Artificial intelligence-based health status intelligent analysis system and method

Abstract

The application discloses a health state intelligent analysis system and method based on artificial intelligence, relates to the technical field of health data analysis, and comprises the following steps: acquiring multi-dimensional data of a user and user-defined meaning anchor point data; acquiring marking data in which the user associates behaviors or feelings with the meaning anchor points; based on all the data, constructing and dynamically maintaining a personalized meaning network graph model, the weight of an edge being combined with data-driven statistical correlation analysis and user-marking-driven semantic connection learning; performing network path analysis, identifying the current state of the user and the effective correlation path between the behaviors and the meaning anchor points; generating personalized and meaning-oriented feedback information based on the identified path, revealing the connection between experience and personal meaning; and presenting the feedback and performing closed-loop optimization through user evaluation. The application provides feedback directly targeting the personal meaning of the user by constructing a personalized meaning network, thereby being expected to significantly improve the intrinsic motivation, long-term compliance of behavior change and overall happiness of the user.

Description

Technical Field

[0001] This invention relates to the field of health data analysis, and in particular to an intelligent health status analysis system and method based on artificial intelligence. Background Technology

[0002] Health self-management is crucial for chronic disease prevention and quality of life improvement. Currently, there are numerous health applications that utilize mobile devices and wearable sensors to collect users' physiological, behavioral, and environmental data. These applications typically provide functions such as data visualization, goal setting, general health knowledge push, and behavioral reminders based on simple rules or basic data patterns. For example, pedometer applications encourage users to reach daily step goals, sleep trackers display the distribution of sleep stages, and some applications even use basic machine learning models for state prediction or risk assessment.

[0003] However, existing technologies generally face the challenge of effectively stimulating users' intrinsic motivation and maintaining long-term behavioral adherence. Users often lose interest after the initial novelty wears off and find it difficult to integrate healthy behaviors into their daily lives and maintain them long-term. This is largely because: 1) the collected data is often disconnected from the user's personal life meaning and deep goals, making it difficult for users to perceive the deeper meaning behind their daily behavioral choices; 2) the feedback provided is usually general, data-based, or directive, lacking connection to individual uniqueness and intrinsic value pursuit, making it difficult to touch the user's intrinsic motivation; 3) although behavior change theories emphasize the importance of autonomy, competence, and consistency with values ​​for intrinsic motivation. Summary of the Invention

[0004] This invention provides an intelligent health status analysis system and method based on artificial intelligence to solve the problem in existing technologies that users often lose interest after the initial novelty wears off, making it difficult to integrate healthy behaviors into daily life and maintain them in the long term.

[0005] In a first aspect, the present invention provides an intelligent health status analysis method based on artificial intelligence, comprising:

[0006] A: Acquire time-stamped physiological, behavioral, and environmental data of the target user, and receive user-inputted anchor data representing their health-related goals through user interaction;

[0007] B: Through the user interaction, obtain the semantic association marker data of the user input. The marker data explicitly records the semantic association between the specific behavioral event or subjective feeling state judged by the user and the semantic anchor data.

[0008] C: Based on all the acquired data, construct and periodically or event-driven dynamically maintain a personalized meaning network model for the target user;

[0009] D: When the preset trigger logic is met, perform network path analysis on the personalized meaning network model to identify a valid association path that meets the preset conditions, starting from the source node corresponding to the current user state or behavior and reaching the target node corresponding to the meaning anchor data.

[0010] E: Based on the attributes of the identified valid association paths, call the feedback strategy library and match the feedback generation rules to generate personalized and meaning-oriented feedback information. The feedback information aims to reveal to the user the specific connection learned by the model between their current experience and their personal meaning anchor.

[0011] F: Through the user interaction, the generated feedback information is presented to the target user in a preset format.

[0012] Preferably, while or after presenting feedback information, an interactive control is provided to the user to evaluate the effectiveness or relevance of the feedback; the collected evaluation data is processed and analyzed through a configured feedback learning module; the analysis results are mapped to the weight adjustment signals of relevant edges in the personalized meaning network model, or used to update the priority or triggering conditions of strategies in the feedback strategy library, so as to achieve adaptive optimization of the system.

[0013] Secondly, the present invention also provides an intelligent health status analysis system based on artificial intelligence, comprising:

[0014] A multi-dimensional data acquisition interface module for interacting with external sensors and user input devices;

[0015] A personalized meaning network modeling engine, which includes a data-driven association learner and a semantic connection learner, and is configured with a dynamic maintenance mechanism;

[0016] The meaning-oriented feedback engine includes a trigger logic judgment unit, a network path analyzer, and a feedback generator.

