Pregnancy risk pregnant and lying-in woman management system based on large language model and gamification

Abstract

The invention discloses a pregnancy risk pregnant and lying-in woman management system based on a large language model and gamification, which belongs to the technical field of intelligent medical treatment and comprises a large language module, a risk monitoring and evaluation module, a personalized intervention module and a gamification interaction module. The large language module is used for analyzing, standardizing and structuring the multi-dimensional pregnancy health data; the risk monitoring and evaluating module is used for dynamically identifying and grading pregnancy risk factors based on the structured data and generating risk grade data and early warning event data; the personalized intervention module is used for generating a customized management scheme according to the risk level data; and the gamification interaction module is used for mapping the personalized intervention scheme data into gamification task parameters and incentive feedback data and recording user interaction behavior data. According to the invention, intellectualization, personalization and participation degree of health management of pregnant and lying-in women are improved, and the problems of poor compliance, slow risk identification and single intervention means in existing pregnancy risk management are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, specifically to an intelligent management system for the entire pregnancy process through virtual scene interaction and AI decision support. Background Technology

[0002] With the adjustment of my country's fertility policy and changes in the social environment, the number of older pregnant women and those with pregnancy risks continues to increase, posing greater challenges to maternal and infant safety. Current pregnancy risk management models mainly rely on regular prenatal checkups, health education, and stratified management of high-risk pregnancies. While these methods have reduced the incidence of adverse pregnancy outcomes to some extent, they generally suffer from problems such as insufficient self-management ability among pregnant women, low adoption rates of healthy behaviors, and heavy workloads for medical staff. Especially for high-risk pregnant women, traditional management models struggle to achieve dynamic and personalized health interventions throughout the entire pregnancy, resulting in limited effectiveness in early risk identification and continuous management. Furthermore, with the diversification and individualization of maternal health needs, existing solutions often fail to adequately address emotional support, behavioral incentives, and multidimensional health education, lacking systematic tools and innovative technologies to promote self-management behaviors. Summary of the Invention

[0003] To address the aforementioned issues, this invention relates to a pregnancy risk management system for pregnant women based on a large language model and gamification. The system aims to enhance the intelligence, personalization, and participation of maternal health management, and to resolve problems such as poor self-management compliance, slow risk identification, and limited intervention methods in existing pregnancy risk management practices.

[0004] To achieve the above objectives, the present invention provides a pregnancy risk management system for pregnant women based on a large language model and gamification, characterized in that the method includes the following: The large language module is used to parse, standardize, and structure multi-dimensional pregnancy health data; The risk monitoring and assessment module is used to dynamically identify and classify pregnancy risk factors based on structured data, and obtain risk level data and early warning event data; The personalized intervention module generates a customized management plan based on the risk level data; The gamified interaction module is used to map personalized intervention plan data into gamified task parameters and incentive feedback data, and to record user interaction behavior data; Preferably, the large language module includes a natural language understanding unit, a multimodal data fusion unit, a knowledge reasoning unit, and an sentiment computing unit, as detailed below: The natural language understanding unit is used to parse free text or voice commands input by the user to obtain structured semantic data; The multimodal data fusion unit is used to perform field mapping, standardization, time alignment and fusion of structured semantic data and multidimensional pregnancy health data to obtain structured health profile data. The knowledge reasoning unit is used to retrieve, match, and reason about structured health profile data from pregnancy risk knowledge graphs, medical guidelines, and clinical evidence bases. The emotion computing unit is used to identify the user's emotional state and obtain interaction strategy parameter data.

