A weight control method and system for constructing an individualized physiological response map
By constructing personalized physiological response maps and utilizing multimodal physiological data and causal inference models, the problem of high universality and low personalization in existing weight control programs has been solved, achieving precise and effective weight control and forward-looking recommendations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING XINDONG INTERNATIONAL SPORTS CULTURE DEVELOPMENT CO LTD
- Filing Date
- 2026-02-25
- Publication Date
- 2026-05-29
AI Technical Summary
Existing technologies are highly universal but lack personalization in weight regulation, failing to reveal the dynamic causal relationship between intervention and physiological response, resulting in highly varied regulatory effects and the potential for adverse reactions.
A personalized physiological response map is constructed by collecting multimodal physiological data, using pre-trained physiological domain models for vectorization, and combining causal inference models to calculate dynamic causal contribution scores. Based on this, candidate weight regulation schemes are deduced and optimized.
It achieves highly personalized physiological modeling, reveals dynamic causal relationships, possesses forward-looking scheme extrapolation capabilities, significantly improves the accuracy and effectiveness of weight regulation, and avoids adverse reactions.
Smart Images
Figure CN122117446A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a health management method and system based on multimodal physiological data analysis.
[0002] More specifically, the present invention relates to a method and system for constructing a personalized physiological response model using knowledge graphs and causal inference techniques, and for weight regulation based thereon. Background Technology
[0003] With socioeconomic development and changing lifestyles, obesity and related chronic metabolic diseases have become a global public health challenge. Therefore, scientific and effective weight management and regulation have received unprecedented attention. Current weight management solutions on the market, whether traditional dietary advice, exercise plans, or digital health management applications, are mostly based on universal physiological and nutritional knowledge, providing standardized guidance for all users. However, the human body is an extremely complex nonlinear dynamic system, and different individuals exhibit significant differences in their physiological responses to the same interventions (such as diet and exercise). These universal solutions ignore individual differences, leading to vastly different regulatory effects and potentially even adverse reactions. To achieve personalization, some technological solutions attempt to utilize knowledge graphs to organize knowledge in the health field. For example, Chinese patent application CN202111333936.2 discloses a method for constructing a next-generation information technology industry knowledge graph. This method mainly uses representation learning techniques (such as the TransH model) to map entities into low-dimensional vectors and constructs static triplet relationships based on vector similarity. The drawback of this method is that the constructed knowledge graph is static; once the relationships are established, they remain unchanged, and the establishment of these relationships relies primarily on the correlation between entities rather than causality. This method cannot be applied to describe human physiological systems because the human body's response to interventions (such as diet and exercise) is highly dynamic, non-linear, and time-sensitive. Simple correlation analysis cannot reveal the true causal chain between intervention and outcome. Furthermore, Chinese patent application CN202111467162.2 discloses a spatiotemporal artificial intelligence expert system based on knowledge graphs and big data. Although this system combines knowledge graphs and big data search, its knowledge graph construction still relies on expert knowledge or keyword crawling, lacking the ability to deeply mine individualized, time-series data. Its core lies in the question-and-answer model of "knowledge graph + search engine," rather than modeling and deducing the internal physiological mechanisms of an individual. Therefore, this system cannot build a "digital twin" model reflecting the unique physiological response patterns of a specific user, thus failing to provide truly personalized and forward-looking health intervention recommendations. In summary, existing technologies for constructing knowledge graphs either focus on static entity relationships or are limited to general knowledge question answering, failing to effectively address the core challenge in weight management: how to extract dynamic, personalized causal relationships between lifestyle interventions and changes in physiological indicators from an individual's multimodal, temporal physiological data, and based on this, provide precise and effective management solutions. Therefore, there is an urgent need for an intelligent weight management method capable of building a personalized physiological response model for each user and dynamically extrapolating it. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for weight regulation that constructs a personalized physiological response map, aiming to solve the technical problems of existing weight regulation schemes having strong universality, low personalization, and inability to reveal the dynamic causal relationship between intervention and physiological response.
