Health management data interaction processing method and system based on cloud computing

Abstract

The application relates to the technical field of data processing, and discloses a health management data interaction processing method and system based on cloud computing, which comprises the following steps: acquiring a user ID, health data flow and context data flow; performing dynamic context inference and personalized baseline loading based on the user ID and the context data flow to obtain an inferred context and a personalized baseline; then, performing adaptive explanation model selection and loading based on the inferred context to obtain an adaptive explainer; further, performing contextual health data explanation and risk scoring on the health data flow based on the personalized baseline and the adaptive explainer to obtain a risk score; and finally, generating an operable message based on the risk score. In this way, through dynamic inference of a user context and matching of a personalized baseline and an adaptive explainer, dynamic and personalized analysis of health data is realized, the accuracy and effectiveness of health risk early warning are significantly improved, and more targeted intelligent feedback is provided for users.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a cloud-based health management data interaction processing method and system. Background Technology

[0002] With the widespread adoption of the Internet of Things (IoT) and smart wearable devices, personal health management is undergoing profound changes. Devices such as smartwatches can continuously monitor users' physiological signs, generating massive streams of personal health data, providing opportunities for real-time, accurate health assessments and risk warnings. However, effectively processing and interpreting this complex data is a major challenge, as the meaning of health data is closely related to the specific context in which it is generated (such as exercise or rest). Therefore, the limited computing power of terminal devices alone cannot meet the needs of in-depth, personalized analysis, making the construction of cloud-based backend processing systems an inevitable trend.

[0003] In existing technologies, some health management systems utilize cloud computing to aggregate and store data, conduct preliminary statistical analysis, and provide periodic reports. However, these solutions have significant drawbacks: their data interpretation is static and generic, employing fixed thresholds and analysis rules, ignoring individual differences and contextual changes. For example, the system cannot distinguish between high heart rates caused by strenuous exercise and stress during static work, easily generating numerous invalid alarms due to exceeding static thresholds, thus reducing user trust. Secondly, existing systems lack real-time contextual awareness; their analysis and feedback are decontextualized, failing to provide personalized guidance tailored to the user's current situation. Therefore, existing technologies face significant technical bottlenecks in deeply integrating users' personalized physiological baselines with dynamic contextual information to achieve adaptive, contextualized health risk assessment, urgently requiring an innovative method that can dynamically adjust interpretation models and analysis strategies based on real-time context. Summary of the Invention

[0004] To overcome the shortcomings of existing health management data processing methods, which are static, universal, and severely lack adaptability to individual user and situational differences, this application is proposed. Embodiments of this application propose a cloud-based health management data interaction processing method and system, which can dynamically perceive user context, deeply integrate individual physiological baselines, and perform adaptive analysis.

[0005] According to one aspect of this application, a cloud-based health management data interaction processing method is provided, comprising: obtaining a user ID, a health data stream, and a contextual data stream; performing dynamic contextual inference and personalized baseline loading based on the user ID and contextual data stream to obtain an inferred context and a personalized baseline; selecting and loading an adaptive interpretation model based on the inferred context to obtain an adaptive interpreter; performing contextualized health data interpretation and risk scoring on the health data stream based on the personalized baseline and the adaptive interpreter to obtain a risk score; and generating an actionable message based on the risk score.

[0006] In one possible implementation, dynamic context inference and personalized baseline loading are performed based on user ID and contextual data stream to obtain inferred context and personalized baseline, including: extracting multimodal features within a window from contextual data stream to obtain selected contextual features; inputting the selected contextual features into a pre-trained context inference model to obtain inferred context; and querying and matching the inferred context and user ID as a joint primary key in the user personalized baseline database to obtain personalized baseline.

[0007] In one possible implementation, the selection contextual features include dominant activities, location clustering, current calendar time, and ambient noise level.

[0008] In one possible implementation, the adaptive interpreter is selected and loaded based on the inference context to obtain an adaptive interpreter, including: using the inference context as the primary key, performing an exact search in the context-interpreter ID mapping table to obtain the parsed interpreter ID; using the parsed interpreter ID as the primary key, querying the interpreter definition library to obtain the interpreter definition record; and performing dynamic instantiation and hydration of the interpreter based on the interpreter definition record to obtain the adaptive interpreter.

