An AI large model-based health information management method and system

Abstract

The application relates to the technical field of AI large models, and discloses a health information management method and system based on an AI large model, which comprises the following steps: performing feature fusion coding on health screening information collected by a health screening device to obtain a fusion feature vector; performing health portrait coding through an AI large model to obtain a user health portrait vector and perform clustering and typing to obtain a crowd typing category; screening candidate health products through a conditioning product matching score model based on the user health portrait vector and the crowd typing category to obtain a health product recommendation list; and generating a physiotherapy conditioning scheme configuration package based on the user health portrait vector and issuing the physiotherapy conditioning scheme configuration package to surrounding physiotherapy equipment to complete remote presetting. The application solves the problem that traditional Chinese medicine diagnosis data and western medicine physiological indexes cannot be jointly modeled in the prior art, and makes the health portrait capable of comprehensively reflecting the comprehensive health status of a user.

Description

Technical Field

[0001] This invention relates to the field of AI large model technology, and in particular to a health information management method and system based on AI large model. Background Technology

[0002] With the accelerating arrival of an aging society, residents' demand for health management continues to grow, and primary healthcare service institutions such as community pharmacies and wellness centers are expected to assume the important responsibility of serving as daily health advisors for residents. However, in the existing health management system, users' health information is stored in various hospital systems in a scattered manner, there is a lack of systematic service connection channels between pharmacists and residents, and traditional Chinese medicine diagnostic data and Western medicine psychological indicators have long been in a state of separation, making it impossible to achieve cross-modal comprehensive health assessment.

[0003] In existing technologies, health management systems rely solely on single-modal physiological data input. They cannot jointly model unstructured diagnostic data from Traditional Chinese Medicine (TCM), such as hand and facial images, with numerical physiological indicators from Western medicine, such as blood pressure and blood oxygen levels. This results in user health profiles that are limited in dimension and lack precision, making it difficult to support personalized health product recommendations and treatment plan formulation. Furthermore, current health product recommendations lack the technical means to quantitatively model the relationship between user health status and product symptoms, and the recommendation results are not based on individual user health profiles. Treatment plans for physiotherapy equipment rely entirely on manual configuration and cannot be automatically generated and remotely distributed based on user health status, leading to high operating costs and hindering large-scale replication and promotion. Summary of the Invention

[0004] This invention provides a health information management method and system based on AI large model. This invention solves the problem that traditional Chinese medicine diagnostic data and Western medicine psychological indicators cannot be jointly modeled in the prior art, so that the health profile can comprehensively reflect the user's overall health status.

[0005] In a first aspect, the present invention provides a health information management method based on an AI big data model, the health information management method based on an AI big data model comprising: The health screening information collected by the health screening equipment is fused and encoded to obtain a fused feature vector. The fused feature vector is input into the AI ​​large model for health profile encoding to obtain the user health profile vector, and the user health profile vector is clustered and classified to obtain the population classification category. Based on the user health profile vector and the population classification, a health product recommendation list is obtained by screening candidate health products through a conditioning product matching and scoring model. Based on the user's health profile vector, a physiotherapy treatment plan configuration package is generated and distributed to surrounding physiotherapy equipment to complete remote preset.

[0006] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of performing feature fusion encoding on the health screening information collected by the health screening device to obtain a fused feature vector includes: When the health screening device includes a hand diagnostic instrument, a face diagnostic instrument, and a body index detection device, image features are extracted from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument to obtain image feature vectors; the physiological indicators collected by the body index detection device are standardized to obtain physiological indicator vectors; the image feature vectors and the physiological indicator vectors are fused and spliced ​​together by cross-modal attention to obtain fused feature vectors. When the health screening device includes only a hand diagnostic instrument and a face diagnostic instrument, image features are extracted from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument to obtain image feature vectors; the image feature vectors are then used as fusion feature vectors.

[0007] In conjunction with the first aspect, in the second implementation of the first aspect of the present invention, the step of inputting the fused feature vector into an AI large model for health profile encoding to obtain a user health profile vector, and then performing clustering and classification on the user health profile vector to obtain a population classification category, includes: The fused feature vector is input into the AI ​​large model for health status semantic encoding to obtain the user health profile vector; Based on the user health profile vector, the Gaussian mixture model's component weights, mean vector, and covariance matrix are iteratively solved using the expectation-maximization algorithm, and the user is assigned to a population category, thus obtaining the population classification category.

[0008] In conjunction with the first aspect, a third implementation of the first aspect of the present invention further includes: Based on the user's testing location information, a set of candidate pharmacists matching the user's testing location is selected from the registered pharmacists. Each pharmacist in the candidate pharmacist set is scored and calculated according to the pharmacist comprehensive scoring model to obtain the comprehensive score of each candidate pharmacist. The comprehensive score ranking results are pushed to the user's front end, and the user can select and bind a pharmacist as a family health consultant. The user's health profile vector and subsequent health monitoring updates are continuously synchronized to the workbench of the family health consultant pharmacist. The family health consultant pharmacist then provides online health consultation services to the user based on the user's health profile vector. The consultation service data of the family health consultant pharmacist is fed back to the pharmacist comprehensive scoring model to complete the iterative update of the pharmacist's score.