[0017] The user interaction module is used to present information, obtain user tags, and provide feedback.

[0018] The feedback learning and model optimization module is used to achieve closed-loop adaptive adjustment.

[0019] Preferably, a model interpretability submodule is constructed. When requested by the user or determined by the system, the model interpretability submodule is configured to perform at least one of the following operations: based on the meaning network model and related raw data fragments, generate an evidence summary about a specific association edge, the summary including the number of data points supporting the association, the association strength index, and applicable user-labeled records; for a certain generated meaning-oriented feedback, visualize the key network path on which it is based or explain its generation logic.

[0020] The technical solution provided in this application has at least the following technical effects or advantages:

[0021] The personalized meaning network model constructed in this invention can directly reveal the personalized connection learned by the model between the user's current experience and the unique meaning anchor points set by the user. This feedback goes beyond simple data presentation or behavioral suggestions, giving daily experiences a deeper personal meaning.

[0022] Enhance users' awareness and understanding of the significance of healthy behaviors: By continuously receiving this meaning-oriented feedback, users can more clearly recognize how their own behavioral choices, even small daily behaviors, are related to their long-term goals, thereby deepening their understanding and recognition of the intrinsic value of a healthy lifestyle.

[0023] Enhancing intrinsic motivation and long-term behavioral adherence: This invention constructs and utilizes a personalized meaning network to provide meaning-oriented feedback that directly impacts the deep-seated driving forces behind behavioral change—the connection between behavior and personal meaning—rather than merely focusing on behavioral appearances or external incentives. This is fundamentally different from existing technologies such as simple data association, goal tracking, general suggestions, and even basic personalized recommendations.

[0024] Because this invention helps users integrate their daily behaviors with their personal meaning framework, promoting a sense of purpose and value consistency in life, it may indirectly but significantly improve deeper mental health indicators such as psychological resilience, subjective well-being, and life satisfaction, in addition to behavioral changes. Attached Figure Description

[0025] Figure 1 This is a flowchart of the intelligent health status analysis method based on artificial intelligence of the present invention;

[0026] Figure 2 This is an architecture diagram of an AI-based intelligent health status analysis system. Detailed Implementation

[0027] This invention relates to an intelligent health status analysis system and method based on artificial intelligence. It aims to overcome the limitations of traditional health advice by constructing a personalized model that reflects the unique connections of users and providing feedback that is connected to users' deep goals and values, thereby stimulating users' intrinsic motivation and improving the effectiveness of long-term self-management.

[0028] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0029] like Figure 1 The flowchart of the AI-based intelligent health status analysis method is shown below. The method includes the following steps:

[0030] A: Acquire time-stamped physiological, behavioral, and environmental data of the target user, and receive user-inputted anchor data representing their health-related goals through user interaction;

[0031] B: Through the user interaction, obtain the semantic association marker data of the user input. The marker data explicitly records the semantic association between the specific behavioral event or subjective feeling state judged by the user and the semantic anchor data.

[0032] C: Based on all the acquired data, construct and periodically or event-driven dynamically maintain a personalized meaning network model for the target user;

[0033] All data is obtained through steps A and B;

[0034] D: When the preset trigger logic is met, perform network path analysis on the personalized meaning network model to identify a valid association path that meets the preset conditions, starting from the source node corresponding to the current user state or behavior and reaching the target node corresponding to the meaning anchor data.

[0035] E: Based on the attributes of the identified valid association paths, call the feedback strategy library and match the feedback generation rules to generate personalized and meaning-oriented feedback information. The feedback information aims to reveal to the user the specific connection learned by the model between their current experience and their personal meaning anchor.

[0036] F: Through the user interaction, the generated feedback information is presented to the target user in a preset format.

[0037] like Figure 2 As shown in the architecture diagram of the AI-based intelligent health status analysis system, the system provided in this embodiment of the invention includes the following modules:

[0038] The system comprises a multi-dimensional data acquisition interface module, a personalized meaning network modeling engine, a meaning-oriented feedback engine, a user interaction module, and optional feedback learning and model optimization modules and a model interpretability subsystem module. These modules can be implemented by one or more processors executing instructions stored in non-transitory memory. Data flows into the acquisition module from external sources (such as sensors and user input), is processed by the modeling engine to construct a personalized meaning network model, and the feedback engine generates meaning-oriented feedback based on this model. This feedback is then presented to the user through the user interaction module, which collects feedback. The feedback learning module uses the feedback to optimize the model or strategy, forming a closed loop.

[0039] The multi-dimensional data acquisition interface module is responsible for comprehensively and accurately acquiring data related to the user's health status and personal significance.