[0005] The natural language understanding unit includes a speech transcription subunit, a semantic preprocessing subunit, an intent recognition subunit, an entity extraction and normalization subunit, and a context state management subunit, as detailed below: The speech-to-text subunit is used to convert user-input speech commands into text data; The voice command undergoes voice preprocessing; the voice preprocessing includes endpoint detection, noise suppression, and voice enhancement. After preprocessing, acoustic feature vectors are extracted, and speech recognition is performed based on an acoustic model and a language model to obtain a text sequence. The acoustic model adopts a speech recognition model based on a hidden Markov model or a deep neural network. The language model is used to perform contextual probability constraints and error correction on the recognition results to obtain text data containing timestamps. The semantic preprocessing subunit uses word segmentation, noise reduction, synonym normalization, and regularization rules to process text data including timestamps and normalize it to obtain normalized text. The intent recognition subunit is used to classify normalized text and calculate intent labels using a pre-trained semantic representation model and classification algorithm. The entity extraction and standardization subunit is used to extract symptom entities, test indicator entities, drug entities, food entities, time entities, and numerical entities from the standardized text and map them to a preset terminology to obtain entity information, and obtain constraint conditions based on the entity information. The context state management subunit is used to update the interaction context by fusing historical dialogue rounds and recent events to obtain structured semantic data; the structured semantic data includes intent tags, entity information, constraints, and interaction context; the interaction context is implemented using a dialogue state tracking and context memory update algorithm; The multimodal data fusion unit includes a data access and field mapping subunit, a data cleaning and standardization subunit, a missing and anomaly handling subunit, a time alignment subunit, and a feature construction subunit, as detailed below: The data access and field mapping subunit is used to receive multi-dimensional pregnancy health data and map the multi-dimensional pregnancy health data into a standard field format; the multi-dimensional pregnancy health data mapping is implemented using a preset field dictionary and encoding mapping rules. The data cleaning and standardization subunit is used to perform unit unification, code standardization, and noise removal on the mapped data; the unit unification and code standardization are achieved by matching unit conversion rules with a terminology table or code table. The missing and anomaly handling subunit is used to impute or mark missing values ​​and to identify and process anomalies; the imputation of missing values ​​is implemented using missing indicator marking and / or statistical imputation algorithms; the identification and processing of anomalies is implemented using threshold rules and statistical distribution detection algorithms. The time alignment subunit is used to align and merge data from different sources according to gestational age or a preset time window; the alignment and fusion of the time windows is achieved using a sliding time window resampling and aggregation algorithm; The feature construction subunit is used to obtain structured health profile data based on the fused data; the structured health profile data is implemented using feature encoding and vectorization representation algorithms; The knowledge reasoning unit includes a retrieval query construction subunit, an evidence retrieval subunit, a relevance ranking and evidence annotation subunit, a rule consistency verification subunit, and an explanation and candidate suggestion generation subunit. The specific process is as follows: The retrieval query construction subunit is used to generate retrieval queries using gestational age information, risk factor tags, and key indicator ranges from structured health profile data as retrieval conditions; the retrieval queries are implemented using field concatenation and templated query generation rules. The evidence retrieval subunit is used to retrieve a set of candidate evidence entries from pregnancy risk knowledge graphs, medical guidelines, and clinical evidence bases; the evidence retrieval subunit is implemented using keyword retrieval and vectorized semantic retrieval. The relevance ranking and evidence labeling subunit is used to rank the candidate evidence item set by relevance score to obtain evidence source identifier and evidence level label; the relevance score ranking is implemented by similarity calculation and ranking algorithm; the evidence level labeling is implemented by preset evidence classification rules and rule engine; The rule consistency verification subunit is used to verify the consistency of candidate suggestions and filter out non-compliant items based on contraindications related to gestational age, comorbidities, and applicable conditions. The consistency verification is implemented by using a rule engine to determine the contraindications and applicable conditions. The explanation and candidate suggestion generation subunit receives structured health profile data, risk status data, and a candidate evidence item set obtained from the evidence retrieval and ranking subunit. Under the rule consistency verification and applicable condition relationships, contraindications, and constraints in the pregnancy risk knowledge graph, the risk status data is mapped and analyzed to obtain risk element explanation data. Based on the verified candidate evidence item set, gestational age information, and risk level, a candidate suggestion set is generated through template matching and a rule engine to obtain suggestion generation intermediate data. The suggestion generation intermediate data includes the candidate suggestion set, evidence citation identifiers, and applicable condition constraints. The emotion computing unit includes an emotion feature extraction subunit, an emotion state calculation subunit, and an interaction strategy selection subunit, as detailed below: The emotion feature extraction subunit is used to receive user interaction text or speech features and obtain emotion feature vectors; the extraction process of the emotion feature vectors is implemented using text emotion feature encoding and speech prosody feature extraction algorithms; The emotion state calculation subunit is used to classify or score the emotion feature vector to obtain emotion level data; the classification or scoring calculation is implemented using an emotion classification model and an emotion scoring model. The interaction strategy selection subunit is used to select an interaction strategy template based on emotion level data and risk level data and output interaction strategy parameter data; the interaction strategy parameter data includes at least prompt tone parameters, information density parameters, interaction frequency parameters, and trigger identifiers for escalation to manual or medical team intervention; the selection of the interaction strategy template is implemented using a rule engine and a state machine. Preferably, the risk monitoring and assessment module includes a physiological indicator monitoring unit, a behavioral pattern analysis unit, a psychological state assessment unit, and a risk early warning unit, specifically including the following: The physiological indicator monitoring unit is used to receive raw vital sign data collected from smart devices and examination, test and diet data uploaded by users, and to record timestamps, convert units and mark outliers in the raw vital sign data and calculate physiological indicator monitoring data. The behavior pattern analysis unit is used to receive the physiological indicator monitoring data and user behavior log data, extract and score or grade the diet, exercise and sleep-related features to obtain behavior assessment data. The psychological state assessment unit is used to receive user interaction content and psychological scales, calculate and map the psychological scale scores and emotion levels to obtain psychological state assessment data. The risk warning unit is used to receive the structured health profile data obtained by the large language module, and integrate physiological indicator monitoring data, behavioral assessment data and psychological state assessment data for feature aggregation. It calls the preset rule model or machine learning model to calculate the risk score and map it to the risk level. When the preset threshold or rule conditions are met, it obtains warning event data and risk status data. The physiological indicator monitoring unit includes a device docking subunit, a data verification subunit, an indicator calculation subunit, and a storage and output subunit, as detailed below: The device interface subunit is used to interface with blood pressure monitors, blood glucose meters, and smart bracelets to collect raw vital sign data; the data collection from blood pressure monitors, blood glucose meters, and smart bracelets is achieved through communication interface adaptation and device protocol parsing. The data verification subunit is used to perform unit unification, timestamp standardization, and outlier marking on the original data to obtain verified data; the unit unification and timestamp standardization are implemented using preset conversion rules and time synchronization rules; the outlier marking is implemented using threshold rules and statistical distribution detection algorithms. The indicator calculation subunit is used to perform interval determination and trend calculation on the verified data to obtain physiological indicator monitoring data; the trend calculation is implemented using a sliding time window statistical and trend fitting algorithm, and the interval determination is implemented using a segmented threshold mapping rule. The storage output subunit receives key physiological indicator feature data obtained from the indicator calculation subunit, encapsulates the physiological indicator feature data in a structured manner, associates it with timestamp identifier, indicator type identifier and data source identifier, and then inputs it into the structured data storage structure. In accordance with the preset data interface specification, the physiological indicator monitoring data is output to the feature aggregation subunit of the risk warning unit in the form of feature fields. The behavior pattern analysis unit includes a behavior log collection subunit, a feature extraction subunit, and a scoring and grading subunit, as detailed below: The behavior log collection subunit is used to receive check-in records, step count synchronization, food photos, and sleep duration records to form behavior log data; the formation of the behavior log data is achieved by encapsulating log schema and normalizing timestamps; The feature extraction subunit is used to extract dietary, exercise, and sleep features from behavioral log data and physiological indicator monitoring data; the extraction of dietary, exercise, and sleep features is achieved using statistical aggregation, rule calculation, and feature engineering algorithms. The scoring and grading subunit is used to calculate compliance scores and risk grading based on preset rules and obtain behavioral assessment data; the compliance scores are implemented using a rule-based scoring model; the risk grading calculation is implemented using threshold mapping rules and regression models or machine learning models; The psychological state assessment unit includes a scale management subunit, an interactive emotion recognition subunit, and a score mapping subunit, as detailed below; The scale management subunit is used to push and receive psychological scales; the psychological scales are implemented using a scale template library and a questionnaire distribution and collection mechanism. The interactive emotion recognition subunit is used to extract emotion features from user interaction content to obtain emotion index data; the emotion feature extraction is implemented using text sentiment feature encoding and speech prosody feature extraction algorithms. The scoring mapping subunit is used to calculate the scores of the psychological scale and fuse them with the emotion index data to map them into psychological levels, thereby obtaining psychological state assessment data. The calculation of the psychological scale scores is implemented using pre-designed scoring rules, weighted fusion, threshold grading mapping rules, and a classification model. The risk warning unit includes a feature aggregation subunit, a risk scoring subunit, a hierarchical mapping subunit, and an event triggering subunit, as detailed below: The feature aggregation subunit is used to aggregate structured health profile data with physiological indicator monitoring data, behavioral assessment data, and psychological state assessment data through time windows to obtain a risk feature vector; the time window aggregation is implemented using a sliding time window resampling and statistical aggregation algorithm; The risk scoring subunit is used to perform weighted fusion calculations on risk feature vectors or to call a machine learning model to obtain a risk score; the weighted fusion calculation adopts a preset weighting rule; the machine learning model uses a regression model and an interpretive machine learning model to implement the risk scoring. The hierarchical mapping subunit is used to map risk scores to risk levels; the risk score mapping is implemented using threshold mapping rules or piecewise functions. The event triggering subunit is used to generate early warning event data and output the risk status data when the key indicator exceeds the threshold, the score jumps, or the continuous abnormal conditions are met. The determination of the key indicator exceeding the threshold, the score jumps, or the continuous abnormal conditions is implemented by a rule engine and a state machine to trigger the event. Preferably, the personalized intervention module includes a health education unit, a behavior guidance unit, an emergency guidance unit, and a remote collaboration unit, as detailed below: The health education unit is used to receive gestational age stage information, individual risk element tags, and risk level and warning event type from the structured health profile data as input conditions, and to perform retrieval and matching in the pregnancy risk knowledge graph and organize evidence to obtain age-appropriate, stage-specific pregnancy knowledge content data and obtain knowledge source identification and evidence basis. The behavior guidance unit is used to receive risk status data and structured health profile data as input to obtain behavior guidance data and task list and task parameter data mapped by the gamified interaction module; The emergency guidance unit is used to receive risk level and early warning event data as input, match the disposal path template, and obtain disposal plan data, upgrade trigger condition data, and re-evaluation node data. The remote collaboration unit is used to receive risk summary data and treatment plan data under authorized conditions, encapsulate them into remote collaboration instruction data and push them to the medical team, and at the same time receive feedback from the medical team to form collaboration receipt data. The health education unit includes a search condition construction subunit, an evidence retrieval and ranking subunit, an evidence classification and tracing subunit, and a content organization subunit, as detailed below: The search condition construction subunit is used to construct search conditions based on gestational age information, risk level, warning event type, and risk element tags; the construction of search conditions is achieved by field concatenation and templated query generation rules. The evidence retrieval and ranking subunit is used to retrieve a set of candidate knowledge items from the pregnancy risk knowledge graph, medical guidelines, and clinical evidence base and rank them by relevance. The retrieval is achieved by keyword retrieval and vectorized semantic retrieval, and the relevance ranking is achieved by similarity calculation and ranking algorithms. The evidence grading and tracing subunit is used to generate evidence levels and source identifiers for a set of candidate knowledge items; the source identifier is implemented using a source identifier field mapping and reference chain recording mechanism; the evidence level is implemented using preset evidence grading rules and a rule engine. The content organization subunit is used to generate a knowledge content data package based on the sorting results and applicable conditions, and to obtain the resulting knowledge content data, as well as the knowledge source identifier and evidence; the content organization subunit is implemented using a content template library and parameter filling algorithm to generate the knowledge content data package. The behavior guidance unit includes a goal setting subunit, an individualized constraint loading subunit, a scheme generation subunit, and a task parameter generation subunit, as detailed below: The target setting subunit is used to determine behavior management targets based on risk status data; the behavior management targets are implemented using rule mapping and threshold determination. The individualized constraint loading subunit is used to load gestational week contraindications, comorbidity contraindications, and user preferences to form a set of constraint conditions; the set of constraint conditions is implemented using a contraindication rule base and a user preference parameter loading mechanism. The scheme generation subunit is used to obtain behavior guidance data based on the set of behavior management objectives and constraints; the behavior guidance data is implemented using an objective-constraint rule calculation and planning algorithm. The task parameter generation subunit is used to convert the behavior guidance data into a task list and task parameter data and output them to the gamification interaction module; the task parameter data generation is implemented using a task template library matching and parameter calculation algorithm. The emergency guidance unit includes an event triage subunit, a response path matching subunit, an escalation rule generation subunit, and a review node generation subunit, as detailed below: The event triage subunit is used to determine the event category and handling priority based on the warning event type and risk level; the risk level is used to determine the event category using event type mapping rules and priority scoring rules. The disposal path matching subunit is used to match a preset disposal path template and generate disposal plan data; the disposal path template is implemented using a disposal path template library and a rule engine. The upgrade rule generation subunit is used to generate upgrade trigger condition data; the upgrade trigger condition data is implemented using threshold rules, continuous anomaly rules, and state machine rules. The review node generation subunit is used to generate review node data and output it to the intervention plan data; the generation of review node data is implemented using a time window rule based on gestational age stage and event category. The remote collaboration unit includes an authorization management subunit, an information encapsulation subunit, a message push subunit, and a receipt processing subunit, as detailed below: The authorization management subunit is used to verify the authorization information of pregnant women and generate an authorization identifier; the authorization identifier is implemented using identity authentication and authorization strategy rules; The information encapsulation subunit is used to encapsulate risk summary data, trigger indicators, risk levels, and disposal plan data into remote collaboration instruction data; the remote collaboration instruction data is implemented using instruction schema and field templates and undergoes necessary desensitization processing. The message push subunit is used to push remote collaboration instruction data to the medical team; the push of remote collaboration instruction data is implemented using interface calls and message queues and supports encrypted transmission; The receipt processing subunit is used to receive feedback from the medical team and generate collaborative receipt data to update the intervention plan status data; the generation of collaborative receipt data is implemented using a receipt state machine and log recording mechanism and triggers the update of the intervention plan status data. Preferably, the gamified interaction module includes: a task mapping unit, an achievement system unit, a virtual character unit, a social interaction unit, a progress visualization unit, and a behavior log collection unit, as detailed below: The task mapping unit includes a scheme parsing subunit, a template matching subunit, a parameter generation subunit, and a task publishing subunit, as detailed below: The intervention plan parsing subunit is used to parse the intervention plan data into a structured set of elements: objectives, frequencies, and constraints; the intervention plan data parsing is implemented using rule parsing and semantic parsing algorithms. The template matching subunit is used to match task types and scene templates from the task template library based on the structured feature set. The task type matching in the task template library is implemented using rule matching and similarity matching algorithms based on label constraints. The parameter generation subunit is used to calculate the target threshold, completion judgment condition, reward trigger condition and deadline based on the target and constraints; the generation of the target threshold, completion judgment condition, reward trigger condition and deadline is implemented by threshold mapping and condition calculation rules; The task publishing subunit is used to output the task list and game task parameter data, and to provide level presentation parameters to the progress visualization unit; the task list is implemented using a task queue and event notification mechanism. The achievement system unit includes a completion calculation subunit, a reward rule engine subunit, and a feedback generation subunit, as detailed below: The completion calculation subunit is used to calculate the task completion rate and continuous achievement status based on user interaction behavior data; the completion calculation subunit is implemented using event counting, threshold determination, and continuous status tracking algorithms; The reward rule engine subunit is used to calculate points, badges, and levels based on the completion rate according to the reward rules and generate reward event data; the reward rules use a rule engine and a state machine to trigger the calculation of points, badges, and levels. The feedback generation subunit is used to convert reward event data into incentive feedback data and output it to the progress visualization unit and the virtual character unit to drive interface feedback and growth presentation; the incentive feedback data is implemented using a feedback template library and parameter filling algorithm. The virtual character unit includes a character parameter generation subunit, a growth path mapping subunit, and a rendering subunit, as detailed below: The role parameter generation subunit is used to generate basic role parameters based on gestational age, risk factor tags, and user preferences in the structured health profile data; the basic role parameters are implemented using profile parameter mapping rules and preference weight fusion algorithms; The growth path mapping subunit is used to map task completion, score level, and stage goal into growth status data; the mapping of task completion, score level, and stage goal into growth status data is implemented by a state machine and a score mapping rule to calculate the growth status. The rendering subunit is used to output character rendering parameters and display character appearance changes, unlocked content, and growth animations in the interface; the character rendering parameters are based on preset resource library and animation template parameters. The social interaction unit includes a content generation and review subunit, a matching and recommendation subunit, and an interactive challenge management subunit, as detailed below: The content generation and review subunit is used to structure and encapsulate experience-sharing texts and challenge content and review them according to preset compliance rules; the structured encapsulation is implemented using content schema and field templates; the compliance rule review is implemented using a rule engine and sensitive information identification algorithm; The matching and recommendation subunit is used to match and recommend interactive objects or content based on gestational age, role presentation task type, and risk level; the matching and recommendation is implemented using rule matching and similarity calculation recommendation algorithms based on tag constraints; The interactive challenge management subunit is used to obtain mutual assistance challenge parameters and output social interaction content data, while recording the interaction process to form user interaction behavior data; the mutual assistance challenge parameters are generated and the process state is updated using a challenge template library and a state machine; The progress visualization unit includes an indicator mapping subunit, a trend calculation subunit, and a graphics rendering subunit, as detailed below: The indicator mapping subunit is used to map structured health profile data, risk level data, and task completion data into a set of visual indicators; the mapping of structured health profile data, risk level data, and task completion data into a set of visual indicators is implemented using an indicator dictionary and rule mapping algorithm; The trend calculation subunit is used to perform time window aggregation and trend calculation on the set of visualization indicators to generate trend feature data; the time window aggregation and trend calculation are implemented using sliding time window statistical aggregation and trend fitting algorithms. The graphics rendering subunit is used to convert the trend feature data into trend charts, progress bars, milestones, or level progress and output visualization parameters; the conversion of the trend feature data into trend charts, progress bars, milestones, or level progress uses a graphics rendering engine and generates the visualization parameters based on a preset chart template. The behavior log collection unit includes an event collection subunit, a log structuring subunit, and a reflow update subunit, as detailed below: The event collection subunit is used to collect user interaction events during task execution, check-in, assessment, and social interaction; the interaction events are implemented using client-side event tracking and event listening mechanisms. The log structuring subunit is used to encapsulate interactive events into timestamped user interaction behavior data; the encapsulation of interactive events into timestamped user interaction behavior data is achieved by encapsulating them using an event model and a log schema, and then normalizing the timestamps. The backflow update subunit is used to provide user interaction behavior data to the core engine of the large language module for updating the structured health profile data, and to provide it to the risk monitoring and assessment module and the personalized intervention module for dynamically adjusting the risk status data or intervention plan data; the dynamic adjustment of risk status data or intervention plan data adopts a message queue or interface call mechanism to realize data distribution and update triggering.