[0009] In conjunction with the first aspect, in the fourth implementation of the first aspect of the present invention, the step of iteratively solving the weights, mean vector, and covariance matrix of each component of the Gaussian mixture model based on the user health profile vector using the expectation-maximization algorithm and assigning the user to a population category to obtain the population classification category includes: Based on the weights of each component, the mean vector, and the covariance matrix, the first posterior probability of each component in the Gaussian mixture model to which the user health profile vector belongs is calculated. The weights, mean vectors, and covariance matrices of each component of the Gaussian mixture model are updated in a weighted manner based on the first posterior probability until the log-likelihood function converges, thus obtaining the parameters of the Gaussian mixture model. Based on the parameters of the Gaussian mixture model, the second posterior probability of each component to which the user health profile vector belongs is calculated, and the population category corresponding to the component with the largest second posterior probability is taken as the user's population classification category.

[0010] In conjunction with the first aspect, in the fifth implementation of the first aspect of the present invention, the step of filtering candidate health products based on the user health profile vector and the population classification using a conditioning product matching scoring model to obtain a recommended list of health products includes: Based on the user health profile vector, a weighted score is calculated on the compatibility between each candidate health product and the user's current symptoms using a product matching scoring model, resulting in a compatibility score for each candidate health product. Based on the aforementioned population classification, user preferences are inferred from the historical interaction data of similar populations through collaborative filtering matrix decomposition to obtain user preference scores. The user preference scores are then combined with the conditioning suitability scores for weighted filtering to obtain a list of recommended health products.

[0011] In conjunction with the first aspect, in the sixth implementation of the first aspect of the present invention, the step of inferring user preferences from historical interaction data of similar groups based on the population classification and obtaining a user preference score through collaborative filtering matrix decomposition, and then performing joint weighted filtering of the user preference score and the conditioning suitability score to obtain a health product recommendation list, including: Based on the aforementioned user segmentation categories, user product interaction records of similar user groups are extracted from historical interaction data. The user latent vector of each user is initialized with the mean of the latent vectors of similar user groups to obtain the initialized user latent vector. The initial user latent vector and the product latent vector of each candidate health product are solved iteratively through matrix factorization objective function to obtain the user preference score of each candidate health product. The user preference score and the conditioning suitability score are weighted and summed to obtain a comprehensive score. The comprehensive score is then filtered and sorted to obtain a list of recommended health products.

[0012] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of generating a physiotherapy treatment plan configuration package based on the user health profile vector and distributing the physiotherapy treatment plan configuration package to surrounding physiotherapy equipment to complete remote preset includes: Based on the user health profile vector, retrieve treatment plan templates from the treatment plan template library that match the user's population category and the type of target physiotherapy equipment; By combining the user health profile vector, each adjustable parameter in the conditioning plan template is assigned a personalized value to obtain a physiotherapy conditioning plan configuration package, and the physiotherapy conditioning plan configuration package is sent to the surrounding physiotherapy equipment through the device communication protocol to complete remote preset.

[0013] In conjunction with the first aspect, in the eighth implementation of the first aspect of the present invention, after the physiotherapy treatment plan configuration package is distributed to the surrounding physiotherapy equipment to complete remote preset, it further includes: The product correction records in the audit feedback signal are analyzed to extract the product identifier that was replaced and the product identifier that was replaced. Combined with the corresponding symptom indication results in the user health profile vector, the direction of the deviation between the symptom category involved in this correction and the candidate health products is determined to obtain the target update signal. Based on the target update signal, the adaptation score coefficient between the replaced product and the corresponding symptom in the conditioning product matching score model is updated by negative gradient, and the adaptation score coefficient between the replaced product and the corresponding symptom is updated by positive gradient, thus completing the iterative update of the conditioning product matching score model.

[0014] Secondly, the present invention provides a health information management system based on an AI big data model, the health information management system based on an AI big data model comprising: The module is used to perform feature fusion encoding on the health screening information collected by the health screening equipment to obtain a fused feature vector. The health profile encoding module is used to input the fused feature vector into the AI ​​big model for health profile encoding to obtain the user health profile vector, and to perform clustering and classification on the user health profile vector to obtain the population classification category. The filtering module is used to filter candidate health products based on the user health profile vector and the population classification, and obtain a recommended list of health products through a conditioning product matching scoring model. The distribution module is used to generate a physiotherapy treatment plan configuration package based on the user health profile vector and distribute the physiotherapy treatment plan configuration package to the surrounding physiotherapy equipment to complete remote preset.