[0009] In one possible implementation, a risk score is obtained by contextualizing and interpreting health data streams based on a personalized baseline and an adaptive interpreter. This includes: extracting windowed multimodal features from the health data stream to obtain aggregated health features, which include average heart rate, maximum blood oxygen saturation, minimum heart rate, and heart rate variability; dynamically thresholding each health indicator in the aggregated health features based on the personalized baseline and the adaptive interpreter to obtain a contextualized indicator set; performing multi-parallel analysis and anomaly signal extraction on the aggregated health features based on the contextualized indicator set and the adaptive interpreter to obtain an anomaly signal set; and aggregating and synthesizing the risk score from the contextualized indicator set and the anomaly signal set using the adaptive interpreter to obtain the risk score.

[0010] In one possible implementation, based on a personalized baseline and an adaptive interpreter, dynamic threshold instantiation is performed on each health indicator in the aggregated health features to obtain a contextualized indicator set. This includes: dynamically thresholding each health indicator in the aggregated health features using the following formula: ;in, This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. and The baseline value and standard deviation of this health indicator are extracted from the personalized baseline. and The parameters are used to calculate the dynamic threshold of this health indicator extracted from the adaptive interpreter.

[0011] In one possible implementation, a risk score is obtained by aggregating and interpreting the contextualized indicator set and the anomalous signal set based on an adaptive interpreter. This includes: aggregating the risk score from the contextualized indicator set and the anomalous signal set based on the adaptive interpreter using the following formula: ;in, To score the risk, For normalization function, As a context weighting factor, Assigning weights to each indicator. This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. These are the baseline values ​​for each health indicator. These are the current values ​​for each health indicator.

[0012] In one possible implementation, a risk score is obtained by aggregating and interpreting the contextualized indicator set and the abnormal signal set based on an adaptive interpreter. This includes: calculating a personalized deviation for each health indicator in the aggregated health features, whereby the personalized deviation measures the degree of deviation of the current value of the health indicator from its baseline mean in a personalized baseline, and normalizing the deviation using the baseline standard deviation of the indicator; determining a context tolerance threshold, which characterizes the acceptable upper limit of the personalized deviation under the inferred context, wherein the context tolerance threshold is determined based on dynamic threshold calculation parameters extracted from the adaptive interpreter; generating a risk contribution value based on the comparison result of the two when the personalized deviation exceeds the context tolerance threshold; and weighting and aggregating the risk contribution values ​​of each health indicator to generate the risk score.

[0013] According to another aspect of this application, a cloud-based health management data interaction processing system is provided, comprising: a multi-source data access module for acquiring user ID, health data stream, and contextual data stream; a context-aware and baseline loading module for performing dynamic context inference and personalized baseline loading based on user ID and contextual data stream to obtain inferred context and personalized baseline; a context-driven interpreter selection module for selecting and loading an adaptive interpretation model based on the inferred context to obtain an adaptive interpreter; a risk score determination module for performing contextualized health data interpretation and risk scoring on the health data stream based on personalized baseline and adaptive interpreter to obtain a risk score; and an intelligent message generation module for generating actionable messages based on the risk score.

[0014] Compared with existing technologies, the cloud-based health management data interaction processing method and system provided in this application first introduces a dynamic context inference and adaptive interpreter selection mechanism. This allows the system to intelligently adopt the most suitable analysis strategy and judgment threshold based on whether the user is in different states such as exercise, work, or rest, thereby greatly improving the accuracy and effectiveness of health risk assessment and significantly reducing invalid alarms caused by contextual misjudgment. Second, by loading the user's personalized baseline in a specific context, each health data interpretation fully considers individual physiological differences. Finally, this application generates not cold, raw data or general suggestions, but actionable messages that combine the user's current state and individual physical condition, possessing high targeting and immediate guidance value, thus greatly enhancing the user experience and providing users with more targeted intelligent feedback. Attached Figure Description

[0015] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 The illustration shows a schematic flowchart of a cloud-based health management data interaction processing method according to an embodiment of this application.

[0017] Figure 2 The illustration shows a schematic flowchart of step S2 in the cloud-based health management data interaction processing method according to an embodiment of this application.

[0018] Figure 3 The illustration shows a schematic flowchart of step S3 in the cloud-based health management data interaction processing method according to an embodiment of this application.

[0019] Figure 4 The illustration shows a schematic flowchart of step S4 in the cloud-based health management data interaction processing method according to an embodiment of this application.