[0040] Objective data acquisition: Configure standardized interfaces for accessing various data sources, including smart wearable devices such as smartwatches and wristbands; mobile terminal built-in sensors such as accelerometers and GPS modules; and external application interfaces such as calendar application interfaces and weather application interfaces.

[0041] Based on the characteristics of data types, implement a hierarchical sampling strategy:

[0042] Acquire high temporal resolution physiological data, such as beat-by-beat or second-level heart rate data for heart rate variability (HRV) analysis, and second-level electrodermal activity (EDA).

[0043] Collect mid-frequency data at the minute or hour level, such as step count, activity intensity classification statistics, duration of each stage in sleep structure reports, and trends in skin temperature or blood oxygen saturation.

[0044] Collect daily, hourly, or event-triggered environmental and behavioral data, such as screen time statistics by application category, location-based scene tags such as home, office, and commuting, ambient light and noise levels, and specific keywords in the calendar.

[0045] All collected data is accompanied by precise timestamps and is aggregated to the processing center through reliable synchronization mechanisms, such as device-side software development kits and cloud application programming interfaces, with local data buffering capabilities to cope with network fluctuations.

[0046] Meaning anchor data acquisition: A guided interface is provided through the user interaction module, allowing users to set their core health-related goals such as "improving energy" and "improving emotional stability".

[0047] The system assigns a unique internal target identifier to each meaning anchor point defined or selected by the user, and allows the user to perform subsequent add, delete, and modify operations.

[0048] Meaning-related tag data acquisition: Provides a convenient user interaction method for receiving meaning-related tags input by users; the tagging process can be initiated by the user after experiencing a specific behavior or feeling, or triggered by the system in a non-mandatory prompt manner when recognizing a specific situation, such as completing an exercise record.

[0049] The tagging process guides users to select or confirm the current behavioral / feeling event and associate it with one or more pre-defined meaning anchors.

[0050] The system records semantic connection data that includes timestamps, behavior node identifiers or perception node identifiers, and a list of associated target identifiers.

[0051] As a preferred option, natural language input processing: a natural language processing unit can be configured, provided that the user has given explicit authorization and has been fully informed of the privacy risks.

[0052] This unit uses techniques such as keyword extraction, sentiment analysis, and relation extraction to process user-input text, such as health diaries, identify potential behaviors, feelings, goals, and their associations, and submits the identification results to the user for confirmation, which then serves as auxiliary input for meaning network modeling.

[0053] II. The personalized meaning network modeling engine is responsible for building and dynamically maintaining a personalized meaning network model for each user. This model is implemented using a graph data structure that includes a set of nodes and a set of weighted edges.

[0054] Node Management: Maintains the user's node collection, including the following node types:

[0055] Objective data feature nodes: represent quantified physiological, behavioral, and environmental characteristics and their state values;

[0056] Subjective feeling node: Represents the user's marked subjective state and its rating or category;

[0057] Behavior nodes: Represent the specific health-related behaviors performed by the user and their occurrence attributes;

[0058] Target node: Represents the health goals set by the user;

[0059] The node set and its attributes are dynamically updated based on new user input or data analysis results.

[0060] Data-driven association edge learner: Input user's long-term, preprocessed, multidimensional feature time series data.

[0061] Core processing flow (executed periodically):

[0062] Step 1: Candidate correlation pair screening: Calculate the long-term preliminary correlation measure between all predefined objective data feature nodes, subjective feeling nodes, and behavioral nodes.

[0063] Set a preliminary threshold for relevance assessment.

[0064] Node pairs whose correlation metrics exceed the threshold are selected as candidate association pairs for further analysis.

[0065] Step 2: Temporal causality or association depth verification (for each candidate association pair, denoted as node A and node B):

[0066] Extract the time-aligned historical sequence data corresponding to node A and node B, and denote them as time series A and time series B.

[0067] Perform temporal causal inference analysis: initialize causal relationship assumptions.

[0068] Apply the selected causal test method based on predictive power: evaluate whether using past values ​​of time series A to predict future values ​​of time series B significantly reduces the prediction error compared to using only past values ​​of time series B; and vice versa.

[0069] Judgment logic: Based on the significance of the reduction in prediction error, determine whether there is a one-way predictive association from A to B or from B to A, or a two-way association / common driving factor, or no direct predictive association. Record the inferred directionality.

[0070] Perform statistical correlation / dependency strength calculation: calculate the numerical value of the correlation strength between time series A and time series B under synchronous or different time lags. The calculation can be performed by methods such as cross-correlation function peak calculation or mutual information calculation.

[0071] Step 3: Comprehensive determination of edge attributes:

[0072] Fusion Logic: Design a set of decision rules to determine whether to add or update an edge connecting node A and node B in the graph, based on the directional evidence and strength values ​​obtained in step two.