[0015] The technical solution provided by this invention integrates image-based TCM diagnostic data collected by hand and face diagnostic instruments with numerical physiological indicators collected by body index detection devices through cross-modal feature fusion encoding. This constructs a user health profile vector that simultaneously incorporates TCM constitution characteristics and Western medicine physiological state characteristics, solving the problem in existing technologies where TCM diagnostic data and Western medicine physiological indicators cannot be jointly modeled. This allows the health profile to comprehensively reflect the user's overall health status. This invention uses a Gaussian mixture model to cluster and classify the user health profile vector, providing accurate population category basis for personalized services. In the health product recommendation stage, a conditioning product matching scoring model quantifies the compatibility between the user's current symptoms and candidate products. Collaborative filtering matrix decomposition utilizes historical interaction data of similar populations to effectively infer new user preferences. The two methods are combined to form a health product recommendation list, solving the problem of existing recommendation systems not considering individual user health status and new user cold start. This invention automatically generates physiotherapy conditioning plan configuration packages based on the user health profile vector and remotely distributes them to surrounding physiotherapy equipment via device communication protocols for pre-setting, eliminating the need for manual configuration and significantly reducing the professional operational threshold for physiotherapy services. This invention continuously feeds back the review and feedback signals of the recommendation results to the product matching and scoring model for iterative updates, so that the recommendation quality can be continuously optimized as operational data accumulates, forming a closed-loop health management mechanism with user health profiles at its core. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating the steps of the health information management method based on an AI large model in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the construction of the fused feature vector in an embodiment of the present invention; Figure 3 This is a schematic diagram of health profile encoding and clustering classification in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the selection of health products in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the configuration of a remote physiotherapy treatment plan in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the health information management system based on an AI large model in an embodiment of the present invention. Detailed Implementation

[0018] This invention provides a health information management method and system based on an AI large-scale model. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0019] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the health information management method based on AI large model in this invention includes: Step S1: Perform feature fusion encoding on the health screening information collected by the health screening device to obtain a fused feature vector; Step S2: Input the fused feature vector into the AI ​​big model to encode the health profile, obtain the user health profile vector, and perform clustering and classification on the user health profile vector to obtain the population classification category. Step S3: Based on the user health profile vector and population classification, the candidate health products are screened through the conditioning product matching scoring model to obtain a recommended list of health products; This invention uses a Gaussian mixture model to cluster and classify user health profile vectors, providing a precise basis for personalized services based on user demographics. In the health product recommendation process, a product matching scoring model quantifies the compatibility between the user's current symptoms and candidate products. Collaborative filtering matrix decomposition utilizes historical interaction data from similar user groups to effectively infer new user preferences. The combined approach forms a health product recommendation list, solving the problem of existing recommendation systems not considering individual user health status and the cold start effect for new users.

[0020] Step S4: Generate a physiotherapy treatment plan configuration package based on the user's health profile vector and distribute the physiotherapy treatment plan configuration package to the surrounding physiotherapy equipment to complete remote preset.

[0021] This invention automatically generates physiotherapy and conditioning plan configuration packages based on user health profile vectors and remotely distributes them to surrounding physiotherapy equipment via device communication protocols to complete the preset process. No manual intervention is required, significantly reducing the professional operational threshold for physiotherapy services. The invention continuously feeds back the review feedback signals of the recommendation results to the conditioning product matching and scoring model for iterative updates, enabling the recommendation quality to continuously self-optimize with the accumulation of operational data, forming a closed-loop health management mechanism centered on the user health profile.

[0022] This embodiment also includes: Based on the user's testing location information, a set of candidate pharmacists matching the user's geographical location is selected from the registered pharmacists. Each pharmacist in the candidate set is scored using a comprehensive pharmacist scoring model to obtain a comprehensive score. The ranking of these comprehensive scores is then pushed to the user's front end, allowing the user to select and bind their pharmacist as a family health consultant. The user's health profile vector and subsequent health testing updates are continuously synchronized to the family health consultant pharmacist's workbench. The family health consultant pharmacist then provides online health consultation services to the user based on the user's health profile vector. The consultation service data from the family health consultant pharmacist is fed back into the comprehensive pharmacist scoring model to complete iterative updates of the pharmacist's score.

[0023] Specifically, after a user completes a health check and a user health profile vector is generated, the system uses the latitude and longitude coordinates of the check location registered by the user during the check to retrieve all registered pharmacists within a specified radius from the connected pharmacist network, forming a candidate pharmacist set. For each pharmacist in the candidate set, a comprehensive score is calculated using a pharmacist comprehensive scoring model. The input dimensions of the pharmacist comprehensive scoring model include: the number of users served by the pharmacist in the past, user satisfaction feedback ratings for the pharmacist's consultation services, the accuracy rate of the pharmacist's professional review and correction of AI recommendation results, the pharmacist's online response timeliness rate, and the user health improvement indicators of the pharmacist's services. The above-mentioned indicators are weighted and summed according to preset weights to obtain the pharmacist's comprehensive score. The candidate pharmacists are then sorted from high to low according to their comprehensive scores, and along with each pharmacist's basic information (including the name of their affiliated pharmacy, professional qualifications, areas of expertise, and a summary of their historical service evaluations), they are pushed to the pharmacist recommendation list interface of the user's front-end APP. Users can then independently select and confirm a pharmacist as their family health consultant pharmacist based on their own needs. Once the binding relationship is confirmed, the user's health profile vector, personalized health product recommendation list, and updated health test data are continuously synchronized to the family health consultant pharmacist's workbench, enabling the pharmacist to monitor the dynamic health status changes of the bound user in real time. The family health consultant pharmacist provides online health consultation services to the user based on the user's health profile vector. Consultation service data (including consultation initiation time, consultation response time, user consultation satisfaction rating, and the content of the pharmacist's professional advice) is fed back to the pharmacist's comprehensive scoring model in real time, updating the corresponding pharmacist's online response timeliness, satisfaction rating, and other indicators online. This allows the pharmacist's comprehensive score to be dynamically adjusted as service data accumulates. High-ranking, high-quality pharmacists receive higher exposure weight in subsequent user test recommendation lists, forming a service quality-oriented self-optimizing allocation mechanism for pharmacist resources.