[0020] Figure 5 The figure shows a schematic block diagram of a cloud-based health management data interaction processing system according to an embodiment of this application. Detailed Implementation

[0021] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0022] Figure 1 The illustration shows a schematic flowchart of a cloud-based health management data interaction processing method and system according to an embodiment of this application. Figure 1 As shown, this application provides a cloud-based health management data interaction processing method, including: S1, obtaining user ID, health data stream, and contextual data stream; S2, performing dynamic contextual inference and personalized baseline loading based on user ID and contextual data stream to obtain inferred context and personalized baseline; S3, selecting and loading an adaptive interpretation model based on the inferred context to obtain an adaptive interpreter; S4, performing contextualized health data interpretation and risk scoring on the health data stream based on the personalized baseline and adaptive interpreter to obtain a risk score; S5, generating an actionable message based on the risk score.

[0023] For example, in step S1, the user ID, health data stream, and contextual data stream are acquired. It should be understood that acquiring the user ID aims to uniquely identify the data subject; without this identifier, real-time data cannot be associated with a specific user's historical health record. Simultaneously, the health data stream is acquired to capture direct indicators reflecting the user's physiological state. However, health data alone is insufficient for accurate judgment; therefore, the contextual data stream must be acquired concurrently. The contextual data stream provides crucial background information for interpreting health data, enabling the differentiation of the rationality of physiological indicator changes in different scenarios. For example, it distinguishes between increased heart rate due to exercise and abnormal heart rate at rest, thereby avoiding analytical biases and misjudgments caused by a lack of contextual awareness in traditional methods.

[0024] Specifically, this is accomplished through a multi-source data access module deployed in the cloud. This module acts as a unified data entry point, responsible for establishing secure data transmission links with multiple user-authorized front-end devices (such as smartphones and smartwatches) and receiving their data in real time. The user ID is a string or number used to uniquely identify a user within the system. It is obtained when a user logs in or a session is established and serves as the associated identifier for all subsequent data packets, ensuring that data is accurately attributed to the corresponding user profile. The health data stream mainly refers to time-series data continuously collected by sensors on the user's smart wearable devices (such as smartwatches) that reflects the user's physiological state. This data is first aggregated to the user's smartphone or other gateway devices via wireless communication protocols (such as Bluetooth), and then uploaded by the gateway devices to the multi-source data access module in the cloud. The contextual data stream refers to multimodal information from various sensors on the user's smartphone or other smart terminal devices that describes the user's current environment, activity status, and social background. For example, it obtains the user's geographical location information through the phone's GPS module, the user's activity status (such as sitting, walking, running) through the accelerometer, the user's schedule by reading the phone's calendar, and the ambient noise level through the microphone.

[0025] For example, in step S2, dynamic context inference and personalized baseline loading are performed based on the user ID and contextual data stream to obtain the inferred context and personalized baseline. It should be understood that the interpretation of health data is highly context-dependent and individual-specific. Without considering the user's current specific context, fluctuations in any health indicator cannot be accurately characterized. For example, without contextual information, it is impossible to distinguish whether a sudden increase in heart rate is a normal physiological response caused by physical exertion or a potential health risk caused by mental stress. Therefore, dynamic context inference is a prerequisite for adaptive analysis; it transforms contextual data into a machine-understandable inferred context, providing a basis for subsequent selection of the correct analysis model. Furthermore, simply knowing the context is insufficient for accurate assessment because physiological baseline levels vary significantly among individuals. By combining the user ID with a personalized baseline for that specific context, this application establishes a dynamically changing normal range reference system specific to that user, thereby ensuring that subsequent risk assessments are based on the user's own historical data, rather than universal statistical standards, thus achieving personalized health management.

[0026] In one embodiment, such as Figure 2As shown, dynamic context inference and personalized baseline loading based on user ID and context data stream are performed to obtain inferred context and personalized baseline, including: S21, extracting multimodal features within a window from the context data stream to obtain selected context features; S22, inputting the selected context features into a pre-trained context inference model to obtain inferred context; S23, using the inferred context and user ID as a joint primary key to perform query matching in the user personalized baseline database to obtain personalized baseline.

[0027] Specifically, firstly, the raw contextual data stream acquired in the previous step is subjected to in-window multimodal feature extraction to obtain selected contextual features. This process aims to organize and refine continuously changing, multi-source raw sensor data (such as GPS coordinates, accelerometer readings, etc.) to calculate key features that can stably represent the user's state within a preset time window. In one embodiment, the selected contextual features include dominant activities, location clustering, current calendar time, and environmental noise level. For example, for location data, GPS coordinate sequences are collected within a time window (e.g., 5 minutes), and clustering analysis such as DBSCAN is performed with the user's historical trajectory data to map them to semantic location labels such as office and home; for activity data, the system processes the accelerometer and gyroscope data within the window, calculates its mean, variance, frequency domain energy, and other statistical features, and then inputs these features into a pre-trained activity recognition classifier (such as a support vector machine) to output dominant activity labels such as sitting and walking. These extracted features together constitute a multi-dimensional contextual feature vector.