[0005] To achieve the above objectives, this invention provides a method for weight regulation by constructing a personalized physiological response map, comprising: collecting multimodal physiological data of a target user, wherein the multimodal physiological data includes time-series data of physiological indicators, lifestyle intervention data, and static physiological baseline data; vectorizing the multimodal physiological data based on a pre-trained physiological domain model to obtain intervention entity vectors, physiological outcome entity vectors, and time-varying user physiological state vectors; calculating a dynamic causal contribution score between any intervention entity and any physiological outcome entity based on a causal inference model; the dynamic causal contribution score is used to quantify the intensity and timeliness of the causal effect of the intervention entity on the physiological outcome entity under the current user physiological state; constructing and dynamically updating a personalized physiological response map of the user by using the intervention entity and the physiological outcome entity as map nodes and the dynamic causal contribution score as weighted directed edges connecting the nodes; and deducing candidate weight regulation schemes based on the personalized physiological response map to generate and output a target weight regulation scheme with optimal expected regulatory efficacy.
[0006] Preferably, the calculation of the dynamic causal contribution score specifically includes: calculating the contribution score at time point using the following formula. By intervention entity Pointing to physiological outcome entity Dynamic causal contribution score : ,
[0007] in, Indicates the first One intervention entity, Indicates the first A physiological outcome entity; Let be an attention function used to calculate right The intensity of basic attention; This is the query vector generated from the entity vector of the physiological results. and These are the key vector and value vector generated from the intervention entity vector, respectively; It is a physiological state gating unit used to determine the user's current physiological state. Adjust the basic attention intensity, wherein This is the user's physiological state vector. and These are the weight matrix and bias vector of the gated unit, respectively. Use the Sigmoid activation function; Let be a time decay term, where The preset time decay coefficient, From the time of the intervention event to the current time point Time difference; It is a normalization coefficient.
[0008] Preferably, the vectorization process specifically includes: inputting the time-series data of physiological indicators, the dietary log text and exercise records in the lifestyle intervention data, and the gene testing report in the static physiological baseline data into a pre-trained, Transformer-based large language model for the physiological domain; extracting the output of a specific hidden layer of the large language model, which is used as the intervention entity vector, the physiological result entity vector, and the user physiological state vector, respectively. Preferably, the user physiological state vector... It is composed of a sleep vector representing the user's recent sleep quality, a stress vector representing the user's recent psychological stress, and an activity vector representing the user's recent activity intensity.
[0009] Preferably, the lifestyle intervention data includes dietary logs, exercise records, and medication records; the time-series physiological indicator data includes continuous blood glucose, heart rate, and skin conductance data collected through wearable devices; and the static physiological baseline data includes the user's genotype data, basal metabolic rate, and allergen information.
[0010] Preferably, the step of deducing candidate weight control schemes and generating and outputting a target weight control scheme specifically includes: for each candidate weight control scheme... Path traversal is performed on the personalized physiological response map to predict its impact on user weight. Expected change Simultaneously, predict the candidate weight control schemes. Risk of triggering negative physiological responses Through the following weight regulation utility function Calculate the utility score for each candidate solution: ,
[0011] in, and These are preset weighting coefficients, representing the degree of importance attached to weight loss effectiveness and side effect avoidance, respectively. A sigmoid function is used to map the weight loss effect to a range of 0 to 1; the function with the highest utility score is selected. The candidate weight control schemes are used as the target weight control scheme.
