding to the three lowest-scoring leaf nodes are extracted as key intervention points for each dimension, and the intervention basis is explained in conjunction with medical history data. The final user-individualized health dimension assessment data includes fields such as dimension name, dimension assessment score, abnormality marker, degree of abnormality, key intervention points, and score composition details, providing core dimensional data for creating health radar charts.

[0112] Furthermore, the health indicator dependency feature data includes the dependency features between subsets of health indicator entity data and the dependency features between subsets of health indicator entity data and subsets of health influencing factor entity data.

[0113] Furthermore, the health radar chart design module includes the following functions:

[0114] We analyze the data volatility, skewness, and trend characteristics of each health dimension in the user's individualized health dimension assessment data, and generate user health volatility data, user health skewness data, and user health trend characteristic data respectively.

[0115] In this embodiment of the invention, the health dimension data feature analysis relies on time series statistics and distribution fitting mechanisms, using historical dimension scores (default 12 months of continuous data) from the user's individualized health dimension assessment data as the core input. Data volatility analysis uses the coefficient of variation (CV) as a quantitative indicator. Since different health dimension scores have the same dimension (0-100 points), the CV can directly compare the degree of volatility. The calculation formula is: CV = (monthly score standard deviation / monthly score mean) × 100%. The quartile interval is calculated simultaneously for verification. Volatility levels are divided according to the CV: ≤10% is low volatility, 10%-20% is medium volatility, and >20% is high volatility. This generates user health volatility data containing dimension name, monthly score sequence, mean, standard deviation, CV, and volatility level. Data skewness analysis determines the distribution pattern through the skewness coefficient, setting coefficient thresholds: <-1 indicates strong left skewness, -1 to -0.5 indicates weak left skewness, -0.5 to 0.5 indicates approximately symmetrical skewness, 0.5 to 1 indicates weak right skewness, and >1 indicates strong right skewness. The distribution characteristics are verified by combining the peak position of the histogram, generating skewed user health data containing dimension name, skewness coefficient, distribution type, and peak position. The data trend feature analysis uses a linear regression model to fit the trend line. The model configuration parameters are: independent variable is a time series (1-12 months), dependent variable is monthly score, the least squares method is used to solve the regression equation, the significance level is set to 0.05, and the significance of the slope is judged by t test. Slope > 0 and p < 0.05 indicates an upward trend, slope < 0 and p < 0.05 indicates a downward trend, and p ≥ 0.05 indicates a stationary trend. The trend strength is divided according to the absolute value of the slope: ≥ 2 indicates a strong trend, 1-2 indicates a medium trend, and < 1 indicates a weak trend. User health trend feature data is generated, including dimension name, regression equation, slope value, p value, trend direction, and trend strength.

[0116] By using a preset health radar morphology decision, user health fluctuation data, user health skewed data, and user health trend feature data are processed to transform the radar morphology features of user health, generating user health radar morphology feature data.

[0117] In this embodiment of the invention, the radar morphology feature conversion processing is implemented based on a preset health radar morphology decision rule base. The rule base binds the mapping relationship between volatility, skewness, trend features and radar chart visual attributes. First, the volatility level in the user's health volatility data is analyzed and mapped to the smoothness parameter of the radar chart polygon edges: low volatility corresponds to a Bezier curve smoothing coefficient of 0.1 (approximately a straight line), medium volatility corresponds to a smoothing coefficient of 0.3 (slight curvature), and high volatility corresponds to a smoothing coefficient of 0.6 (obvious jaggedness). Combining the distribution type of the user's health skewness data, the vertex offset feature is converted: strong left skewness shifts the corresponding dimension vertex towards the center of the circle by 10% of the radius length, strong right skewness shifts the vertex towards the circumference by 10% of the radius length, approximately symmetrical with no offset, and weak skewness with the offset amplitude halved, and the offset direction along the extension line of the dimension coordinate axis. Based on the trend direction and intensity of user health trend data, fill color gradients are configured: an upward trend uses a green-light green gradient (0.4 color difference for strong trends, 0.2 for medium trends, and 0.1 for weak trends); a downward trend uses a red-light red gradient (color difference same as the upward trend); and a stable trend uses a blue-light blue gradient. Vertex styles are simultaneously set: strong trends use rhombuses (8px side length), medium trends use squares (6px side length), and weak trends use circles (4px diameter). Rule validation is performed during the conversion process to ensure that the morphological feature combinations of a single dimension do not conflict. For example, when high volatility and strong trends coexist, the jagged edge features corresponding to volatility are retained first. The final generated user health radar morphological feature data includes fields such as dimension name, edge smoothing coefficient, vertex offset, fill color parameters, vertex style parameters, and feature priority.

[0118] Based on user health volatility data, user health skewness data, and user health trend characteristic data, we conduct user health risk correlation factor analysis to generate user health risk correlation factor data.