[0024] In one specific embodiment, such as Figure 2 The process of executing step S1 can specifically include the following steps: Step S11: When the health screening equipment includes a hand diagnostic instrument, a face diagnostic instrument, and a body index detection device, extract image features from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument to obtain image feature vectors; standardize the physiological indicators collected by the body index detection device to obtain physiological indicator vectors; and perform cross-modal attention fusion and splicing on the image feature vectors and physiological indicator vectors to obtain fused feature vectors. Step S12: When the health screening equipment only includes a hand diagnostic instrument and a face diagnostic instrument, extract image features from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument respectively to obtain image feature vectors; use the image feature vectors as fusion feature vectors.

[0025] Specifically, the data streams output from the hand and face diagnostic instruments are connected to the same front-end acquisition link, and a separate channel processing method is maintained before entering the fusion calculation. The hand diagnostic image focuses on preserving visual cues such as palm print texture, palm color distribution, and local morphological undulations, while the face diagnostic image focuses on preserving visual cues such as facial color levels, regional brightness variations, and contour states. In the image feature extraction stage, the original image is subjected to denoising, color space conversion, and effective region cropping before being input into the image encoding network to form a unified-dimensional image feature representation. The acquired image is first subjected to bilateral filtering to suppress random noise, and then the color representation is converted from RGB to LAB space to reduce the interference caused by ambient light fluctuations on facial color judgment and palm print boundary recognition. The target detection model is used to locate the palm or facial region, and the region of interest is cropped to a fixed size. The hand and face diagnostic images are respectively input into a convolutional neural network for layer-by-layer convolution extraction and fully connected mapping, outputting a unified-dimensional image feature vector. For example, the image feature dimension can be set to 256 dimensions, so that the texture, color, and structural information can maintain sufficient expression density in the fusion process. Meanwhile, the physiological indicators output by the body index detection device, such as blood pressure, blood oxygen saturation, body fat percentage, and heart rate variability, are first standardized due to their different units and magnitudes. These indicators are compressed to a comparable scale before being input into a multi-layer fully connected network for nonlinear mapping, forming a physiological indicator vector with the same dimension as the image feature vector (e.g., 256 dimensions). The image feature vector is used as the primary query information, and the physiological indicator vector is used as the auxiliary information retrieved. Attention weights adaptively determine the contribution of each physiological indicator to the current image semantics, and the attention weight matrix is ​​applied to the image feature vector. The mapped physiological indicator vector yields weighted physiological semantic features, which are then concatenated with the original image feature vector to form a fused feature vector.

[0026] Specifically, in scenarios where body index detection equipment is not deployed, the hand and face images collected by the hand and face diagnostic instruments undergo bilateral filtering for noise reduction, RGB-to-LAB color space conversion, and region of interest cropping, respectively. These are then input into a convolutional neural network for image feature extraction to obtain image feature vectors. Since there is no physiological indicator data channel in this scenario, the image feature vectors do not need to be fused with physiological indicator vectors across modalities. Instead, they are directly input into the portrait encoding layer of the AI ​​large model. The portrait encoding layer performs single-path semantic encoding on the image feature vectors along the constitution semantic parsing path, outputting a fused feature vector containing only TCM constitution characteristics. This fused feature vector serves as input for subsequent Gaussian mixture model clustering and classification, product matching and scoring, and the generation of physiotherapy and conditioning plan configuration packages. Compared to the complete mode that simultaneously connects three types of equipment, this mode lacks the physiological indicator dimension of the fused feature vector, but the expression of the constitution characteristic dimension remains complete and effective, supporting basic health product recommendation and conditioning plan configuration functions. It is suitable for community pharmacies or health centers that only have TCM diagnostic equipment.

[0027] In one specific embodiment, such as Figure 3 The process of executing step S2 can specifically include the following steps: Step S21: Input the fused feature vector into the AI ​​large model to perform health status semantic encoding to obtain the user health profile vector; Step S22: Based on the user health profile vector, the Gaussian mixture model is iteratively solved by the expectation-maximization algorithm to obtain the weights, mean vector and covariance matrix of each component and the user is assigned to the population category, thus obtaining the population classification category.