[0012] On the other hand, the present invention also provides a weight regulation system for constructing a personalized physiological response map, comprising: a multimodal physiological data acquisition module for acquiring multimodal physiological data of a target user; a physiological entity and state vectorization module for vectorizing the multimodal physiological data based on a pre-trained physiological domain model to obtain intervention entity vectors, physiological result entity vectors, and time-varying user physiological state vectors; a causal relationship quantification module for calculating the dynamic causal contribution score between any intervention entity and any physiological result entity based on a causal inference model; a map construction and update module for constructing and dynamically updating the user's personalized physiological response map by using the intervention entity and the physiological result entity as map nodes and the dynamic causal contribution score as weighted directed edges connecting the nodes; and an intervention strategy recommendation module for deducing candidate weight regulation schemes based on the personalized physiological response map, generating and outputting a target weight regulation scheme with optimal expected regulation efficacy. Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: 1. It achieves highly personalized physiological modeling. By constructing a personalized physiological response map for each user, this invention breaks away from the traditional "one-size-fits-all" control approach. It accurately reflects the unique response patterns of individuals to different interventions (such as specific foods or types of exercise), significantly improving the precision and effectiveness of weight management. 2. It reveals dynamic causal relationships. This invention innovatively introduces a dynamic causal contribution score based on causal inference, which not only quantifies the strength of the relationship between intervention and outcome but also considers the user's current physiological state and time decay effects. This allows the map to evolve dynamically, capturing the nonlinear and time-varying characteristics of physiological responses, far superior to existing static correlation-based analysis methods. 3. It possesses forward-looking scheme deduction and optimization capabilities. Based on the constructed personalized map, this invention can conduct "sandbox deduction" of multiple candidate control schemes, predict their impact on weight and potential negative physiological responses, and perform quantitative evaluation through utility functions to recommend the optimal scheme. This forward-looking decision support capability is not available in existing health management tools. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0014] Figure 1 This is an overall flowchart of a weight regulation method for constructing a personalized physiological response map, provided in an embodiment of the present invention.
[0015] Figure 2 This is a detailed flowchart of the steps for calculating the dynamic causal contribution score in an embodiment of the present invention. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of the present invention. Please refer to... Figure 1 This invention provides a method for weight regulation by constructing a personalized physiological response map, which can be applied to personal computers, servers, mobile terminals, or cloud computing platforms. The method specifically includes the following steps:
[0017] Step S101: Collect multimodal physiological data from the target user. This multimodal physiological data aims to comprehensively characterize the user's physiological state and lifestyle. Specifically, the multimodal physiological data includes, but is not limited to, three categories:
[0018] First, time-series data of physiological indicators, such as continuous blood glucose levels, heart rate, heart rate variability (HRV), and skin conductance response (GSR) data collected through wearable devices (e.g., smartwatches, continuous glucose monitors). This data, recorded in time-series format, reflects real-time fluctuations in the user's physiological indicators. Second, lifestyle intervention data, primarily entered manually by the user or imported through third-party application interfaces. Examples include dietary logs containing food types, portions, and cooking methods; exercise records including exercise type, duration, and intensity; and medication records for prescription drugs or health supplements. Third, static physiological baseline data, which consists of relatively stable or slowly changing individual characteristics. Examples include genotype data related to metabolism and obesity obtained through gene sequencing (e.g., FTO genotype), basal metabolic rate (BMR) measured by specialized equipment, and known allergen information.
[0019] Step S102: Based on the pre-trained physiological domain model, the multimodal physiological data is vectorized to obtain intervention entity vectors, physiological result entity vectors, and time-varying user physiological state vectors.
[0020] In one specific implementation, this step is accomplished using a pre-trained, Transformer-based Large Language Model (LLM) for the physiological domain. This model is pre-trained on massive corpora of medical literature, health information, and nutrition databases, giving it deep knowledge in the physiological domain. The specific processing involves inputting dietary log text (e.g., "100g brown rice and 150g steamed salmon for lunch") and exercise records (e.g., "30 minutes of jogging in the evening") from lifestyle intervention data, as well as gene testing report text from static physiological baseline data, into the LLM. The model identifies and encodes key entities in these inputs (e.g., "brown rice," "jogging," "FTO gene rs9939609 A / T type"). Simultaneously, time-series data of physiological indicators (e.g., blood glucose levels) are also input into the model.
[0021] By extracting the output of specific hidden layers of the model, vectors that can represent these entities in a high-dimensional semantic space can be obtained. For example, the vector of the entity "brown rice" extracted from a dietary log is an intervention entity vector; the vector of the entity "2-hour postprandial blood glucose" extracted from physiological indicator data is a physiological outcome entity vector.