[0028] Secondly, the extracted selection context features are input into a pre-trained context inference model to obtain the inferred context, such as a random forest classifier model. This model consists of multiple decision trees. During the training phase, it learns the mapping relationship between a large number of labeled context feature vectors and the final context (such as high-intensity office work). During the inference phase, the input feature vector is judged on each decision tree in the forest, and finally, a voting mechanism is used to output the inferred context with the highest probability and clear semantics.

[0029] Finally, using the inferred context and the user's unique identifier ID as a composite primary key, a query and match is performed in the pre-built user personalized baseline database to obtain a personalized baseline that perfectly corresponds to the current user and current context. Here, the construction of the user personalized baseline database is an offline or periodic process. The system processes the user's long-term historical health data and contextual data, grouping them by user ID and context. For each user's data subset under each context, statistical parameters of various physiological indicators (such as heart rate and blood oxygen saturation), mainly the mean and standard deviation, are calculated. The database table structure uses the combination of user ID and context as a composite primary key, storing the corresponding physiological indicator mean and standard deviation. During real-time loading, the current user ID and the newly obtained inferred context are used as the composite primary key for querying and matching in this database.

[0030] For example, in step S3, an adaptive interpreter is obtained by selecting and loading an adaptive interpreter based on the inferred context. It should be understood that the scientific interpretation of health data is not static; its inherent analytical logic, judgment criteria, and the weights of various physiological indicators all dynamically change with the user's context. For example, in a strenuous exercise context, a significant increase in heart rate is a normal physiological phenomenon with a low risk weight; however, in a quiet office context, the same heart rate level may indicate a higher health risk, and its weight should be increased accordingly. Using a fixed, static interpreter model to process data in all contexts would inevitably lead to numerous misjudgments and omissions, making accurate health management impossible. Therefore, this application, based on the known inferred context, precisely selects and loads an interpreter model specifically designed for that context from a pre-set model library, thereby generating an adaptive interpreter containing specific rules, parameters, and algorithms, providing a unique and accurate analysis engine for subsequent contextualized risk scoring.

[0031] In one embodiment, such as Figure 3 As shown, the adaptive interpreter is selected and loaded based on the inference context to obtain an adaptive interpreter, including: S31, using the inference context as the primary key, performing an exact search in the context-interpreter ID mapping table to obtain the parsed interpreter ID; S32, using the parsed interpreter ID as the primary key, querying the interpreter definition library to obtain the interpreter definition record; S33, performing dynamic instantiation and hydration of the interpreter based on the interpreter definition record to obtain the adaptive interpreter.

[0032] Specifically, firstly, using the inferred context as the primary key, a precise lookup is performed in a pre-configured context-interpreter ID mapping table to obtain a resolved interpreter ID. Specifically, the context-interpreter ID mapping table is a key-value data structure pre-built and maintained by domain experts and system engineers. Its construction process involves using all standardized context labels that the context inference model might output as keys, and assigning each context label a unique string or code within the system as the value (i.e., the interpreter ID). This establishes a one-to-one correspondence between human-understandable context descriptions (such as high-intensity office work) and the IDs (such as a specific string or code) used internally by the system to uniquely identify the interpreter program.

[0033] Next, using the parsed interpreter ID as the primary key, a query is performed in the interpreter definition repository to retrieve the interpreter definition record corresponding to that ID. This interpreter definition repository is a database or file repository that stores detailed configuration information for all available interpreters. Its construction process involves developers and health management experts jointly defining all the parameters and rules required for each interpreter, storing them in the database or configuration file with the interpreter ID as the primary key. The interpreter definition record contains all the blueprint information needed to build a specific interpreter, such as the weights of specific physiological indicators during risk calculations, the calculation parameters for dynamic thresholds, and the algorithm models that need to be invoked.

[0034] Finally, based on the retrieved interpreter definition record, the interpreter is dynamically instantiated and hydrated in the server memory to obtain the final usable adaptive interpreter. In software engineering, instantiation and hydration refer to creating a concrete, executable object instance based on the definition record template, and filling the corresponding attributes of the instance with the parameter values ​​stored in the record, making it a fully functional, configurable, and ready-to-use analysis tool.