[0119] In this embodiment of the invention, the following abnormal feature dimensions are selected: high volatility dimensions in volatile data, strong skewness dimensions in skewed data, and strong downward trend dimensions in trend data are included in the abnormal feature dimension set, and the number of abnormal features in each dimension is counted (multiple anomalies have doubled weight). Based on the health indicator-influencing factor knowledge graph, a breadth-first traversal algorithm is used to track the upstream influencing factor nodes of the core indicators of the abnormal dimensions, with a traversal depth limited to 2 layers. The contribution factor values ​​and effect attributes of the health indicator contribution factor data corresponding to the nodes are extracted. The risk correlation is calculated as follows: Risk correlation = (contribution factor value × number of abnormal features) × trend influence coefficient (downward trend coefficient 1.5, stable / upward trend coefficient 1.0). A risk correlation threshold of 0.8 is set, and influencing factor nodes above the threshold are retained. Attribution analysis is performed on the selected factors. Combining the causal relationship attributes of the knowledge graph edges, the mechanism of action between factors and abnormal features is clarified (e.g., high-salt diet → blood pressure fluctuation → high volatility in cardiovascular dimensions). Simultaneously, the number of abnormal dimensions associated with the factors is counted, and core risk factors (correlation dimension ≥ 2) and secondary risk factors (correlation dimension = 1) are marked. The final generated user health risk correlation factor data includes structured fields such as factor node ID, factor name, risk correlation degree, list of correlation anomaly dimensions, description of mechanism of action, factor level, and contribution factor value.

[0120] The user health radar chart is designed and generated by using user health risk correlation factor data and user health radar morphology feature data.

[0121] In this embodiment of the invention, the health radar chart design employs a multi-layer overlay drawing mechanism, integrating visual parameters from user health risk correlation factor data and radar morphological feature data. The basic framework of the radar chart is constructed using user health radar morphological feature data, and an overlay risk factor annotation layer is established using user health risk correlation factor data. After drawing, visual verification is performed to ensure that the scoring points of each dimension are aligned with the coordinate axis scales, the annotation boxes do not overlap, and the fill color gradient transition is natural. The final generated user health radar chart includes basic framework layer data, polygon core layer data, risk annotation layer data, legend layer data, and overall size parameters, and can be directly output as a vector graphics format.

[0122] This specification provides a method for designing a health radar chart based on a fused knowledge graph, characterized in that it is executed based on the health radar chart design system based on a fused knowledge graph as described in claim 1, and the method includes the following steps:

[0123] Acquire medical and health information data from the medical system; build a knowledge graph of health indicators and influencing factors based on the medical and health information data, and generate a health indicator-influencing factor knowledge graph.

[0124] Acquire users' real-time health data; perform personalized health status analysis and processing on users' real-time health data through a health indicator-influencing factor knowledge graph to generate personalized health status data for users;

[0125] A mathematical model of the logical dependency tree of health dimensions is established based on the knowledge graph of health indicators and influencing factors, and a logical dependency tree model of health dimensions is generated; the user's individual health status data is transmitted to the logical dependency tree model of health dimensions for evaluation processing of each health dimension, and user individual health dimension evaluation data is generated.

[0126] A user health radar chart is designed and generated based on user-individualized health dimension assessment data.

[0127] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0128] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A health radar chart design system integrating knowledge graphs, characterized in that, It includes the following modules: a knowledge graph design module, used to acquire medical and health information data from the medical system; A knowledge graph of health indicators and influencing factors is established based on medical and health information data, generating a health indicator-influencing factor knowledge graph. The health status analysis module is used to acquire users' real-time health data; it performs personalized health status analysis on users' real-time health data through a health indicator-influencing factor knowledge graph, and generates personalized health status data for users. The health dimension assessment module is used to establish a mathematical model of the logical dependency tree of health dimensions based on the knowledge graph of health indicators and influencing factors, and to generate a logical dependency tree model of health dimensions. The module transmits the user's individual health status data to the health dimension logical dependency tree model for evaluation of each health dimension, generating user-individual health dimension evaluation data; the health radar chart design module is used to design user health radar charts based on user-individual health dimension evaluation data, generating user health radar charts.

2. The health radar chart design system integrating knowledge graphs according to claim 1, characterized in that, The knowledge graph design module includes the following functions: acquiring medical and health information data from the medical system; Perform structured preprocessing on medical and health information data to generate structured health information data; Entity extraction is performed on the structured health information data to obtain structured health entity data. Structured health entity data is divided into health indicator entities and health influencing factor entities, resulting in health indicator entity data and health influencing factor entity data, respectively. Based on these entities, causal feature analysis of health indicators and influencing factors is performed to generate causal feature data of health indicators and influencing factors. The health indicator entity data is then divided into health indicator types to generate health indicator type data. A preliminary knowledge graph of health indicator types is constructed using this type data, generating a preliminary health indicator-influencing factor knowledge graph. Finally, a graph analysis of the co-occurrence coupling mapping of influencing factors is performed based on this preliminary knowledge graph to generate a comprehensive health indicator-influencing factor knowledge graph.