[0028] Specifically, the fused feature vector is input into the portrait encoding layer trained for health management scenarios. The portrait encoding layer simultaneously performs deep representation learning around two analysis paths: TCM constitution semantics and physiological state semantics. This allows the already aligned image information and physiological indicator information in the fused feature vector to be further compressed, recombined, and enhanced within a unified semantic space, forming constitution feature vectors and physiological state feature vectors respectively. After the two types of semantic results are aligned, they are then concatenated and encoded to form the user health portrait vector. This allows the user health portrait vector to retain both the constitution characteristics reflected in the hand and face diagnosis images and the physiological state characteristics output by the body index detection device, resulting in a comprehensive health expression. Based on user health profile vectors, population segmentation is performed. In this segmentation process, a Gaussian mixture model trained on historical samples is used as the foundation for describing the population structure. Iterative estimation is then performed around the weights, mean vectors, and covariance matrices of each component, allowing different health status distributions to be represented by multiple probability components. The posterior probability of a user belonging to each Gaussian component is calculated based on the user health profile vector, and the parameters of each component are continuously updated based on the posterior probability results until the model converges, forming a stable set of population segmentation parameters. In the online application phase, the membership degree of each component corresponding to the current user health profile vector is recalculated using the converged model parameters, and the category corresponding to the component with the highest posterior probability is determined as the user's population category. The probability density function of the above Gaussian mixture model is expressed as: ; in, Vectorization of user health profile For the number of population subtypes, For the first The mixing weights of Gaussian components, For the first The mean vector of each component For the first The covariance matrix of the n components satisfies the constraints that the sum of the weights of all components equals 1 and that the weights of each component are all greater than 0. In the E-step of the expectation maximization iteration, the nth... User sample Belonging to the The posterior probability (responsibility) of each Gaussian component is calculated as follows: ; The sum of the responsibility of the same sample to all components is 1. In M steps, the mean vector, covariance matrix and component weights are updated by weighting according to the responsibility of all samples. The two steps are alternated and iterated until the change of the log-likelihood function in two adjacent rounds is lower than the convergence threshold, and then the converged Gaussian mixture model parameter set is output.

[0029] In one specific embodiment, the process of performing step S21 may specifically include the following steps: (1) Input the fused feature vector into the image encoding layer of the AI ​​large model to perform physical condition semantic parsing and physiological state semantic parsing to obtain physical condition feature vector and physiological state feature vector; (2) The physical feature vector and the physiological state feature vector are concatenated and encoded to obtain the user health profile vector.

[0030] Specifically, the fused feature vector is directly used as input to the portrait encoding layer. Within this layer, a constitution semantic analysis pathway and a physiological state semantic analysis pathway are established. This allows the same fused feature to be specifically represented and refined in two semantic directions. The constitution semantic analysis pathway focuses on identifying palm print texture, color distribution, facial contour changes, and implicit patterns related to TCM constitution assessment in hand and face diagnosis images that have already incorporated the fused features. After multiple layers of nonlinear mapping, it outputs a constitution feature vector. The physiological state semantic analysis pathway focuses on analyzing the numerical correlation, fluctuation relationship, and overall trend of physiological indicators such as blood pressure, blood oxygen, body fat percentage, and heart rate variability within the fused features. After semantic compression and feature reshaping, it outputs a physiological state feature vector. The portrait encoding layer introduces a feature alignment mechanism in the middle layer between the two pathways to coordinate the scale distribution, semantic center of gravity, and importance weights of constitution and physiological state semantic information. This ensures that the constitution feature vector does not obscure the physiological state feature vector due to visual information dominance, nor does it weaken the diagnostic value of constitution information due to large fluctuations in numerical indicators. After completing the two-way semantic parsing, the physical constitution feature vector and the physiological state feature vector are concatenated and encoded in a predetermined order, and then unified and integrated through a subsequent mapping layer to obtain the user health profile vector, thereby forming a comprehensive health representation that simultaneously covers traditional Chinese medicine physical constitution features and modern physiological state features.

[0031] In one specific embodiment, the process of performing step S22 may specifically include the following steps: (1) Based on the weights, mean vectors and covariance matrices of each component, calculate the first posterior probability of each component in the Gaussian mixture model to which the user health profile vector belongs; (2) The weights, mean vector and covariance matrix of each component of the Gaussian mixture model are updated by weighting according to the first posterior probability until the log-likelihood function converges, and the parameters of the Gaussian mixture model are obtained. (3) Based on the parameters of the Gaussian mixture model, calculate the second posterior probability of each component to which the user health profile vector belongs, and take the population category corresponding to the component with the largest second posterior probability as the user's population classification category.