[0035] For example, in step S4, based on a personalized baseline and an adaptive interpreter, the health data stream is contextualized for health data interpretation and risk scoring to obtain a risk score. That is, all prepared contextual information (context, personal baseline, analysis model) is deeply integrated and calculated with real-time physiological data to obtain a value that can accurately and dynamically reflect the user's current health risk level.

[0036] In one embodiment, such as Figure 4As shown, based on a personalized baseline and an adaptive interpreter, contextualized health data interpretation and risk scoring are performed on the health data stream to obtain a risk score. This includes: S41, windowed multimodal feature extraction of the health data stream to obtain aggregated health features, which include average heart rate, maximum blood oxygen saturation, minimum heart rate, and heart rate variability; S42, dynamic threshold instantiation of each health indicator in the aggregated health features based on the personalized baseline and the adaptive interpreter to obtain a contextualized indicator set; S43, multi-parallel analysis and abnormal signal extraction of the aggregated health features based on the contextualized indicator set and the adaptive interpreter to obtain an abnormal signal set; and S44, risk score aggregation and interpretation results synthesis of the contextualized indicator set and the abnormal signal set based on the adaptive interpreter to obtain a risk score.

[0037] Specifically, firstly, windowed multimodal feature extraction is performed on the raw health data stream to obtain aggregated health features. This operation aims to calculate and refine the continuous, potentially noisy, raw sensor data stream within a time window, transforming it into a set of more stable and representative statistical indicators.

[0038] Next, based on the personalized baseline and adaptive interpreter, dynamic threshold instantiation is performed on each health indicator in the aggregated health features to obtain a contextualized indicator set. This contextualized indicator set includes a set of personalized alert thresholds dynamically generated for the current user and the current context. Specifically, the dynamic threshold instantiation of each health indicator in the aggregated health features is performed using the following formula: ;in, This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. and The baseline value and standard deviation of this health indicator are extracted from the personalized baseline. and The parameters are used to calculate the dynamic threshold of this health indicator extracted from the adaptive interpreter.

[0039] Then, based on this contextualized indicator set and adaptive interpreter, the aggregated health features are subjected to multi-parallel analysis and abnormal signal extraction to obtain an abnormal signal set. This means that the system compares the current value of each aggregated health feature with its corresponding dynamic alarm threshold. Once the current value exceeds the threshold range, a specific abnormal signal (such as high heart rate) is generated. All these signals together constitute the abnormal signal set.

[0040] Finally, based on the adaptive interpreter, the risk score aggregation and interpretation results of the contextualized indicator set and the abnormal signal set are synthesized to obtain the final risk score. Specifically, based on the adaptive interpreter, the risk score aggregation of the contextualized indicator set and the abnormal signal set is performed using the following formula: ;in, To score the risk, This is a normalization function used to calculate the current value of each indicator. Relative to its baseline value and dynamic alarm thresholds The degree of deviation, As a context weighting factor, The weights for each indicator are obtained from the adaptive interpreter, representing the importance of each indicator in the current context. This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. These are the baseline values ​​for each health indicator. The values ​​represent the current values ​​of various health indicators. Through weighted summation and normalization, the multidimensional abnormal signals are integrated into a single, standardized risk score.

[0041] Preferably, the calculation process of the risk score can be broken down into more statistically significant steps, first measuring the statistical significance of the current value relative to the user's personal baseline, and then using context to adjust the tolerability of this significance.

[0042] In a preferred embodiment, a risk score is obtained by aggregating and interpreting the contextualized indicator set and the abnormal signal set based on an adaptive interpreter. This includes: calculating a personalized deviation for each health indicator in the aggregated health features, whereby the personalized deviation measures the degree of deviation of the current value of the health indicator from its baseline mean in a personalized baseline, and normalizing the deviation using the baseline standard deviation of the indicator; determining a context tolerance threshold, whereby the context tolerance threshold characterizes the acceptable upper limit of the personalized deviation under the inferred context, wherein the context tolerance threshold is determined based on dynamic threshold calculation parameters extracted from the adaptive interpreter; generating a risk contribution value based on the comparison result of the two when the personalized deviation exceeds the context tolerance threshold; and weighting and aggregating the risk contribution values ​​of each health indicator to generate the risk score.

[0043] Specifically, firstly, for each health indicator in the aggregated health features, the personalized deviation score (PDS) is calculated, which measures the current observation value. Deviation from user's personal baseline The degree, and using standard deviation Normalization is performed, that is: ;in As the standard deviation of a personalized baseline in a specific context, if If the value is zero or extremely small, a preset minimum fluctuation value is used to prevent the denominator from being zero.