3. The health radar chart design system integrating knowledge graphs according to claim 2, characterized in that, Causal characteristic analysis of health indicators and influencing factors based on entity data of health indicators and entity data of health influencing factors includes: conducting correlation analysis of health indicators and health influencing factors based on entity data of health indicators and entity data of health influencing factors to generate correlation data of health indicators and influencing factors, and conducting causal characteristic analysis of correlation data of health indicators and influencing factors to generate causal characteristic data of health indicators and influencing factors.

4. The health radar chart design system integrating knowledge graphs according to claim 1, characterized in that, The real-time health data of users described in the health status analysis module includes real-time user vital sign monitoring data, real-time user behavior record data, and user health history data.

5. The health radar chart design system integrating knowledge graphs according to claim 4, characterized in that, The health status analysis module includes the following functions: acquiring real-time user health data; transmitting real-time user health data to the health indicator-influencing factor knowledge graph for user health node mapping processing, and generating user health graph mapping data; Perform health indicator association path analysis on user health graph mapping data to obtain user health indicator association path depth data; Based on the deep data of the association path of user health indicators, the influence characteristics of indicator nodes in the knowledge graph are analyzed to generate influence characteristic data of user health indicator nodes. Health domain dimensional analysis is performed using a health indicator-influencing factor knowledge graph to generate health domain dimensional data; based on the health domain dimensional data and user health indicator node influence feature data, user individual health status analysis is performed to generate user individual health status data.

6. The health radar chart design system integrating knowledge graphs according to claim 5, characterized in that, The user-individualized health status analysis based on the influence feature data of user health indicator nodes includes: performing user health indicator specificity analysis based on the influence feature data of user health indicator nodes to generate user health indicator specific data; performing user health dimension-related candidate indicator analysis on the user health indicator specific data based on health domain dimension data to obtain user health dimension-related candidate indicator data; and performing user-individualized health status analysis based on the user health dimension-related candidate indicator data to generate user-individualized health status data.

7. The health radar chart design system integrating knowledge graphs according to claim 5, characterized in that, The health dimension assessment module includes the following functions: analyzing the dependence characteristics and contribution factors of health indicators based on the health indicator-influencing factor knowledge graph, generating health indicator dependence characteristic data and health indicator contribution factor data respectively; establishing a model architecture of the health dimension logical dependency tree based on the health domain dimension data and health indicator dependence characteristic data, generating the health dimension logical dependency tree model architecture. The model health assessment weights are designed using health indicator dependency feature data and health indicator contribution factor data to generate model health assessment weights; the model health assessment weights are then used to adjust the model parameter weights of the health dimension logical dependency tree model architecture to generate a health dimension logical dependency tree model. The user's individual health status data is transmitted to the health dimension logical dependency tree model for evaluation of each health dimension, generating user-individual health dimension evaluation data.

8. The health radar chart design system integrating knowledge graphs according to claim 7, characterized in that, The health indicator dependency feature data includes the dependency features between subsets of health indicator entity data and the dependency features between subsets of health indicator entity data and subsets of health influencing factor entity data.

9. The health radar chart design system integrating knowledge graphs according to claim 1, characterized in that, The health radar chart design module includes the following functions: analyzing the data volatility, data skewness, and data trend characteristics of each health dimension of the user's individualized health dimension assessment data, and generating user health volatility data, user health skewness data, and user health trend characteristic data respectively; By using a preset health radar morphology decision, user health fluctuation data, user health skewed data, and user health trend feature data are processed to transform the radar morphology features of user health, generating user health radar morphology feature data. Based on user health volatility data, user health skewness data, and user health trend characteristic data, we conduct user health risk correlation factor analysis to generate user health risk correlation factor data. The user health radar chart is designed and generated by using user health risk correlation factor data and user health radar morphology feature data.

10. A method for designing a health radar chart integrating knowledge graphs, characterized in that, Based on the health radar chart design system using the fusion knowledge graph as described in claim 1, the method for designing a health radar chart using the fusion knowledge graph includes the following steps: acquiring medical and health information data from a medical system; establishing a knowledge graph of health indicators and influencing factors based on the medical and health information data to generate a health indicator-influencing factor knowledge graph; acquiring real-time user health data; performing personalized health status analysis on the real-time user health data using the health indicator-influencing factor knowledge graph to generate personalized user health status data; establishing a mathematical model of a logical dependency tree for health dimensions based on the health indicator-influencing factor knowledge graph to generate a logical dependency tree model for health dimensions; transmitting personalized user health status data to the logical dependency tree model for evaluation of each health dimension to generate personalized user health dimension evaluation data; and designing a user health radar chart based on the personalized user health dimension evaluation data to generate the user health radar chart.

Images