[0032] Specifically, the user health profile vector output from the health profile encoding layer is used as the input to the Gaussian mixture model, and a population distribution representation composed of multiple Gaussian components is established during the offline training phase, where the user health profile vector is denoted as... , The sample number is represented by , and the total number of components is denoted as . , Indicates the first The weight of each component, Indicates the first The mean vector of each component Indicates the first The covariance matrix of each component is used as a basis for the expectation maximization iteration. First, the first posterior probability (i.e., responsibility) of each sample's health profile vector to each Gaussian component is calculated based on the current round's component weights, mean vector, and covariance matrix. This yields the probability assignment of the current sample to each population component, with the sum of the responsibility scores for all components corresponding to the same sample remaining constant at 1. This establishes a soft assignment relationship between a single health profile vector and multiple population subtypes. After calculating the first posterior probability, the model parameters are updated with weighted averages based on the responsibility scores of all samples. After one round of updates, the changes in the log-likelihood function are compared between adjacent rounds. When the change is lower than a preset convergence threshold, the iteration stops, and a stable set of Gaussian mixture model parameters is output, achieving unified modeling of different health status groups in terms of mean position, dispersion, and distribution proportion. In the online application stage, the current user's health profile vector is input into the converged Gaussian mixture model, and the second posterior probability of the current user to each Gaussian component is recalculated. The component with the highest second posterior probability is used as the current user's population subtype.

[0033] In one specific embodiment, such as Figure 4 The process of executing step S3 can specifically include the following steps: Step S31: Based on the user's health profile vector, the matching score model of conditioning products is used to calculate the weighted score of the compatibility between each candidate health care product and the user's current symptoms, so as to obtain the conditioning compatibility score of each candidate health care product. Step S32: Based on the population classification, infer user preferences from the historical interaction data of the same population through collaborative filtering matrix decomposition to obtain user preference scores. Combine user preference scores with conditioning suitability scores for joint weighted filtering to obtain a list of recommended health products.

[0034] Specifically, the health profile analysis unit performs symptom identification on the user's health profile vector, extracting the symptom indication results and symptom weight results corresponding to the current user. Then, candidate health products are fed into the conditioning product matching and scoring model one by one. Combining the conditioning effect coefficients pre-established for each product for different symptom categories, a conditioning suitability score is calculated between the candidate health products and the current user's symptom state. When the user's current health state matches a certain symptom category, the corresponding indication result is taken as valid. The weight corresponding to that symptom category and the product effect coefficient are then added to the final score, thus obtaining the conditioning suitability score for each candidate health product for the current user.

[0035] The formula for calculating the above conditioning fit score is: ; in, As a candidate health and wellness product, This is the current user's health profile vector. The total number of predefined disease categories. For the first The weight of each disease is calculated by weighting the disease severity score and the user's attention to the disease. For products Targeting the symptoms The pre-assessment adaptation effect coefficient is established by the AI ​​big model after analyzing the product efficacy instructions and the traditional Chinese medicine knowledge base, and is continuously iterated and updated according to the pharmacist review feedback signal; For the symptom indication function, when the user's health profile vector is determined to match the first... The value is 1 when a certain symptom characteristic is present, and 0 otherwise. The product set that meets the recommendation threshold τ is then filtered. ; Recommended threshold On the historical user product interaction verification set, different candidates are calculated. The recommended precision rate for the given value is selected by choosing the minimum precision rate that meets the minimum precision requirement. The value is determined and is periodically reassessed and updated as operational data accumulates.

[0036] This method uses user segmentation as a filtering criterion for historical interaction data. It extracts user samples belonging to the same segment as the current user from the historical interaction database and constructs a relationship between users and products based on interactions such as browsing, consultation, adding to cart, purchasing, and feedback. Then, it extracts the potential preference features of similar user groups for candidate health products through collaborative filtering matrix decomposition, resulting in the current user's preference score for each candidate health product. This user preference score reflects the product acceptance tendency and selection patterns exhibited by similar health segment groups during long-term interactions. Therefore, it can compensate for cold-start bias, interest bias, and selection sparsity that may occur when relying solely on health status scores. Especially when the current user has limited historical interactions, it can initialize and correct the current user's preferences using common preferences from similar groups, ensuring that the recommendation results balance treatment effectiveness and actual acceptance. The user preference score and treatment suitability score are jointly weighted and filtered, giving higher overall rankings to candidate health products with strong health needs and high group preference, while naturally shifting them down the ranking. Finally, a health product recommendation list is output based on the overall ranking results.

[0037] In one specific embodiment, the process of performing step S32 may specifically include the following steps: (1) Based on the user classification, extract the user product interaction records of the same group from the historical interaction data, and initialize the user latent vector of the user with the mean of the latent vector of the same group to obtain the initialized user latent vector. (2) The initial user latent vector and the product latent vector of each candidate health care product are solved iteratively by matrix factorization objective function to obtain the user preference score of each candidate health care product; (3) The user preference score and the conditioning suitability score are weighted and summed to obtain a comprehensive score. The comprehensive score is then filtered and sorted to obtain a list of recommended health products.