[0006] This invention is based on a large language model intelligent engine, which can deeply analyze multimodal health data of pregnant and postpartum women, dynamically identify and classify pregnancy risks, and realize personalized and dynamic management of maternal health throughout the entire pregnancy, significantly improving the timeliness and scientific nature of risk identification. This invention combines health education, behavioral guidance, and a virtual reward system through an innovative and serious gamified interactive mechanism, enhancing user engagement and interactive experience, effectively promoting the adoption and long-term adherence to healthy behaviors, and improving the initiative and sustainability of self-management; This system integrates real-time monitoring and evaluation of multi-source data, including physiological, behavioral, and psychological data. By combining big data knowledge reasoning and medical evidence, it can accurately push personalized intervention plans at different levels, covering health education, behavior correction, risk warning, and emergency guidance. Furthermore, by utilizing artificial intelligence and automated processes, the system significantly reduces the health management burden on medical staff, enabling intelligent screening, dynamic intervention, and continuous follow-up of large-scale high-risk pregnant women, thereby improving management efficiency and service coverage. Attached Figure Description

[0007] Figure 1 This is a structural diagram of a pregnancy risk management system for pregnant women based on a large language model and gamification, according to the present invention. Detailed Implementation

[0008] The following detailed description of specific embodiments is provided to better understand the present invention and is not limited to the preferred embodiments. It does not limit the content and scope of protection of the present invention. Any product that is the same as or similar to the present invention, derived by any person under the guidance of the present invention or by combining the features of the present invention with other prior art, falls within the protection scope of the present invention.