[0044] Then, the context tolerance threshold CTT is calculated, which is the upper limit of tolerance for PDS, determined by the contextualization parameter. and The common definition represents what level of standardization deviation is acceptable in the current context, namely: That is to say, It is directly used as a multiple of the standard deviation, while the offset is... quilt After normalization and Add them on the same scale.

[0045] Finally, calculate the final risk contribution. This involves comparing a user's personalized deviation score with the context tolerance threshold for that situation to obtain a standardized risk metric, namely: ;in, It is an activation function used to map ratios to the risk space, such as a variant of the ReLU activation function: This means that only when Exceed Risk contribution is only calculated when the ratio is greater than 1.

[0046] Similarly, the total risk score is aggregated as follows: In other words, the calculation of the total risk score has been changed from comparing absolute values ​​to comparing statistical significance; that is, in comparing... and These two absolute differences may ignore the physiological fluctuations of different users. For example, for user A with a baseline heart rate of 60 bpm and a standard deviation of 2 bpm, a heart rate of 70 bpm ( =10) is a huge deviation of 5 standard deviations. For user B, whose baseline is 60 bpm and standard deviation is 5 bpm, a heart rate of 70 bpm ( =10) is only a deviation of 2 standard deviations, which makes it impossible to distinguish the severity of the two situations. In contrast, PDS transforms the deviation of all indicators to the scale of a standard normal distribution, making the deviation of different indicators and different users comparable. For example, PDS=2 always means that there has been a deviation of 2 standard deviations, which is a metric with universal statistical significance.

[0047] Furthermore, the context influences the threshold. When it works, Make The effects of this are offset, resulting in situational conditioning using a window size independent of baseline levels, which is physiologically imprecise. In contrast, by changing the role of the situation to define the tolerance to statistical significance (PDS) (CTT), for example, the CTT for a strenuous exercise situation might be set to 5.0, meaning the system can tolerate heart rate deviations of up to 5 standard deviations, while the CTT for a sleep situation might be only 1.5, making the system very sensitive to even small statistical anomalies. This design is closer to real-world medical and physiological conditions.

[0048] Moreover, the physical meaning of the core ratio PDS / CTT here is clear: PDS / CTT < 1 indicates a deviation from the situation tolerance range, with a risk of 0 (after...). (After processing); PDS / CTT=1 indicates that the deviation is exactly at the boundary of the situation tolerance; while PDS / CTT>1 indicates that the deviation exceeds the range of the situation tolerance, and the greater the deviation, the greater the risk.

[0049] In general, while taking into account the average user level ( ) and volatility level ( In situations where physiological indicators are stable, the system can provide more sensitive monitoring while avoiding excessive false alarms for users with large fluctuations in physiological indicators, thus improving personalized accuracy. Furthermore, the system's adaptation to different situations is no longer simply adjusting absolute thresholds, but rather adjusting its tolerance to statistical biases. This makes the model more robust and reasonable in the face of various complex situations, enhancing its situational adaptability. For example, it can correctly determine that a heart rate of 150 bpm while running is normal (PDS may be high, but CTT is even higher), while a heart rate of 115 bpm while sitting is abnormal (PDS is not high, but CTT is very low).

[0050] For example, in step S5, an actionable message is generated based on the risk score. It should be understood that an isolated risk score, no matter how precise its calculation, cannot realize its value if the user cannot correctly understand it and take appropriate action. Therefore, generating an actionable message based on the risk score translates the complex backend calculation results into frontend information that the user can directly perceive and respond to, thus achieving a key link in realizing the closed loop from data insight to behavioral guidance and completing the goal of proactive health intervention.

[0051] Specifically, generating actionable messages based on risk scores goes beyond simply mapping risk scores to preset warning levels; it involves a more refined content synthesis process. It receives the risk score generated in the previous step, but more importantly, it also receives the underlying causes of that score—the set of abnormal signals. Internally, the module determines the urgency and tone of the message based on the risk score and selects an appropriate message template based on the specific content of the abnormal signals (e.g., high heart rate or low blood oxygen saturation). Once a template is selected, a series of dynamic information is populated into it, including the user's current situation, specific physiological indicator readings, and dynamically calculated personalized thresholds for that situation, thus making the message content highly personalized and contextualized. Furthermore, based on the combination of situation and abnormal signals, it matches and generates the most targeted health advice from a preset knowledge base. This preset health knowledge base is a structured database, pre-built and entered by health management experts, medical experts, and system engineers. This database is not a simple text list but stores a large number of health advice entries in a rule-based format. Each entry is associated with one or more specific trigger conditions. Specifically, the database table structure typically uses a unique suggestion ID as the primary key and includes key fields for matching, such as context identifiers, abnormal signal combination patterns, and content fields for generating messages, such as cause explanation templates and health suggestion templates. The context identifier directly corresponds to the standardized context that the context inference model might output (e.g., "high-intensity office work"); the abnormal signal combination pattern is a unique code or string describing one or more abnormal signal combinations. For example, the signal combination "high heart rate" and "low heart rate variability" is encoded as "HR_HIGH&HRV_LOW".