[0073] Example of a rule:

[0074] Strong association rule: If the causal inference shows a clear direction and the association strength value exceeds the "high confidence threshold", then create or update a strong data-driven edge with a weight that is positively correlated with the strength value.

[0075] Medium association rule: If there is only evidence of association strength and the strength exceeds the "medium confidence threshold", then create or update an undirected data-driven edge with weight related to strength.

[0076] Weak or unrelated rule: If the evidence is insufficient or the strength is below the threshold, the weight of the existing edge is not created or is reduced.

[0077] Weight sign determination: The weight sign (positive / negative) of the edges is determined based on the sign of the correlation coefficient of the original sequence or pre-defined domain knowledge.

[0078] User-labeled semantic connection edge learner:

[0079] Input: A stream of meaning-related tags created by the user, including timestamps, source node identifiers (behavior / feeling nodes), a list of target node identifiers (target nodes), and optional contextual information.

[0080] Core processing flow (event-driven):

[0081] Step 1: Identify the association pair: For each tagged event, determine the source node and target node pair.

[0082] Step 2: Initialize or find edges: Locate the semantic edge connecting this pair of nodes in the graph. If none exists, create it and assign it a basic weight value.

[0083] Step 3: Calculate the weight increment: Set a default weight increment value.

[0084] The weight increment is adjusted based on the context adjustment factor calculation logic: the logic considers the contextual information of the labeled event, such as the emotional intensity of the user's annotation, the importance level of the labeled context, or the time interval since the last labeling of the same connection, and calculates a context adjustment factor.

[0085] The final weight increment is equal to the default increment multiplied by the context adjustment factor.

[0086] Step 4: Update weights: Add the calculated weight increments to the current weights of the edges, and optionally apply a maximum weight limit.

[0087] Step 5: Handling negative user feedback: If the system receives a clear negative evaluation from a user regarding the feedback generated based on this edge, a punitive weight decay factor (less than 1) is applied to reduce the weight of that edge, and a minimum weight limit can be set.

[0088] Network Integration and Dynamic Maintenance Unit:

[0089] Integration: Data-driven associative edges and semantically connected edges are stored in a unified graph data structure, and their origins are distinguished by the attributes of the edges.

[0090] Dynamic maintenance logic (including at least one of the following):

[0091] Weight decay: A time-dependent decay function is periodically applied to data-driven edges to reduce the weight of edges that have not received new data support for a long time. Semantic connections are primarily updated through user tags and feedback.

[0092] Node attribute update: Update the attribute values ​​of the corresponding nodes in the graph in real time or near real time with the latest data.

[0093] Topology adjustment: In response to user editing operations on meaning anchor points, nodes and related edges are added, deleted, or modified synchronously in the graph.

[0094] Third, the meaning-oriented feedback and reflection engine is responsible for querying the meaning network model at the appropriate time, generating personalized feedback, and stimulating the user's intrinsic motivation.

[0095] Context and trigger monitoring: Use state machines or rule engines to manage trigger logic.

[0096] Continuously monitor predefined triggering conditions, which include at least: the user completing a key health behavior event, the user submitting a subjective feeling marker, a significant change in the state related to the user's goal, reaching a preset periodic reflection time point, or receiving an explicit query request from the user.

[0097] Personalized Meaning Network Query and Analysis: Input the contextual information of the triggering event, as well as the user's personalized meaning network model.

[0098] Core processing flow:

[0099] Step 1: Determine query parameters: Based on the trigger type, determine the starting node set and target node set for graph traversal, which are usually the user's target nodes.

[0100] Step 2: Perform a constrained graph traversal: Initialize the queue to be visited, and maintain the set of visited nodes and the current path information.

[0101] Iteratively expand the path, performing pruning and constraint checks during the expansion process: check if the current path length exceeds the maximum depth limit; check if the current path cumulative weight meets the minimum path weight threshold; perform loop checks to avoid infinite loops.

[0102] If the end node of the path belongs to the target node set, it is recorded as a valid path.

[0103] Continue iterating until the termination condition is met.

[0104] Step 3: Path Evaluation and Filtering: For all valid paths found, a pre-defined path evaluation function is applied to calculate the overall score for each path. The path evaluation function aims to quantify the overall strength, reliability, and relevance to the current context of the association represented by a path, and its calculation principle incorporates at least the following information:

[0105] a) Cumulative edge weights of the path: Calculated cumulatively based on the weights of all edges on the path. For weights representing the strength of association (e.g., ranging from 0 to 1), a weighted product method can be used; a weighted sum method can be used; or the minimum edge weight on the path can be used. The specific calculation method is preset.