[0022] It is worth noting that the user's physiological state vector It is a time-varying vector used to capture a specific point in time. The macroscopic background conditions that affect the user's physiological response. In this embodiment, This can be constructed by concatenating a sleep vector representing the user's recent (e.g., the past 24 hours) sleep quality, a stress vector representing the user's recent psychological stress, and an activity vector representing the user's recent activity intensity. These sub-vectors can also be obtained by analyzing sleep stage data, heart rate variability data, and step count data collected by wearable devices, and encoded using the aforementioned large language model. Step S103: Based on the causal inference model, calculate the dynamic causal contribution score between any of the intervention entities and any of the physiological outcome entities. Please refer to... Figure 2 This step is one of the core innovations of this invention, aiming to quantify the intensity and timeliness of the causal effect of a certain intervention entity on a certain physiological outcome entity under the current physiological state of the user.
[0023] Specifically, this embodiment uses the following formula to calculate at time point By intervention entity (For example, "coffee consumption") refers to the physiological outcome entity. Dynamic causal contribution score of (e.g., "heart rate") :
[0024] ,
[0025] The following is a detailed explanation of each component of the formula: Indicates the first One intervention entity, Indicates the first A physiological outcome entity. This is an attention function, such as the standard scaled dot product attention mechanism. It is used to compute... right The basic level of concern, i.e., intervention without considering other contexts. With results The basic physiological correlation strength. Among them, For entities resulting from physiological outcomes The query vector generated by the vector. and They are respectively by the intervention entity The vectors generate key and value vectors. These vectors are all generated by the physiological domain model in step S102. This is a physiological state gating unit. This unit is key to achieving personalization and dynamism. It is based on the user's current physiological state vector. (As mentioned above, it consists of vectors of sleep, stress, and activity levels), through a learnable linear transformation (weight matrix) and bias vector ) and Sigmoid activation function This generates a gating value between 0 and 1. This value is used to dynamically adjust the base attention intensity. For example, when the user is sleep-deprived ( (As reflected in the sleep vector), this gating value may amplify the contribution of "caffeine" to "heart rate," and vice versa. This is a time decay term.
[0026] It is used to simulate the timeliness of causal effects. Among them... The preset time decay coefficient, From the time of the intervention event (such as drinking coffee) to the current calculation point. The time difference. As time goes by... As the index increases, the causal contribution of the intervention decreases over time, which is consistent with physiological laws. This is a normalization coefficient used to ensure the scale of the final score is desirable. Step S104: Construct and dynamically update the user's personalized physiological response graph by using the intervention entity and the physiological outcome entity as graph nodes, and the dynamic causal contribution score as weighted directed edges connecting the nodes.
[0027] Specifically, the system maintains a knowledge graph organized by user. The nodes in the graph are various entities extracted in step S102, such as "rice," "running," "blood sugar," and "weight." Directed edges between nodes represent causal relationships, and their weights are the dynamic causal contribution scores calculated in step S103. For example, an edge pointing from "rice" to "blood sugar" has a weight representing the expected impact of rice consumption on blood sugar under the current user state. Since the contribution score is dynamically calculated, the edge weights of this graph are constantly updated with the inflow of new data and the passage of time, thus forming a vivid "digital twin" model that reflects the evolution of the user's physiological state.