[0052] Once the intelligent message generation module receives the inferred context and the set of abnormal signals, its matching process begins. The module first normalizes the received set of abnormal signals (e.g., containing two signals: high heart rate and low heart rate variability) to generate a query key that matches the format of abnormal signal combination patterns in the knowledge base. Then, using the current context identifier and the generated abnormal signal combination pattern query key as joint query conditions, it performs a precise match search in the health knowledge base.

[0053] The goal of the query is to find a suggestion entry that perfectly matches the user's current situation. If an exact match is found, the explanation template and health advice template are extracted from that entry. These two templates are text strings containing placeholders, such as, "Under [situation description], your [metric name] remains high, which may be related to [presumed cause]" and "It is recommended that you try [specific action] to help your body return to a stable state."

[0054] After obtaining the template, the final dynamic content synthesis stage begins. It utilizes all contextual information from the current session to dynamically populate placeholders in the template. For example, it replaces "[Context Description]" with "During Office Work," "[Indicator Name]" with "Heart Rate," and also fills in specific physiological values ​​(such as current values ​​and personalized thresholds), thereby generating a complete, highly personalized health recommendation with clear explanations and actionable guidance. If no entry in the knowledge base can be found that perfectly matches the current combination of abnormal signals, a downgraded matching strategy is executed. For example, it attempts to match only based on the most significant abnormal signal or only based on the current context to ensure that the most relevant and valuable health guidance is provided to the user in any situation.

[0055] In summary, the cloud-based health management data interaction processing method provided in this application first introduces a dynamic contextual inference and adaptive interpreter selection mechanism. This allows it to intelligently adopt the most suitable analysis strategy and judgment threshold based on whether the user is in different states such as exercise, work, or rest, thereby greatly improving the accuracy and effectiveness of health risk assessment and significantly reducing invalid alarms caused by contextual misjudgment. Second, by loading the user's personalized baseline in a specific context, each health data interpretation fully considers individual physiological differences. Finally, this application generates not cold, raw data or general suggestions, but actionable messages that combine the user's current state and individual physical condition, possessing highly targeted and immediate guidance value, thus greatly enhancing the user experience and providing users with more targeted intelligent feedback.

[0056] This application also provides a cloud-based health management data interaction processing system for executing the aforementioned cloud-based health management data interaction processing method, such as... Figure 5 As shown, the cloud-based health management data interaction processing system 500 includes: a multi-source data access module 510 for acquiring user ID, health data stream, and contextual data stream; a context-aware and baseline loading module 520 for performing dynamic context inference and personalized baseline loading based on user ID and contextual data stream to obtain inferred context and personalized baseline; a context-driven interpreter selection module 530 for selecting and loading an adaptive interpretation model based on the inferred context to obtain an adaptive interpreter; a risk score determination module 540 for performing contextualized health data interpretation and risk scoring on the health data stream based on personalized baseline and adaptive interpreter to obtain a risk score; and an intelligent message generation module 550 for generating actionable messages based on the risk score.

[0057] This application also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-mentioned related method steps to implement the cloud-based health management data interaction processing method provided in the above embodiments.

[0058] This application also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to realize the cloud-based health management data interaction processing method provided in the above embodiments.

[0059] In this application, the system, computer-readable storage medium, or computer program product provided in the embodiments are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0060] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments.

[0061] The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous. The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A cloud-based health management data interaction processing method, characterized in that, include: Obtain user ID, health data stream, and contextual data stream; Dynamic context inference and personalized baseline loading are performed based on user ID and contextual data stream to obtain inferred context and personalized baseline; An adaptive interpreter is obtained by selecting and loading an adaptive interpretation model based on the inference context. Based on a personalized baseline and an adaptive interpreter, contextualized health data interpretation and risk scoring are performed on the health data stream to obtain a risk score; Based on risk scores, actionable messages are generated.