[0038] Specifically, based on the user segmentation results, a sample set belonging to the same segmentation category as the current user is identified in the historical interaction database. Then, user product interaction records corresponding to browsing, inquiries, adding to cart, purchasing, and manual review feedback are extracted from this sample set. Based on this, the user latent vector set trained by the same user group is read. The mean of the latent vectors of the same user group is used as the initial representation of the current user, allowing the current user to inherit the common preference direction of the same user group even with limited historical interactions or before a stable behavioral trajectory has been formed, thus completing the user preference initialization in the cold start phase. The initialized user latent vector and the product latent vectors corresponding to each candidate health product are fed into the matrix decomposition solution process. Through iterative minimization of errors on the observed user product interaction set, the user latent vector matrix and the product latent vector matrix are continuously corrected, gradually approximating the vector positions in the latent space to the actual interaction relationships. After the objective function continuously converges iteratively, the user's preference score for each candidate health product can be given by the inner product of the user latent vector and the corresponding product latent vector, which is used to characterize the current user's acceptance tendency for different candidate health products under the constraints of the behavioral patterns of the same user group.

[0039] The objective function for the above matrix decomposition is: ; in, For users For the product The implicit feedback score is derived by weighting user browsing, consultation, adding to cart, purchasing, and positive feedback from pharmacists according to preset weights. For users The user's latent vector, For products The product's latent vector, This is a collection of all observed user product interaction records. The regularization coefficient is used to prevent overfitting of the latent vectors. It is the Frobenius norm. The objective function is obtained by stochastic gradient descent. Iterative optimization is performed on the samples, and the user finds the solution after convergence. For the product User preference rating by and The inner product is given.

[0040] User preference scores and conditioning fit scores are jointly weighted to ensure that the intensity of health needs and group behavioral preferences work together within the same ranking process. The conditioning fit score reflects the degree of matching between candidate health products and the current symptom state, while the user preference score reflects the potential acceptance of candidate health products by the current user under the consumption patterns of similar groups. The two scores are weighted and summed to form a comprehensive score. Then, the products are filtered and sorted from high to low according to the comprehensive score, and a list of recommended health products that meet the recommendation criteria is output.

[0041] In one specific embodiment, such as Figure 5 The process of executing step S4 can specifically include the following steps: Step S41: Based on the user's health profile vector, retrieve a treatment plan template from the treatment plan template library that matches the user's population category and the type of target physiotherapy equipment; Step S42: Personalize the adjustable parameters in the treatment plan template by combining the user health profile vector to obtain the physiotherapy treatment plan configuration package, and send the physiotherapy treatment plan configuration package to the surrounding physiotherapy equipment through the device communication protocol to complete the remote preset.

[0042] Specifically, the user's health profile vector is used as input to the physiotherapy execution link. The physiotherapy indication determination unit performs an adaptability analysis on the user's current health status to determine whether the user meets the usage conditions of the target physiotherapy methods such as moxibustion chambers and physiotherapy devices, and outputs the corresponding physiotherapy type identifier. When the physiotherapy indication determination result is valid, the background application service system combines the user's population category, the type of target physiotherapy equipment, and the real-time availability status of the equipment network to limit the range of candidate physiotherapy equipment. Then, using "equipment type + population category" as a dual index condition, the system retrieves the corresponding treatment plan template from the treatment plan template library. This ensures that the templates entering the search result set not only match the control capabilities of the target physiotherapy equipment, but also maintain consistency with the common treatment patterns of the current user's population classification, thereby avoiding the direct application of incompatible equipment parameter templates to the current user. The treatment plan template records the range of basic treatment parameters and parameter constraints that can be executed by this type of device. For example, when the target physiotherapy device is a moxibustion chamber, the treatment plan template can include programmable fields such as the combination code of moxibustion acupoints, the temperature gradient setting value of each acupoint, and the duration of a single moxibustion treatment; when the target physiotherapy device is a physiotherapy instrument, the treatment plan template can include device-specific parameters such as the output frequency value, intensity level, and waveform type code.

[0043] The physical characteristics, physiological state characteristics, and population category information from the user's health profile vector are fed into the parameter mapping unit. Personalized assignment processing is performed on each adjustable parameter in the treatment plan template, transforming the basic template from a "group-applicable template" into an "individually exclusive configuration." Corresponding assignment rules are established according to parameter categories, with value mapping logic set for temperature, duration, frequency, intensity, and waveform parameters. Based on the quantification results of each health feature dimension in the user's health profile vector, candidate values ​​within the template parameter range are filtered and refined to obtain the target parameter combination corresponding to the current user's health status. The target physiotherapy device identifier, the names of each programmable parameter, the set values ​​of each programmable parameter, the treatment duration, and necessary task control fields are encapsulated into a physiotherapy treatment plan configuration package. Standardized encoding is performed according to the communication format corresponding to the target device, enabling the physiotherapy treatment plan configuration package to be directly parsed by the device control unit. Before sending, the physiotherapy treatment plan configuration package is encrypted and signed. Then, according to the equipment manufacturer's access specifications, the physiotherapy treatment plan configuration package is sent to the target physiotherapy equipment control unit using the MQTT protocol or RESTful API. The target physiotherapy equipment control unit completes parameter writing, status verification, and preset activation, thus completing the remote preset. The backend application service system synchronously generates appointment confirmation information and pushes the arrival time, physiotherapy store address, equipment location number, and treatment plan summary to the user's front end. After the user arrives at the store, the target physiotherapy equipment enters the ready state according to the sent physiotherapy treatment plan configuration package, and the user can directly start the treatment process after identity verification.