[0028] Step S105: Based on the personalized physiological response map, the candidate weight control schemes are deduced, and a target weight control scheme with the optimal expected control effect is generated and output. In a specific implementation, this step includes: First, the system generates a series of candidate weight control schemes. For example, "Option A: Replace dinner with chicken breast salad and add 20 minutes of brisk walking," and "Option B: Reduce lunch by half and add 15 minutes of HIIT training," etc. Secondly, for each candidate option... The system performs path traversal and reasoning based on the user's personalized physiological response profile. For example, for option A, the system would follow the path of "chicken breast salad". … "Weight" and "Brisk Walking" … The study deduced the impact of the proposed solution on user weight by considering multiple paths, including "weight," and by integrating the dynamic causal contribution scores of each edge along these paths. Expected change At the same time, the system will also predict the potential negative physiological responses to this solution, such as "chicken breast salad". … Gastrointestinal discomfort or brisk walking … "Knee pain" is used to calculate a comprehensive risk of negative physiological responses. Then, through the following weight regulation utility function To calculate the overall utility score for each candidate solution: ,
[0029] in, and The preset weighting coefficients can be adjusted by users or health managers according to personal preferences (e.g., whether they value the speed of weight loss more or the comfort of the process more), representing the degree of importance attached to the weight loss effect and the avoidance of side effects, respectively. This is an S-shaped function used to map the weight loss effect (usually negative) to a range of 0 to 1, so that the more weight lost, the higher the score for this item. Finally, the system selects the score with the maximum utility. The proposed weight management schemes are selected as the final target weight management scheme recommended to the user and output in a visual form (such as suggested diets or exercise plans). Accordingly, the present invention also provides a weight management system for constructing a personalized physiological response map, which is used to execute the above method. In one embodiment, the system may include: a multimodal physiological data acquisition module, used to execute step S101, collecting multimodal physiological data of the target user; a physiological entity and state vectorization module, used to execute step S102, vectorizing the data based on a pre-trained physiological domain model; a causal relationship quantification module, used to execute step S103, calculating a dynamic causal contribution score based on a causal inference model, the specific implementation of which may employ the aforementioned formula including an attention mechanism, a physiological state gating unit, and a time decay term; a map construction and update module, used to execute step S104, constructing and dynamically updating the user's personalized physiological response map; and an intervention strategy recommendation module, used to execute step S105, performing scheme deduction based on the map and evaluating it using the aforementioned weight management utility function, ultimately generating and outputting the optimal target weight management scheme.
[0030] Those skilled in the art will understand that the modules in the above system can be implemented using computer program instructions, which can be stored in a computer-readable storage medium, such as a hard disk, flash memory, or optical disk. When the processor executes these instructions, the steps of the method described in this invention are completed. The system can be deployed on a single device or as part of a distributed system.
[0031] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for weight regulation by constructing a personalized physiological response map, characterized in that, include: Collect multimodal physiological data of target users, including time-series data of physiological indicators, lifestyle intervention data, and static physiological baseline data; Based on a pre-trained physiological domain model, the multimodal physiological data is vectorized to obtain intervention entity vectors, physiological outcome entity vectors, and time-varying user physiological state vectors; based on a causal inference model, the dynamic causal contribution score between any intervention entity and any physiological outcome entity is calculated. The dynamic causal contribution score is used to quantify the intensity and timeliness of the causal effect of the intervention entity on the physiological outcome entity under the current physiological state of the user; the intervention entity and the physiological outcome entity are used as graph nodes, and the dynamic causal contribution score is used as a weighted directed edge connecting the nodes to construct and dynamically update the user's personalized physiological response graph; based on the personalized physiological response graph, candidate weight control schemes are deduced, and a target weight control scheme with the optimal expected control effect is generated and output.
2. The method for weight regulation by constructing a personalized physiological response map according to claim 1, characterized in that, The calculation of the dynamic causal contribution score specifically includes: calculating the contribution score at time point using the following formula. By intervention entity Pointing to physiological outcome entity Dynamic causal contribution score : ,in, Indicates the first One intervention entity, Indicates the first A physiological outcome entity; Let be an attention function used to calculate right The intensity of basic attention; This is the query vector generated from the entity vector of the physiological results. and These are the key vector and value vector generated from the intervention entity vector, respectively; It is a physiological state gating unit used to determine the user's current physiological state. Adjust the basic attention intensity, wherein This is the user's physiological state vector. and These are the weight matrix and bias vector of the gated unit, respectively. Use the Sigmoid activation function; Let be a time decay term, where The preset time decay coefficient, From the time of the intervention event to the current time point Time difference; It is a normalization coefficient.