2. The cloud-based health management data interaction processing method according to claim 1, characterized in that, Dynamic context inference and personalized baseline loading are performed based on user ID and contextual data stream to obtain inferred context and personalized baseline, including: Multimodal feature extraction within a window is performed on the contextual data stream to obtain the selection contextual features; Selected context features are input into a pre-trained context inference model to obtain the inferred context; The inferred context and user ID are used as a composite primary key to perform a query and match in the user personalization baseline database to obtain the personalization baseline.

3. The cloud-based health management data interaction processing method according to claim 2, characterized in that, The selected contextual features include dominant activities, location clustering, current calendar time, and ambient noise level.

4. The cloud-based health management data interaction processing method according to claim 1, characterized in that, Adaptive interpreter selection and loading are performed based on the inference context to obtain an adaptive interpreter, including: Using the inferred context as the primary key, a precise lookup is performed in the context-interpreter ID mapping table to obtain the parsed interpreter ID; Using the parsed interpreter ID as the primary key, a query is performed in the interpreter definition library to obtain the interpreter definition record; The adaptive interpreter is obtained by dynamically instantiating and hydrating the interpreter based on the interpreter definition record.

5. The cloud-based health management data interaction processing method according to claim 1, characterized in that, Based on a personalized baseline and adaptive interpreter, contextualized health data interpretation and risk scoring are performed on the health data stream to obtain a risk score, including: Windowed multimodal feature extraction is performed on the health data stream to obtain aggregated health features, which include average heart rate, maximum blood oxygen saturation, minimum heart rate, and heart rate variability. Based on a personalized baseline and an adaptive interpreter, dynamic threshold instantiation is performed on each health indicator in the aggregated health features to obtain a contextualized indicator set. Based on a contextualized index set and an adaptive interpreter, multi-parallel analysis and abnormal signal extraction are performed on aggregated health features to obtain an abnormal signal set; Based on an adaptive interpreter, risk scores are obtained by aggregating and synthesizing the risk scores from the contextualized indicator set and the abnormal signal set.

6. The cloud-based health management data interaction processing method according to claim 5, characterized in that, Based on a personalized baseline and an adaptive interpreter, dynamic threshold instantiation is performed on each health indicator in the aggregated health features to obtain a contextualized indicator set. This includes: dynamically instantiating the thresholds of each health indicator in the aggregated health features using the following formula: ;in, This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. and The baseline value and standard deviation of this health indicator are extracted from the personalized baseline. and The parameters are used to calculate the dynamic threshold of this health indicator extracted from the adaptive interpreter.

7. The cloud-based health management data interaction processing method according to claim 5, characterized in that, Based on an adaptive interpreter, risk scores are aggregated and interpreted from a set of contextualized indicators and a set of anomalous signals to obtain a risk score. This includes: Based on the adaptive interpreter, risk scores are aggregated from the set of contextualized indicators and the set of anomalous signals using the following formula: ;in, To score risk, For normalization function, As a context weighting factor, Assigning weights to each indicator. This sets dynamic alarm thresholds for various health indicators within a contextualized indicator set. These are the baseline values ​​for each health indicator. These are the current values ​​for each health indicator.

8. The cloud-based health management data interaction processing method according to claim 5, characterized in that, Based on an adaptive interpreter, risk scores are aggregated and interpreted from contextualized indicator sets and anomalous signal sets to obtain a risk score, including: For each health indicator in the aggregated health features, a personalized deviation is calculated. The personalized deviation measures the degree of deviation of the current value of the health indicator from its baseline mean in the personalized baseline, and the deviation is normalized using the baseline standard deviation of the indicator. A context tolerance threshold is determined, which is used to characterize the acceptable upper limit of the degree of personalization deviation under the inference context, wherein the context tolerance threshold is determined based on dynamic threshold calculation parameters extracted from the adaptive interpreter; When the personalization deviation exceeds the situation tolerance threshold, a risk contribution value is generated based on the comparison result of the two. The risk contribution values ​​of each health indicator are weighted and aggregated to generate the risk score.

9. A cloud-based health management data interaction processing system, characterized in that, include: The multi-source data access module is used to acquire user ID, health data stream, and contextual data stream; The context-aware and baseline loading module is used to perform dynamic context inference and personalized baseline loading based on user ID and context data stream to obtain inferred context and personalized baseline; The context-driven interpreter selection module is used to select and load an adaptive interpreter model based on the inference context to obtain an adaptive interpreter. The risk scoring determination module is used to perform contextualized health data interpretation and risk scoring on health data streams based on personalized baselines and adaptive interpreters to obtain risk scores. The intelligent message generation module is used to generate actionable messages based on risk scores.

Images