[0009] The technical solution of the present invention will be described in more detail and completely below with reference to the accompanying drawings; like Figure 1 As shown, the pregnancy risk management system for pregnant women based on a large language model and gamification proposed in this invention includes the following: The large language module is used to parse, standardize, and structure multi-dimensional pregnancy health data; The large language module includes a natural language understanding unit, a multimodal data fusion unit, a knowledge reasoning unit, and an affective computing unit. This engine can parse text or voice input from pregnant women, integrate clinical indicators, behavioral data, and environmental factors, generate personalized health recommendations based on medical knowledge, and identify the user's emotional state to adjust interaction strategies. The emotional state is assessed using the Psychological Evaluation Scale (EPDS), which consists of 10 items, each scored from 0 to 3 points, for a total score of 0 to 30 points. Users complete the scale periodically or as prompted by the system; the system automatically records and calculates the total score. The system uses EPDS scores to classify user emotional states: scores below 9 indicate a daily health management mode, focusing on pushing maternal and infant health knowledge and daily behavioral incentives; scores between 9 and 12 suggest possible mild depressive symptoms, and the system automatically switches to "Attention Mode," pushing more psychological care content, using a gentler tone, and providing timely emotional guidance suggestions and positive encouragement; scores ≥ 13 suggest possible postpartum depression symptoms, and the system enters "High Attention Mode," pushing mental health science information, access to psychological counseling appointments, family support suggestions, and advising users to proactively contact professional doctors; simultaneously, risk information is pushed to the medical team (with user authorization) for follow-up of key populations.

[0010] The risk monitoring and assessment module is used to dynamically identify and classify pregnancy risk factors based on structured data, and generate risk level data and early warning event data. The risk monitoring and assessment module includes a physiological indicator monitoring unit, a behavioral pattern analysis unit, a psychological state assessment unit, and a risk early warning unit. This module can collect real-time data on pregnant and postpartum women's vital signs and behavioral habits, assess emotional fluctuations using professional questionnaires or interactive content, and establish a multi-parameter comprehensive risk assessment model to achieve dynamic risk grading and early warning. The system supports connection to Bluetooth / smart wearable devices (such as blood pressure monitors, blood glucose meters, and smart bracelets) to automatically collect data on blood pressure, blood glucose, weight, heart rate, etc., enabling vital sign data collection. Behavioral habit data is collected through the app's built-in check-in, synchronized step count, food photo uploads, and automatic / automatic recording / self-entry of sleep duration, forming a daily behavioral log of diet, exercise, and sleep. Psychological state assessment data is collected by periodically pushing the EPDS scale, combined with user-initiated emotional words. The multi-parameter comprehensive risk assessment model and tiered early warning system utilizes weighted scoring or machine learning methods to comprehensively analyze the collected physiological parameters, dietary and exercise behaviors, sleep quality, and psychological scale scores, assigning risk indicator weights. For example, hypertension and abnormal blood sugar are hard high-risk factors, while poor behavioral compliance, sleep disorders, and abnormal emotional scores are auxiliary risk factors. The system classifies risk states into three levels: "normal," "attention," and "high-risk," dynamically updating with the entered data. If key indicators (such as blood pressure, blood sugar, and EPDS score) exceed preset thresholds, the system automatically issues a pop-up warning, pushing it to the user and their family, and, with authorization, simultaneously notifying the medical team to prompt immediate intervention.

[0011] The personalized intervention module generates a customized management plan based on the risk level data; The personalized intervention module includes health education, behavioral guidance, emergency guidance, and remote collaboration units. Based on risk assessment results, the system automatically pushes phased health knowledge, gamified health tasks, and emergency response plans, and supports remote guidance and follow-up by medical teams. Health education pushes are automatically tailored to age, personalized health knowledge, dietary recommendations, and exercise guidance based on gestational week, risk level, and behavioral performance. Behavioral task guidance generates daily health task lists, such as dietary check-ins, step count goals, and psychological scales, with virtual rewards for completing tasks. Emergency guidance and remote collaboration automatically pop up emergency response suggestions and remote consultation appointment portals when high-risk events (such as high blood pressure or high depression scores) are detected, supporting remote intervention by medical teams.

[0012] The gamified interaction module is used to map personalized intervention plan data into gamified task parameters and incentive feedback data, and to record user interaction behavior data.

[0013] The gamified interaction module includes an achievement system, a virtual character system, a social interaction system, and a progress visualization system. By setting health management milestones, virtual rewards, and growth paths, as well as experience sharing and mutual support challenges among pregnant women and new mothers, user engagement and adherence are enhanced. The virtual reward system offers each user various reward formats, including points, badges, virtual items, and a personalized honor wall. Users automatically receive points, honor badges, and cartoon character accessories for completing tasks such as daily health check-ins, dietary management, regular vital sign measurements, participation in psychological questionnaires, and answering questions. Points can be redeemed for maternal and infant knowledge booklets, virtual character clothing, etc. The growth path is structured as a "health growth tree" or "virtual baby development" route for users in stages. As users achieve milestones at different stages of pregnancy, their virtual character (such as a baby or mother) "grows"—for example, a sapling grows leaves, the baby unlocks new movements, or a new photo is added to the family digital album. The growth path visually reflects the user's health management progress, encouraging continued participation.

[0014] The health knowledge, dietary recommendations, and exercise guidance content of this invention are derived from authoritative medical literature, clinical guidelines, expert consensus, and relevant national / regional regulations. The system has a built-in knowledge base, which mainly includes, but is not limited to: "Guidelines for the Health Management of Pregnant and Lactating Women in China", "Dietary Guidelines for Pregnant and Lactating Women in China (2022)", "Dietary Guidelines for Gestational Diabetes Mellitus", "Guidelines for the Diagnosis and Treatment of Hyperglycemia in Gestation (2024)", "Expert Consensus on the Management of Hypertensive Disorders Complicated with Pregnancy", "Expert Consensus on the Management of Blood Pressure in Hypertensive Disorders of Pregnancy (2019)", relevant maternal and child health guidelines from the World Health Organization (WHO), and the latest evidence-based medical literature from authoritative domestic and international journals. The pregnancy risk knowledge graph assigns the following tags to each knowledge point: applicable gestational week (e.g., early / mid / late pregnancy), risk level (normal / concern / high risk), behavioral characteristics (e.g., high / low exercise adherence, dietary preferences, psychological state), and applicable scenarios (e.g., pregnancy complications, special nutritional needs).