[0106] Weighting: When performing cumulative calculations, it is necessary to consider the edge type, whether it is data-driven or semantic connection, and its direction (positive or negative). For example, the strength of positive and negative paths can be calculated separately, or negative weights can be converted into penalty terms.

[0107] b) Path length factor: Shorter, more direct paths between nodes represent stronger connections or influences.

[0108] Calculation method: Introduce a factor that is inversely proportional to the path length.

[0109] c) Node recent activity or importance factor: If the nodes involved in the path are recently active, for example, if the relevant objective data has changed significantly recently, or if the user has recently marked the relevant behavior / feeling, or if the node itself has been given high importance, then the path may be more relevant.

[0110] Calculation method: Assign an activity / importance score to each node on the path. The score can be calculated based on the node's latest update time, change range, user settings, etc., and these scores are integrated into the total score of the path by means such as summation, averaging, or weighted averaging.

[0111] As a preferred option, d) Confidence or strength of evidence factor for edges: If the meaningful network model can provide a confidence score for the learning outcome of each edge, for example, based on the amount of data supporting the association, the statistical significance p-value, or user feedback evaluation, then the path with higher confidence is more reliable.

[0112] Calculation method: The confidence scores of each side on the path are combined into the total score by means such as multiplication, taking the minimum value, or weighted average.

[0113] Overall Score Calculation: The final overall path score is a quantified result obtained by weighting or multiplying the above factors using a preset method: Path Score = f(Cumulative Edge Weight, Path Length Factor, Node Activity Factor, [Edge Confidence Factor]). The specific form of the function f is preset by the system or configurable.

[0114] Feedback content generation and personalization: the critical path list and its attributes selected in the previous step, and a structured feedback strategy library.

[0115] Core processing flow:

[0116] Step 1: Matching Feedback Strategy / Template:

[0117] Analyze the characteristics of the critical path, such as the starting point, ending point, main intermediate node types, and overall correlation trends.

[0118] These path features are matched with the rule conditions in the feedback strategy library to find the most suitable feedback template identifier.

[0119] Step Two: Content Filling and Structuring:

[0120] The feedback strategy library stores a natural language expression framework for implementing various preset meaning-oriented functions, including behavioral meaning confirmation and reinforcement, attribution and understanding of the source of feelings, feedback related to goal progress, and switching of meaning perspectives for challenge response.

[0121] Read the specific names or labels of the nodes involved in the path from the diagram, accurately fill them into the reserved positions of the selected template, and form structured feedback content.

[0122] As a preferred option, step three involves applying template-based or more complex natural language generation techniques to convert structured content into fluent and natural text.

[0123] Based on learned user communication preferences, such as historical feedback or user settings, the tone, style, and level of detail of the text are personalized.

[0124] User feedback processing and learning: Input user evaluation data on the effectiveness or relevance of feedback information provided through interactive controls.

[0125] Core processing flow:

[0126] Step 1: Feedback Attribution: Associate user reviews with the specific network path and edges on which the feedback was generated.

[0127] Step 2: Calculate the adjustment signal: Weight adjustment logic: If a user repeatedly rates a feedback based on a certain edge as "irrelevant," then a negative weight adjustment is calculated and applied to the weight of that edge. If the rating is "very useful," then a positive adjustment is calculated. The magnitude of the adjustment is related to the strength or consistency of the rating.

[0128] Strategy optimization logic: Calculate the average evaluation score obtained from different feedback templates or strategies, periodically update the selection priority of each strategy in the strategy library, and give priority to strategies with better historical evaluations.

[0129] Step 3: Application Adjustment: Update the edge weights in the semantic network model, or adjust the parameters in the feedback policy library.

[0130] IV. The User Interaction and Visualization Module is responsible for all user interactions with the system: it is used for user authorization of data collection, setting of meaning anchors, marking of meaning associations, and receiving feedback information. It presents meaning-oriented feedback in a non-intrusive manner, provides interactive controls for user feedback evaluation, and offers user-configurable privacy settings.

[0131] As a preferred option, the fifth module, Model Interpretability Subsystem, aims to enhance system transparency and user trust. It is configured to respond to user queries or specific system triggers.

[0132] The functionality should include at least one of the following:

[0133] Generate an evidence summary for a specific association edge in a meaning network model, displaying quantitative indicators supporting the association and related user-labeled records. For a specific feedback generation process, visualize the key network paths it follows, or provide a simplified explanation of its generation logic.

[0134] In one specific embodiment, coping with work stress and maintaining exercise habits:

[0135] User A: A software engineer who has been under a lot of work pressure lately and often needs to work overtime.