3. The method for weight regulation by constructing a personalized physiological response map according to claim 1, characterized in that, The vectorization process specifically includes: inputting the time-series data of physiological indicators, the dietary log text and exercise records in the lifestyle intervention data, and the gene detection report in the static physiological baseline data into a pre-trained physiological domain large language model based on the Transformer architecture; extracting the output of a specific hidden layer of the large language model as the intervention entity vector, the physiological result entity vector, and the user physiological state vector, respectively.
4. The method for weight regulation by constructing a personalized physiological response map according to claim 2, characterized in that, The user physiological state vector It is composed of a sleep vector representing the user's recent sleep quality, a stress vector representing the user's recent psychological stress, and an activity vector representing the user's recent activity intensity.
5. The method for weight regulation by constructing a personalized physiological response map according to claim 1, characterized in that, The lifestyle intervention data includes dietary logs, exercise records, and medication records; the time-series physiological indicator data includes continuous blood glucose, heart rate, and skin conductance data collected through wearable devices; and the static physiological baseline data includes the user's genotype data, basal metabolic rate, and allergen information.
6. The method for weight regulation by constructing a personalized physiological response map according to claim 1, characterized in that, The process of deducing candidate weight control schemes and generating and outputting a target weight control scheme specifically includes: for each candidate weight control scheme... Path traversal is performed on the personalized physiological response map to predict its impact on user weight. Expected change Simultaneously, predict the candidate weight control schemes. Risk of triggering negative physiological responses Through the following weight regulation utility function Calculate the utility score for each candidate solution: ,in, and These are preset weighting coefficients, representing the degree of importance attached to weight loss effectiveness and side effect avoidance, respectively. A sigmoid function is used to map the weight loss effect to a range of 0 to 1; the function with the highest utility score is selected. The candidate weight control schemes are used as the target weight control scheme.
7. A weight regulation system for constructing personalized physiological response maps, characterized in that, include: A multimodal physiological data acquisition module is used to collect multimodal physiological data of target users, including time-series data of physiological indicators, lifestyle intervention data, and static physiological baseline data. The physiological entity and state vectorization module is used to vectorize the multimodal physiological data based on a pre-trained physiological domain model to obtain intervention entity vectors, physiological result entity vectors, and time-varying user physiological state vectors. The causal relationship quantification module is used to calculate the dynamic causal contribution score between any of the intervention entities and any of the physiological outcome entities based on the causal inference model. The graph construction and update module is used to construct and dynamically update the user's personalized physiological response graph by using the intervention entity and the physiological result entity as graph nodes and the dynamic causal contribution score as weighted directed edges connecting the nodes. The intervention strategy recommendation module is used to deduce candidate weight control schemes based on the personalized physiological response map, and generate and output the target weight control scheme with the best expected control effect.
8. A weight regulation system for constructing a personalized physiological response map according to claim 7, characterized in that, The causal relationship quantification module is specifically used to: calculate the causal relationship at time point using the following formula. By intervention entity Pointing to physiological outcome entity Dynamic causal contribution score : ,in, Indicates the first One intervention entity, Indicates the first A physiological outcome entity; Let it be an attention function; This is the query vector generated from the entity vector of the physiological results. and These are the key vector and value vector generated from the intervention entity vector, respectively; It is a physiological state gating unit, in which This is the user's physiological state vector. and These are the weight matrix and the bias vector, respectively; Let be a time decay term, where The time decay coefficient, For time difference; It is a normalization coefficient.
9. A weight regulation system for constructing a personalized physiological response map according to claim 7, characterized in that, The intervention strategy recommendation module is specifically used for: recommending each candidate weight control plan. Predict its expected change in user weight. and the risk of triggering negative physiological responses Through the following weight regulation utility function Calculate the utility score for each candidate solution: ,in, and The preset weighting coefficients are used; the score with the highest utility is selected. The candidate weight control schemes are used as the target weight control scheme.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a weight regulation method for constructing a personalized physiological response map according to any one of claims 1-6.