[0015] The pregnancy risk knowledge graph is stored in the form of a graph database or graph structure index. Graph nodes include at least risk factor entities, clinical indicator entities, behavioral factor entities, psychosocial factor entities, gestational age stage entities, outcome / event entities, and intervention entity entities. Node attributes include at least standard terminology codes, synonym sets, units and value ranges, applicable gestational age labels, risk level labels, applicable context labels, and evidence level / source identifiers. Graph edges include at least association relationships, causal / facilitative relationships, contraindication / constraint relationships, and applicable condition relationships. Edge attributes include at least relationship direction, conditional expression, confidence parameter, and evidence citation index. The conditional expression describes the gestational age stage, contraindications for complications, and indicator threshold conditions for the relationship to hold. The confidence parameter can be based on evidence level labeling, guideline source weights, and / or statistical consistency results. The system initializes and updates data; it establishes inverted and vector indexes for frequently queried fields. The inverted index is used to quickly recall a set of candidate knowledge items based on standard terminology encoding, tag fields, and relation types. The vector index is used to vectorize user input text, structured health profile features, or retrieval queries and recall a set of relevant evidence fragments based on semantic similarity. The recall results are then scored and ranked in the relevance ranking and evidence annotation logic to output a traceable evidence citation index, thereby supporting rapid retrieval and traceable citation for risk interpretation and health education. The inverted index can be implemented using key-value indexes and field indexes, the vector index can be implemented using vectorized representation and approximate nearest neighbor retrieval algorithms, and the relevance ranking can be implemented using similarity calculation and ranking algorithms while retaining the retrieval hit fields and evidence citation indexes to support auditing.

[0016] The construction process of the pregnancy risk knowledge graph includes at least the following steps: acquiring knowledge source data and forming a construction dataset, wherein the knowledge source data includes at least medical guideline texts, clinical evidence literature and / or expert rule entries, and may further include structured fields from historical cross-sectional survey data and / or follow-up data; defining entity types and relation types for the knowledge source data based on a preset pregnancy risk ontology or data model; performing entity extraction and relation extraction on the knowledge source data and performing synonym normalization, unit normalization and encoding mapping to generate graph triple data; performing consistency verification and conflict resolution on the graph triple data and writing it into a graph database to form a queryable pregnancy risk knowledge graph; in the inference stage, the system constructs query conditions containing gestational age, risk level, key indicator anomalies and behavioral characteristics based on structured health profile data and enters the retrieval query construction and evidence retrieval logic to perform the pregnancy risk knowledge graph construction. The risk knowledge graph performs retrieval and matching, and outputs risk element explanation data, evidence citation indexes, and / or suggested intermediate data. The output further serves as constraints and traceable basis for the large language module to generate health education content, behavioral guidance task parameters, and emergency response prompts. When the system receives new guideline versions, new literature evidence, or new labeled data, it performs synonym normalization, encoding mapping, and conflict detection on the new data and incrementally writes the graph triples, recording the version number and change log to ensure the auditability of the reasoning results. Among these, entity extraction and relation extraction can be implemented using sequence labeling models and / or rule extraction algorithms; synonym normalization and encoding mapping can be implemented using terminology matching and similarity matching algorithms; consistency verification and conflict resolution can be implemented using rule engines and constraint checking algorithms; and incremental writing and version management can be implemented using version control and incremental update mechanisms, recording the source identifier and update timestamp.

[0017] Assigning and updating user gestational age, risk level, and behavioral performance: (1) Gestational age assignment: When registering, users enter their last menstrual period / expected delivery date, and the system automatically calculates the current gestational age, which automatically increases over time. Users / doctors can manually correct the gestational age.

[0018] (2) Risk grading assignment: The risk grading is automatically calculated based on vital signs (such as blood pressure and blood sugar), weight gain, psychological scale scores, past medical history, and obstetric history. The system has grading rules (for example: blood pressure ≥140 / 90 mmHg is high risk, EPDS ≥13 is psychological high risk, and weight gain exceeding the recommended range is of concern). The risk grading is divided into three categories: "normal", "concern", and "high risk", and the grading results are dynamically updated with the data.

[0019] (3) Assignment of behavioral performance: Through daily check-in, wearable devices, health logs and other data collection, the system records behavioral indicators such as diet, exercise, rest and psychology. The system sets scoring rules for each behavior (e.g., ≥5 times of exercise per week is "high compliance", <3 times is "low compliance"), and dynamically adjusts them based on historical performance.

[0020] The system's knowledge and user status matching and push mechanism automatically calls the large language module and knowledge graph reasoning unit daily / weekly to read the user's gestational week, risk level, and scores for various behaviors. Matching principles prioritize pushing content whose "Applicable Gestational Week" tag matches (or is adjacent to) the user's gestational week; content with a risk level tag greater than or equal to the user's current risk level (e.g., high-risk users receive high-risk + attention + regular content, normal users receive only regular content); and content whose behavioral characteristic tags match or indicate problems with the user's current behavior is prioritized (e.g., those with low exercise adherence receive priority exercise guidance, those with high adherence receive advanced content or challenge tasks). A specific push process example: A user is 22 weeks pregnant, has a "high-risk" risk level, and low exercise adherence. The system searches the knowledge base for exercise suggestions and mental health guidance applicable to those tagged "mid-pregnancy," "high-risk," and "low exercise adherence," and pushes content such as "Exercise Precautions for High-Risk Pregnant Women in Mid-Pregnancy" and "Emotional Adjustment Techniques." Simultaneously, the system automatically updates the pushed content regularly based on the latest medical guideline changes.

[0021] Knowledge delivery formats and feedback mechanisms: Delivery content includes various formats such as text-based science popularization, video tutorials, interactive Q&A, and daily tips. After each learning session or task completion, users can proactively provide feedback on whether the knowledge was useful or not. The system records this feedback and optimizes subsequent delivery strategies, creating a personalized knowledge curve. Medical teams can customize and supplement the knowledge delivery content and conduct manual reviews to ensure its scientific accuracy and applicability.

[0022] The auxiliary modules include a family support system (allowing family members to participate in health management), a medical data platform (connecting to hospital information systems), a privacy protection engine (data anonymization and tiered authorization), and a multi-terminal adapter (compatible with mobile devices, VR, etc.). The family support system allows family members to authorize joining the maternal and infant health management account, view real-time health data and task completion status, and receive family care suggestions pushed by the system. The system encourages family members to participate in health tasks (such as accompanying exercise and sharing meal check-ins) and features a dedicated "family achievement wall." The medical data platform automatically synchronizes medical orders such as outpatient examinations, tests, and medications for pregnant women through standardized system integration with hospital HIS and maternal and infant follow-up systems, assisting doctors in dynamically adjusting management plans. The privacy protection engine encrypts and stores sensitive user data (such as medical information and psychological scale scores), with tiered data authorization; users can set the visibility range for family members and doctors. All operations are logged, and anonymized data export is supported. The multi-terminal adapter supports access from Android and iOS apps, mini-programs, web pages, and smart wearable devices; some interactive content supports voice broadcasting and VR visualization, facilitating user use in multiple scenarios.