[0136] Set health goals (meaning anchors):

[0137] Goal_ReduceStress (reduce work stress);

[0138] Goal_MaintainExercise (Maintain the habit of running at least 3 times a week);

[0139] Goal_BetterSleep (improves sleep quality);

[0140] Authorized data sources: smartwatches (heart rate, HRV, sleep, activity), mobile phones (screen time, location, calendar), in-app subjective feeling tagging interface, meaning association tagging interface.

[0141] 2. Initial Meaning Network Construction (assuming it has been running for some time):

[0142] Data-driven associations, some examples:

[0143] Long working hours (calendar / screen time) --- [negative, strong] --> sleep duration;

[0144] Sleep duration --- [positive, moderate] --> HRV (RMSSD) the next day;

[0145] The following day's HRV (RMSSD) --- [Positive, Moderate] --> Subjective Energy Score;

[0146] Running behavior --- [positive, strong] --> short-term increase in HRV -- [subsequent] --> long-term trend of decreasing resting heart rate;

[0147] Running behavior --- [positive, moderate] --> subjective stress relief feeling (occasional marker);

[0148] High-intensity workdays (calendar / screen time) --- [negative, moderate] --> probability of running behavior occurring;

[0149] Semantic connections (partial examples, from user-marked tags):

[0150] Running behavior --- [positive, strong] --> Goal_MaintainExercise;

[0151] Running behavior --- [positive, weak] --> Goal_ReduceStress (Users occasionally mark that they feel less stress after running);

[0152] Sufficient sleep duration --- [positive, strong] --> Goal_BetterSleep;

[0153] 3. Specific scenario triggers and system response (Wednesday evening):

[0154] Scenario: User A had a stressful day and didn't get home until 9 p.m. He was exhausted and missed his planned run.

[0155] Data collection:

[0156] The calendar shows a packed schedule of meetings for the day.

[0157] Screen time displays excessively long usage times for work-related applications.

[0158] The smartwatch recorded a low HRV during the day, indicating that the activity level did not meet the target.

[0159] Users tagged their subjective feelings in the app: "Low energy (2 / 5)" and "High stress (4 / 5)".

[0160] Triggering event: User submits a subjective feeling tag.

[0161] Meaningful network query and analysis (performed by the feedback engine):

[0162] Source nodes: subjective feelings (low energy), subjective feelings (high stress), behavioral nodes (missed exercise), objective data (low HRV), objective data (high workload).

[0163] Target node: The target node for all users.

[0164] Path analysis revealed:

[0165] Path 1: Perceived Attribution: High workload -> Low HRV -> Low energy (strong data-driven correlation)

[0166] Path 2 behavior conflicts with the goal: High workload -> Missed motion -> Negative connection to Goal_MaintainExercise strong semantic connection

[0167] Path 3: Potential positive connections: Searching for behavioral nodes that are positively connected to Goal_ReduceStress reveals a strong historical association between running behavior and sufficient sleep.

[0168] Feedback generation, select a strategy and populate content:

[0169] Matching Strategy: The system identifies a user's state of "goal frustration" and "negative feelings," with a clear objective data correlation: workload -> physiological indicators -> feelings, and a conflict between behavior and goals. A feedback strategy emphasizing "understanding attribution + meaning reconstruction / perspective shift" is selected.

[0170] Generate feedback text:

[0171] Attribution and Empathy: "We understand that you feel low on energy and under a lot of stress today. Data shows that today's high workload may have affected your physiological state, with a low HRV, which is often associated with feelings of low energy."

[0172] Connecting behavior with goals: "Notice that you missed your planned run today, which deviates from your goal of maintaining an exercise habit (Goal_MaintainExercise)."

[0173] Reconstructing Meaning and a Positive Perspective: Prioritizing rest tonight might be a good way to restore balance and support your long-term goals.

[0174] If the model learns that users have improved their sleep in the past by using specific relaxation techniques after similar overtime work, the data might show: "Data shows that you have experienced improved sleep quality when you tried [specific relaxation techniques, such as reading before bed] in similar situations in the past. Consider trying it tonight?"

[0175] User interaction: The user receives this push notification or card in the app.

[0176] Feedback and rating: Users can rate this feedback as "helpful", "average", or "irrelevant".

[0177] Subsequent tagging: If users adopt the suggestions, such as ensuring adequate sleep, they are tagged with positive feelings the next day and may be associated with "rest" or "wellness," further strengthening the relevant network connection.

[0178] 4. Systematic learning and iteration:

[0179] User feedback: If a user rates the feedback as "helpful," the system will slightly increase the confidence level of the network path on which the feedback was based or its future priority. If the user rates it as "irrelevant," the system will decrease the confidence level of the relevant path.

[0180] Model update: The system periodically retrains or fine-tunes the semantic network model using new objective and labeled data, so that the associations more accurately reflect the user's latest state and cognition.

[0181] The method and system described in this invention can be implemented using a computing device equipped with a processor and a memory. The memory stores computer-executable instructions, and the processor executes these instructions to implement the functions of the aforementioned modules. The computing device can be a user's mobile terminal, an edge computing device, or a cloud server, or a combination thereof. The various modules of the system can communicate and exchange data via a network.

[0182] This detailed embodiment elaborates on the internal operating logic of each major functional module of the system, aiming to provide a detailed basis and explanation for those skilled in the art to understand and implement it. It should be emphasized that the above description constitutes a specific, preferred embodiment, but the concept of the present invention is not limited thereto. Any equivalent transformations, modifications, or improvements based on the core spirit of the present invention, without departing from the technical principles and scope disclosed in this specification, should be considered to fall within the scope of protection claimed by the present invention, as long as they achieve the same or similar technical effects.

Claims

1. An artificial intelligence-based health state intelligent analysis method, characterized by, The method comprises: A: obtaining time-stamped physiological data, behavior data and environmental data of a target user, and receiving user input meaning anchor data representing health-related goals through user interaction; B: obtaining user input meaning association label data through the user interaction, the label data explicitly recording the semantic association between a specific behavior event or subjective feeling state determined by the user and the meaning anchor data; C: based on all the data obtained, constructing and periodically or event-driven dynamically maintaining a personalized meaning network model for the target user; the dynamic maintenance includes performing at least one of the following processes: periodic edge weight decay process, reducing the weight of the associated edge that has not obtained new data-driven evidence or user label support within a preset time window according to a preset decay function; node state update process, updating the attribute value of the objective data node, subjective feeling node or behavior node according to the latest obtained data; dynamic adjustment of network topology in response to user editing operation on the target node; D: when the preset trigger logic is met, performing network path analysis on the personalized meaning network model to identify an effective association path that meets the preset conditions from the source node corresponding to the current user state or behavior to the target node corresponding to the meaning anchor data; the preset trigger logic at least includes: user completes a key health behavior event, user submits a subjective feeling label, a state related to the user's goal changes significantly, or receives an explicit query request from the user; the execution of network path analysis includes: using a graph traversal algorithm configured with a maximum search depth limit or a minimum path weight threshold to search all effective paths from the source node to the target node in the meaning network model; E: based on the attributes of the effective association path, calling a feedback strategy library and matching a feedback generation rule to generate personalized and meaning-oriented feedback information, which aims to reveal to the user the specific connection between his current experience and personal meaning anchor learned by the model; the feedback strategy library pre-stores a natural language expression framework for realizing various meaning-oriented functions including behavior meaning confirmation and reinforcement, feeling source attribution and understanding, goal progress association feedback, and challenge coping perspective switching; F: presenting the generated feedback information to the target user in a preset format through the user interaction. 2.The artificial intelligence-based health state intelligent analysis method of claim 1, wherein The personalized meaning network model is stored and represented using a graph data structure, wherein the nodes of the graph include objective data feature nodes, subjective feeling nodes, behavior nodes and target nodes generated according to the data classification, and the edges of the graph represent the learned weighted associations between the nodes; the weights and types of the edges are determined by combining the following steps: C1: performing data-driven association analysis on the time series corresponding to the objective data, subjective feeling and behavior data to determine the statistical association strength and direction between the nodes; C2: based on the meaning association label data, establishing or strengthening the semantic connection weight between the behavior node or feeling node and the target node. 3.The artificial intelligence-based health state intelligent analysis method of claim 2, wherein The data-driven association analysis comprises: preprocessing the physiological data, behavior data and environmental data, and extracting a plurality of quantified feature time series; selecting at least one algorithm in a preset set of time series causal inference algorithms to process the feature time series to identify a node pair having a predictive leading relationship; selecting at least one algorithm in a preset set of statistical correlation or dependence measure algorithms to calculate the association strength between the feature time series; and integrating the output results of the time series causal inference algorithm and the correlation or dependence measure algorithm to determine the existence, direction and quantified weight value of a data-driven association edge according to a preset fusion rule. 4.The AI-based health state intelligent analysis method of claim 2, wherein The establishment or strengthening of the semantic connection weight between the behavior node or feeling node and the target node comprises: analyzing the meaning association mark data to extract, in each mark event, the associated behavior node or feeling node as a source node and the target node as a target node; for each identified association pair, finding or creating a corresponding semantic connection edge in the meaning network model; and applying at least one of an event frequency-based cumulative update rule and a context-based weight adjustment rule to calculate the updated weight of the semantic connection edge. 5.The artificial intelligence-based health status intelligent analysis method of claim 1, wherein The network path analysis further comprises: for each searched valid path, calculating its comprehensive weight or significance score according to a preset path evaluation function, the evaluation function at least fusing the weight of each edge on the path, a path length factor, and recent activity information of the nodes involved in the path. 6.The artificial intelligence-based health status intelligent analysis method of claim 1, wherein, The generation of the personalized and meaning-oriented feedback information comprises: matching a corresponding feedback template identifier from the feedback strategy library according to the attributes of the highest-weight valid path identified by the network path analysis; and filling specific node information in the attributes of the identified highest-weight valid path into the matched feedback template to generate a preliminary feedback text. 7.The artificial intelligence-based health status intelligent analysis method of claim 1, wherein The method further comprises: providing an interactive control for evaluating the feedback effectiveness or relevance to the user while or after presenting the feedback information; processing and analyzing the collected evaluation data through a configured feedback learning module; and mapping the analysis result as a weight adjustment signal for the relevant edges in the personalized meaning network model or as an update of the priority or trigger condition of the strategy in the feedback strategy library to realize adaptive optimization of the system.