[0023] System Operation Flow: User Registration and Initial File Creation: Pregnant women register via mobile app, web page, or WeChat mini-program, entering basic information (name, age, last menstrual period / due date, past pregnancy and childbirth history, complications, etc.). The system automatically creates a personal health record and calculates gestational age. Multidimensional health data collection follows, including physiological indicators, behavioral habits, and psychological scales. After information collection, user status is assigned and risk is categorized. The system automatically determines the current gestational age based on the collected data. Based on physiological indicators, psychological scales, and behavioral scores, the system uses a multi-parameter model to dynamically categorize users as "Normal," "Attention," or "High Risk." Intelligent Knowledge / Task Push: The system utilizes a structured knowledge base and a large language module, combined with the user's gestational age, risk level, and behavioral performance, to accurately match and push age-appropriate, risk-appropriate, and behaviorally appropriate health knowledge, dietary advice, and exercise guidance. Push formats include text, video, task lists, and Q&A interactions. Furthermore, the system establishes gamified incentives and growth feedback, helping users complete daily health check-ins, behavioral tasks, and psychological scales to earn virtual rewards such as points, badges, and virtual character growth values, with a dynamic and visually displayed growth path. The system then monitors key indicators for tiered warnings and personalized interventions. If abnormalities are detected (such as high blood pressure, high blood sugar, or high scores on depression scales), the system automatically sends a pop-up warning, pushes emergency treatment suggestions, and, when necessary (with authorization), synchronizes information with family members and doctors, supporting remote intervention. A family-doctor collaboration model is also established, allowing family members to be invited to join collaborative management, view some of the pregnant woman's health data, and participate in family tasks. The medical team can remotely follow up, access data, and customize personalized knowledge pushes and intervention suggestions through the doctor's end. The large language module further includes: The clinical decision support unit provides treatment recommendations based on the latest medical evidence. The terminology conversion unit enables the mutual translation of professional medical concepts and colloquial expressions; Multilingual support unit to meet the needs of pregnant women in different regions; Cultural adaptation unit, taking into account regional customs and religious differences; The gamified interaction module further includes: The narrative design unit constructs the storyline of the pregnancy process; the mini-game library provides a variety of health-promoting mini-games; the virtual companion system features an AI prenatal doctor role; and the data art unit transforms health data into visual art works. The risk monitoring and assessment module further includes: The genetic risk assessment unit analyzes family medical history and genetic data. Environmental risk scanning unit monitors hazards in living and working environments; A social support assessment unit to quantify the level of family and social support; Compliance prediction unit, assessing the likelihood of implementing the intervention plan; The personalized intervention module further includes: The nutrition guidance unit provides personalized dietary plans. The exercise prescription unit designs safe exercise plans. The psychological adjustment unit provides mindfulness training and relaxation techniques. The delivery preparation unit provides prenatal education and simulation training. The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0024] Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit its scope of protection. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that any modifications, substitutions, or equivalent variations to the technical content without departing from the core idea of ​​the present invention should be considered to fall within the scope of protection of the present invention, and the scope of protection of the present invention should be determined by the claims.

Claims

1. A pregnancy risk management system for pregnant women based on a large language model and gamification, characterized in that, The management system includes a large language module, a risk monitoring and assessment module, a personalized intervention module, and a gamified interaction module; The large language module is used to parse, standardize, and structure multi-dimensional pregnancy health data; The risk monitoring and assessment module is used to dynamically identify and classify pregnancy risk factors based on structured data, and obtain risk level data and early warning event data. The personalized intervention module generates a customized management plan based on risk level data; The gamified interaction module is used to map personalized intervention plan data into gamified task parameters and incentive feedback data, and to record user interaction behavior data.

2. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 1, characterized in that, The large language module includes a natural language understanding unit, a multimodal data fusion unit, a knowledge reasoning unit, and an sentiment computing unit, as detailed below: The natural language understanding unit is used to parse free text or voice commands input by the user to obtain structured semantic data; The multimodal data fusion unit is used to perform field mapping, standardization, time alignment and fusion of structured semantic data and multidimensional pregnancy health data to obtain structured health profile data. The knowledge reasoning unit performs retrieval, matching, and reasoning on structured health profile data based on a pregnancy risk knowledge graph; The emotion computing unit is used to identify the user's emotional state and obtain interaction strategy parameter data.

3. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 2, characterized in that, Specifically, it includes the following: The Natural Language Understanding unit includes a speech transcription subunit, a semantic preprocessing subunit, an intent recognition subunit, an entity extraction and normalization subunit, and a context state management subunit, as detailed below: The speech transcription subunit is used to convert user-inputted voice commands into text data; the voice commands undergo speech preprocessing; the speech preprocessing includes endpoint detection, noise suppression, and speech enhancement; after preprocessing, acoustic feature vectors are extracted, and speech recognition is performed based on an acoustic model and a language model to obtain a text sequence; the acoustic model adopts a speech recognition model based on a hidden Markov model or a deep neural network; the language model is used to perform contextual probability constraints and error correction on the recognition results to obtain text data containing timestamps; The semantic preprocessing subunit uses word segmentation, noise reduction, synonym normalization, and regularization rules to process text data including timestamps and normalize it to obtain normalized text. The intent recognition subunit is used to classify normalized text and calculate intent labels using a pre-trained semantic representation model and classification algorithm. The entity extraction and standardization subunit is used to extract symptom entities, test indicator entities, drug entities, food entities, time entities, and numerical entities from the standardized text and map them to a preset terminology to obtain entity information, and obtain constraint conditions based on the entity information. The context state management subunit is used to update the interaction context by fusing historical dialogue rounds and recent events to obtain structured semantic data; the structured semantic data includes intent tags, entity information, constraints, and interaction context; the interaction context is implemented using a dialogue state tracking and context memory update algorithm; The multimodal data fusion unit includes a data access and field mapping subunit, a data cleaning and standardization subunit, a missing and anomaly handling subunit, a time alignment subunit, and a feature construction subunit, as detailed below: The data access and field mapping subunit is used to receive multi-dimensional pregnancy health data and map the multi-dimensional pregnancy health data into a standard field format; the multi-dimensional pregnancy health data mapping is implemented using a preset field dictionary and encoding mapping rules. The data cleaning and standardization subunit is used to perform unit unification, code standardization, and noise removal on the mapped data; the unit unification and code standardization are achieved by matching unit conversion rules with a terminology table or code table. The missing and anomaly handling subunit is used to impute or mark missing values ​​and to identify and process anomalies; the imputation of missing values ​​is implemented using missing indicator marking and / or statistical imputation algorithms; the identification and processing of anomalies is implemented using threshold rules and statistical distribution detection algorithms. The time alignment subunit is used to align and merge data from different sources according to gestational age or a preset time window; the alignment and fusion of the time windows is achieved using a sliding time window resampling and aggregation algorithm; The feature construction subunit is used to obtain structured health profile data based on the fused data; the structured health profile data is implemented using feature encoding and vectorization representation algorithms; The knowledge reasoning unit includes a retrieval query construction subunit, an evidence retrieval subunit, a relevance ranking and evidence annotation subunit, a rule consistency verification subunit, and an explanation and candidate suggestion generation subunit. The specific process is as follows: The retrieval query construction subunit is used to generate retrieval queries using gestational age information, risk factor tags, and key indicator ranges from structured health profile data as retrieval conditions; the retrieval queries are implemented using field concatenation and templated query generation rules. The evidence retrieval subunit is used to retrieve a set of candidate evidence entries from pregnancy risk knowledge graphs, medical guidelines, and clinical evidence bases; the evidence retrieval subunit is implemented using keyword retrieval and vectorized semantic retrieval. The relevance ranking and evidence labeling subunit is used to rank the candidate evidence item set by relevance score to obtain evidence source identifier and evidence level label; the relevance score ranking is implemented by similarity calculation and ranking algorithm; the evidence level labeling is implemented by preset evidence classification rules and rule engine; The rule consistency verification subunit is used to verify the consistency of candidate suggestions and filter out non-compliant items based on contraindications related to gestational age, comorbidities, and applicable conditions. The consistency verification is implemented by using a rule engine to determine the contraindications and applicable conditions. The explanation and candidate suggestion generation subunit receives structured health profile data, risk status data, and evidence retrieval and ranking subunit to obtain a set of candidate evidence items. Under the rule consistency verification and applicable condition relationship, contraindication and constraint relationship in the pregnancy risk knowledge graph, the risk status data is mapped and analyzed to obtain risk element explanation data. Based on the set of verified candidate evidence items, gestational age information, and risk level, a set of candidate suggestions is generated through template matching and a rule engine to obtain intermediate data for suggestion generation. The intermediate data for suggestion generation includes the set of candidate suggestions, evidence citation identifiers, and applicable condition constraints. The emotion computing unit includes an emotion feature extraction subunit, an emotion state calculation subunit, and an interaction strategy selection subunit, as detailed below: The emotion feature extraction subunit is used to receive user interaction text or speech features and obtain emotion feature vectors; the extraction process of the emotion feature vectors is implemented using text emotion feature encoding and speech prosody feature extraction algorithms; The emotion state calculation subunit is used to classify or score the emotion feature vector to obtain emotion level data; the classification or scoring calculation is implemented using an emotion classification model and an emotion scoring model. The interaction strategy selection subunit is used to select an interaction strategy template based on emotion level data and risk level data and output interaction strategy parameter data; the interaction strategy parameter data includes at least prompt tone parameters, information density parameters, interaction frequency parameters, and trigger identifiers for escalation to manual or medical team intervention; the selection of the interaction strategy template is implemented using a rule engine and a state machine.

4. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 1, characterized in that, The risk monitoring and assessment module includes a physiological indicator monitoring unit, a behavioral pattern analysis unit, a psychological state assessment unit, and a risk early warning unit, specifically including the following: The physiological indicator monitoring unit is used to receive raw vital sign data collected from smart devices and examination, test and diet data uploaded by users, and to record timestamps, convert units and mark outliers in the raw vital sign data and calculate physiological indicator monitoring data. The behavior pattern analysis unit is used to receive the physiological indicator monitoring data and user behavior log data, extract and score or grade the diet, exercise and sleep-related features to obtain behavior assessment data. The psychological state assessment unit is used to receive user interaction content and psychological scales, calculate and map the psychological scale scores and emotion levels to obtain psychological state assessment data. The risk warning unit is used to receive structured health profile data obtained from the large language module, and integrate physiological indicator monitoring data, behavioral assessment data and psychological state assessment data for feature aggregation. It calls a preset rule model or machine learning model to calculate a risk score and map it to a risk level. When a preset threshold or rule condition is met, it obtains warning event data and risk status data.

5. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 4, characterized in that, Specifically, it includes the following: The physiological indicator monitoring unit includes a device docking subunit, a data verification subunit, an indicator calculation subunit, and a storage and output subunit, as detailed below: The device interface subunit is used to interface with blood pressure monitors, blood glucose meters, and smart bracelets to collect raw vital sign data; the data collection from blood pressure monitors, blood glucose meters, and smart bracelets is achieved through communication interface adaptation and device protocol parsing. The data verification subunit is used to perform unit unification, timestamp standardization, and outlier marking on the original data to obtain verified data; the unit unification and timestamp standardization are implemented using preset conversion rules and time synchronization rules; the outlier marking is implemented using threshold rules and statistical distribution detection algorithms. The indicator calculation subunit is used to perform interval determination and trend calculation on the verified data to obtain physiological indicator monitoring data; The trend calculation is achieved using a sliding time window statistical and trend fitting algorithm, and the interval determination is achieved using a segmented threshold mapping rule. The storage output subunit receives key physiological indicator feature data obtained from the indicator calculation subunit, encapsulates the physiological indicator feature data in a structured manner, associates it with timestamp identifier, indicator type identifier and data source identifier, and then inputs it into the structured data storage structure. In accordance with the preset data interface specification, the physiological indicator monitoring data is output to the feature aggregation subunit of the risk warning unit in the form of feature fields. The behavior pattern analysis unit includes a behavior log collection subunit, a feature extraction subunit, and a scoring and grading subunit, as detailed below: The behavior log collection subunit is used to receive check-in records, step count synchronization, food photos, and sleep duration records to form behavior log data; the formation of the behavior log data is achieved by encapsulating log schema and normalizing timestamps; The feature extraction subunit is used to extract dietary, exercise, and sleep features from behavioral log data and physiological indicator monitoring data; the extraction of dietary, exercise, and sleep features is achieved using statistical aggregation, rule calculation, and feature engineering algorithms. The scoring and grading subunit is used to calculate compliance scores and risk grading based on preset rules and obtain behavioral assessment data; the compliance scores are implemented using a rule-based scoring model; the risk grading calculation is implemented using threshold mapping rules and regression models or machine learning models; The psychological state assessment unit includes a scale management subunit, an interactive emotion recognition subunit, and a score mapping subunit, as detailed below; The scale management subunit is used to push and receive psychological scales; the psychological scales are implemented using a scale template library and a questionnaire distribution and collection mechanism. The interactive emotion recognition subunit is used to extract emotion features from user interaction content to obtain emotion index data; the emotion feature extraction is implemented using text sentiment feature encoding and speech prosody feature extraction algorithms. The scoring mapping subunit is used to calculate the scores of the psychological scale and fuse them with the emotion index data to map them into psychological levels, thereby obtaining psychological state assessment data. The calculation of the psychological scale scores is implemented using pre-designed scoring rules, weighted fusion, threshold grading mapping rules, and a classification model. The risk warning unit includes a feature aggregation subunit, a risk scoring subunit, a hierarchical mapping subunit, and an event triggering subunit, as detailed below: The feature aggregation subunit is used to aggregate structured health profile data with physiological indicator monitoring data, behavioral assessment data, and psychological state assessment data through time windows to obtain a risk feature vector; the time window aggregation is implemented using a sliding time window resampling and statistical aggregation algorithm; The risk scoring subunit is used to perform weighted fusion calculations on risk feature vectors or to call a machine learning model to obtain a risk score; the weighted fusion calculation adopts a preset weighting rule; the machine learning model uses a regression model and an interpretive machine learning model to implement the risk scoring. The hierarchical mapping subunit is used to map risk scores to risk levels; the risk score mapping is implemented using threshold mapping rules or piecewise functions. The event triggering subunit is used to generate early warning event data and output the risk status data when the key indicator exceeds the threshold, the score jumps, or the continuous abnormal conditions are met. The determination of the key indicator exceeding the threshold, the score jumps, or the continuous abnormal conditions is implemented by a rule engine and a state machine to trigger the event.

6. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 1, characterized in that, The personalized intervention module includes a health education unit, a behavior guidance unit, an emergency guidance unit, and a remote collaboration unit, as detailed below: The health education unit is used to receive gestational age stage information, individual risk element tags, and risk level and warning event type from the structured health profile data as input conditions, and to perform retrieval and matching in the pregnancy risk knowledge graph and organize evidence to obtain age-appropriate, stage-specific pregnancy knowledge content data and obtain knowledge source identification and evidence basis. The behavior guidance unit is used to receive risk status data and structured health profile data as input to obtain behavior guidance data and task list and task parameter data mapped by the gamified interaction module; The emergency guidance unit is used to receive risk level and early warning event data as input, match the disposal path template, and obtain disposal plan data, upgrade trigger condition data, and re-evaluation node data. The remote collaboration unit is used to receive risk summary data and treatment plan data under authorized conditions, encapsulate them into remote collaboration instruction data and push them to the medical team, and at the same time receive feedback from the medical team to form collaboration receipt data.

7. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 6, characterized in that, Specifically as follows: The health education unit includes a search condition construction subunit, an evidence retrieval and ranking subunit, an evidence classification and tracing subunit, and a content organization subunit, as detailed below: The search condition construction subunit is used to construct search conditions based on gestational age information, risk level, warning event type, and risk element tags; the construction of search conditions is achieved by field concatenation and templated query generation rules. The evidence retrieval and ranking subunit is used to retrieve a set of candidate knowledge items from the pregnancy risk knowledge graph, medical guidelines, and clinical evidence base and rank them by relevance. The retrieval is achieved by keyword retrieval and vectorized semantic retrieval, and the relevance ranking is achieved by similarity calculation and ranking algorithms. The evidence grading and tracing subunit is used to generate evidence levels and source identifiers for a set of candidate knowledge items; the source identifier is implemented using a source identifier field mapping and reference chain recording mechanism; the evidence level is implemented using preset evidence grading rules and a rule engine. The content organization subunit is used to generate knowledge content data packages and obtain nurturing knowledge content data based on the sorting results and applicable conditions; the content organization subunit is implemented using a content template library and parameter filling algorithm to generate knowledge content data packages; The behavior guidance unit includes a goal setting subunit, an individualized constraint loading subunit, a scheme generation subunit, and a task parameter generation subunit, as detailed below: The target setting subunit is used to determine behavior management targets based on risk status data; the behavior management targets are implemented using rule mapping and threshold determination. The individualized constraint loading subunit is used to load gestational week contraindications, comorbidity contraindications, and user preferences to form a set of constraint conditions; the set of constraint conditions is implemented using a contraindication rule base and a user preference parameter loading mechanism. The scheme generation subunit is used to obtain behavior guidance data based on the set of behavior management objectives and constraints; the behavior guidance data is implemented using an objective-constraint rule calculation and planning algorithm. The task parameter generation subunit is used to convert the behavior guidance data into a task list and task parameter data and output them to the gamification interaction module; the task parameter data generation is implemented using a task template library matching and parameter calculation algorithm. The emergency guidance unit includes an event triage subunit, a response path matching subunit, an escalation rule generation subunit, and a review node generation subunit, as detailed below: The event triage subunit is used to determine the event category and handling priority based on the warning event type and risk level; the risk level is used to determine the event category using event type mapping rules and priority scoring rules. The disposal path matching subunit is used to match preset disposal path templates and generate disposal plan data; The disposal path template is implemented using a disposal path template library and a rule engine; The upgrade rule generation sub-unit is used to generate upgrade trigger condition data; The upgrade triggering condition data is implemented using threshold rules, continuous anomaly rules, and state machine rules. The review node generation subunit is used to generate review node data and output it to the intervention plan data; the generation of review node data is implemented using a time window rule based on gestational age stage and event category. The remote collaboration unit includes an authorization management subunit, an information encapsulation subunit, a message push subunit, and a receipt processing subunit, as detailed below: The authorization management subunit is used to verify the authorization information of pregnant women and generate an authorization identifier; the authorization identifier is implemented using identity authentication and authorization strategy rules; The information encapsulation subunit is used to encapsulate risk summary data, trigger indicators, risk levels, and disposal plan data into remote collaboration instruction data; the remote collaboration instruction data is implemented using instruction schema and field templates and undergoes necessary desensitization processing. The message push subunit is used to push remote collaboration instruction data to the medical team; the push of remote collaboration instruction data is implemented using interface calls and message queues and supports encrypted transmission; The receipt processing subunit is used to receive feedback from the medical team and generate collaborative receipt data to update the intervention plan status data; the generation of collaborative receipt data is implemented using a receipt state machine and log recording mechanism and triggers the update of the intervention plan status data.

8. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 1, characterized in that, The gamified interaction module includes: a task mapping unit, an achievement system unit, a virtual character unit, a social interaction unit, a progress visualization unit, and a behavior log collection unit. The specific process is as follows: The task mapping unit is used to break down the intervention plan data into a task list according to the preset task template and scenario rules and output the game task parameter data; the game task parameter data includes at least the task type, target threshold, execution frequency, deadline, completion judgment condition and reward trigger condition. The achievement system unit is used to configure pregnancy health management milestones, points, badges, and reward rules based on intervention program data and reward rules, and to obtain incentive feedback data. The virtual character unit is used to generate a personalized pregnant woman image and growth path based on the intervention plan data and structured health profile data, and to obtain character presentation parameters. The social interaction unit is used to support experience sharing and mutual challenges among pregnant and postpartum women and to obtain social interaction content data. The progress visualization unit is used to display the visualization parameters of the changing trends of multi-dimensional pregnancy health data. The behavior log collection unit is used to collect and structure the user's interaction events during task execution, check-in, assessment and social interaction to obtain user interaction behavior data.

9. The pregnancy risk management system for pregnant women based on a large language model and gamification as described in claim 8, characterized in that, Specifically as follows: The task mapping unit includes a scheme parsing subunit, a template matching subunit, a parameter generation subunit, and a task publishing subunit, as detailed below: The intervention plan parsing subunit is used to parse the intervention plan data into a structured set of elements: objectives, frequencies, and constraints; the intervention plan data parsing is implemented using rule parsing and semantic parsing algorithms. The template matching subunit is used to match task types and scene templates from the task template library based on the structured feature set. The task type matching in the task template library is implemented using rule matching and similarity matching algorithms based on label constraints. The parameter generation subunit is used to calculate the target threshold, completion judgment condition, reward trigger condition and deadline based on the target and constraints. The generation of the target threshold, completion judgment condition, reward trigger condition and deadline is achieved by threshold mapping and condition calculation rules; The task publishing subunit is used to output the task list and game task parameter data, and to provide level presentation parameters to the progress visualization unit; the task list is implemented using a task queue and event notification mechanism. The achievement system unit includes a completion calculation subunit, a reward rule engine subunit, and a feedback generation subunit, as detailed below: The completion calculation subunit is used to calculate the task completion rate and continuous achievement status based on user interaction behavior data; the completion calculation subunit is implemented using event counting, threshold determination, and continuous status tracking algorithms; The reward rule engine subunit is used to calculate points, badges, and levels based on the completion rate according to the reward rules and generate reward event data; the reward rules use a rule engine and a state machine to trigger the calculation of points, badges, and levels. The feedback generation subunit is used to convert reward event data into incentive feedback data and output it to the progress visualization unit and the virtual character unit to drive interface feedback and growth presentation; the incentive feedback data is implemented using a feedback template library and parameter filling algorithm. The virtual character unit includes a character parameter generation subunit, a growth path mapping subunit, and a rendering subunit, as detailed below: The role parameter generation subunit is used to generate basic role parameters based on gestational age, risk factor tags, and user preferences in the structured health profile data; the basic role parameters are implemented using profile parameter mapping rules and preference weight fusion algorithms; The growth path mapping subunit is used to map task completion, score level, and stage goal into growth status data; the mapping of task completion, score level, and stage goal into growth status data is implemented by a state machine and a score mapping rule to calculate the growth status. The rendering subunit is used to output character rendering parameters and display character appearance changes, unlocked content, and growth animations in the interface; the character rendering parameters are based on preset resource library and animation template parameters. The social interaction unit includes a content generation and review subunit, a matching and recommendation subunit, and an interactive challenge management subunit, as detailed below: The content generation and review subunit is used to structure and encapsulate experience-sharing texts and challenge content and review them according to preset compliance rules; the structured encapsulation is implemented using content schema and field templates; the compliance rule review is implemented using a rule engine and sensitive information identification algorithm; The matching and recommendation sub-unit is used to match and recommend interactive objects or content based on pregnancy stage, role presentation task type, and risk level; The matching recommendation is implemented using a rule-based matching and similarity calculation recommendation algorithm based on label constraints; The interactive challenge management subunit is used to obtain mutual assistance challenge parameters and output social interaction content data, while recording the interaction process to form user interaction behavior data; the mutual assistance challenge parameters are generated and the process state is updated using a challenge template library and a state machine; The progress visualization unit includes an indicator mapping subunit, a trend calculation subunit, and a graphics rendering subunit, as detailed below: The indicator mapping subunit is used to map structured health profile data, risk level data, and task completion data into a set of visual indicators; the mapping of structured health profile data, risk level data, and task completion data into a set of visual indicators is implemented using an indicator dictionary and rule mapping algorithm; The trend calculation subunit is used to perform time window aggregation and trend calculation on the set of visualization indicators to generate trend feature data; The time window aggregation and trend calculation are implemented using sliding time window statistical aggregation and trend fitting algorithms; The graphics rendering subunit is used to convert the trend feature data into trend charts, progress bars, milestones or level progress and output visualization parameters. The trend feature data is converted into trend charts, progress bars, milestones or level progress using a graphics rendering engine and the visualization parameters are generated based on a preset chart template. The behavior log collection unit includes an event collection subunit, a log structuring subunit, and a reflow update subunit, as detailed below: The event collection subunit is used to collect user interaction events during task execution, check-in, assessment, and social interaction; the interaction events are implemented using client-side event tracking and event listening mechanisms. The log structuring subunit is used to encapsulate interactive events into timestamped user interaction behavior data; the encapsulation of interactive events into timestamped user interaction behavior data is achieved by encapsulating them using an event model and a log schema, and then normalizing the timestamps. The backflow update subunit is used to provide user interaction behavior data to the core engine of the large language module for updating the structured health profile data, and to provide it to the risk monitoring and assessment module and the personalized intervention module for dynamically adjusting the risk status data or intervention plan data; the dynamic adjustment of risk status data or intervention plan data adopts a message queue or interface call mechanism to realize data distribution and update triggering.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the system according to any one of claims 1 to 9.

Images