8. The health state intelligent analysis system based on artificial intelligence, characterized in that, The system is configured to perform the method of any one of claims 1 to 7, and the system comprises: a multi-dimensional data acquisition interface module for interacting with external sensors and user input devices; a personalized meaning network modeling engine comprising a data-driven association learner and a semantic connection learner, and being configured with a dynamic maintenance mechanism; a meaning-oriented feedback engine comprising a trigger logic judgment unit, a network path analyzer and a feedback generator; a user interaction module for presenting information, obtaining user marks and feedback; a feedback learning and model optimization module for realizing closed-loop adaptive adjustment. The data-driven association analysis comprises: preprocessing the physiological data, behavior data and environmental data, and extracting a plurality of quantified feature time series; selecting at least one algorithm in a preset set of time series causal inference algorithms to process the feature time series to identify a node pair having a predictive leading relationship; selecting at least one algorithm in a preset set of statistical correlation or dependence measure algorithms to calculate the association strength between the feature time series; and integrating the output results of the time series causal inference algorithm and the correlation or dependence measure algorithm to determine the existence, direction and quantified weight value of a data-driven association edge according to a preset fusion rule. The establishment or strengthening of the semantic connection weight between the behavior node or feeling node and the target node comprises: analyzing the meaning association mark data to extract, in each mark event, the associated behavior node or feeling node as a source node and the target node as a target node; for each identified association pair, finding or creating a corresponding semantic connection edge in the meaning network model; and applying at least one of an event frequency-based cumulative update rule and a context-based weight adjustment rule to calculate the updated weight of the semantic connection edge. The network path analysis further comprises: for each searched valid path, calculating its comprehensive weight or significance score according to a preset path evaluation function, the evaluation function at least fusing the weight of each edge on the path, a path length factor, and recent activity information of the nodes involved in the path. The generation of the personalized and meaning-oriented feedback information comprises: matching a corresponding feedback template identifier from the feedback strategy library according to the attributes of the highest-weight valid path identified by the network path analysis; and filling specific node information in the attributes of the identified highest-weight valid path into the matched feedback template to generate a preliminary feedback text. The method further comprises: providing an interactive control for evaluating the feedback effectiveness or relevance to the user while or after presenting the feedback information; processing and analyzing the collected evaluation data through a configured feedback learning module; and mapping the analysis result as a weight adjustment signal for the relevant edges in the personalized meaning network model or as an update of the priority or trigger condition of the strategy in the feedback strategy library to realize adaptive optimization of the system. The system is configured to perform the method of any one of claims 1 to 7, and the system comprises: a multi-dimensional data acquisition interface module for interacting with external sensors and user input devices; a personalized meaning network modeling engine comprising a data-driven association learner and a semantic connection learner, and being configured with a dynamic maintenance mechanism; a meaning-oriented feedback engine comprising a trigger logic judgment unit, a network path analyzer and a feedback generator; a user interaction module for presenting information, obtaining user marks and feedback; a feedback learning and model optimization module for realizing closed-loop adaptive adjustment. 9.The artificial intelligence-based health state intelligent analysis system of claim 8, wherein The system further comprises a model explainability submodule configured to perform at least one of the following operations when requested by the user or determined by the system: generating an evidence summary about a certain specific association edge based on the meaning network model and the relevant original data segments, the summary including the number of data points supporting the association, the association strength indicator, and the applicable user label records; visualizing the key network path on which a certain generated meaning-oriented feedback is based or explaining the generation logic thereof.

Images