[0044] In one specific embodiment, after the physiotherapy treatment plan configuration package is distributed to the surrounding physiotherapy equipment to complete remote preset, the method further includes: (1) Analyze the product correction records in the audit feedback signal, extract the product identifier that was replaced and the product identifier after replacement, and combine them with the corresponding symptom indication results in the user health profile vector to determine the direction of the adaptation relationship deviation between the symptom category involved in this correction and the candidate health products, and obtain the target update signal; (2) Based on the target update signal, perform negative gradient update on the matching score coefficient between the replaced product and the corresponding disease in the treatment product matching score model, and perform positive gradient update on the matching score coefficient between the replaced product and the corresponding disease to complete the iterative update of the treatment product matching score model.

[0045] Specifically, the product correction records in the review feedback signals are subjected to structured parsing, and the replaced product identifier, the replaced product identifier, the corresponding user identifier, the correction time, and the correction source mark are extracted from the product correction records. Then, the user health profile vector and symptom indication results corresponding to the same user at the time of recommendation generation are retrieved, and the symptom categories that have been determined to be valid in the user health profile vector are associated and matched with the current correction record, thereby locking in the range of symptom categories that the current correction actually affects. The product replacement action is interpreted as a directional correction to the disease-product fit relationship. The recommendation result corresponding to the replaced product identifier is identified as "current fit score is too high", and the recommendation result corresponding to the replaced product identifier is identified as "current fit score is too low". Combined with the disease indication results, it is determined whether the deviation occurs under a single disease category or in a complex scenario where multiple disease categories participate in the scoring. When only one disease category is active in the disease indication results, the target update signal can be directly bound to the fit relationship between the corresponding disease category and the two product identifiers. When multiple disease categories are active at the same time, the contribution of this correction is attributed and split according to the disease weight distribution formed by each disease category in the recommendation calculation stage. This ensures that the target update signal not only contains the identification information of "which product was replaced and which product was replaced", but also the directional information of "which disease caused this replacement" and "in which direction should the fit relationship be corrected", forming a structured update basis for model parameter updates. In the review results pushed by the health service center management system, the approval of recommended products is recorded as a positive feedback signal, while the modification or replacement of recommended products is recorded as a correction feedback signal. The correction content includes the original recommended product identifier and the identifier of the replaced product, and is fed back to the AI ​​central processing system in real time as an update signal source. At the same time, the matrix elements in the symptom-product pre-assessment effect coefficient matrix are used to represent the expected treatment effect score of the product on the corresponding symptom. The matrix elements will be iteratively updated as the pharmacist feedback signals are continuously fed back.

[0046] The AI ​​central processing system sends the target update signal to the parameter update unit of the product matching scoring model. It then performs gradient corrections in opposite directions for both the replaced and new products. Specifically, the matching score coefficients between the replaced product and its corresponding symptoms are updated in the negative adjustment direction, while the matching score coefficients between the new and new products are updated in the positive adjustment direction. This gradually reduces previously overestimated matching relationships and increases previously underestimated matching relationships. After continuous feedback accumulation, the symptom-product pre-assessment effect coefficient matrix gradually approaches the actual business judgment after review by licensed pharmacists. When the same replacement direction occurs repeatedly under the same symptom category, the parameter update unit can accumulate update intensity according to the frequency of correction operations, allowing high-frequency correction relationships to converge faster. When different pharmacists give opposite corrections for the same product relationship, the parameter update unit retains the superposition of positive and negative update amounts, ensuring that the scoring coefficients reflect the long-term statistical trend of matching. After an update is completed, the new adaptation score coefficient is re-entered into the recommendation calculation process and directly participates in the formation of the adaptation score in the next round of recommendations. Therefore, the recommendation results will be continuously corrected as review feedback signals continue to flow in.

[0047] This embodiment also includes: based on user health profile vectors and personalized health product recommendation lists, using an AI big data model to match and analyze the correlation between the store's main products and the user's current health status, and obtaining product health correlation analysis results; inputting the product health correlation analysis results into the AI ​​big data model to generate sales scripts, and obtaining personalized sales scripts and sales guidance materials for the current user's health status.

[0048] In this embodiment, addressing the problem that pharmacy or wellness center sales staff lack professional medical knowledge and are unable to effectively correlate users' health status with the wellness products promoted by the store, this invention, based on the completion of user health profile vector construction and personalized wellness product recommendation list generation, further utilizes an AI big data model to automatically generate personalized sales scripts and sales guidance materials.

[0049] The specific implementation process is as follows: The system obtains the information of the main products uploaded by the store in the health service center management system, including product name, product efficacy description and applicable population description; the AI ​​big model takes the user health profile vector as input and analyzes the correlation between the store's main products and the user's current health status one by one. It performs semantic matching between the specific symptom characteristics identified in the user health profile vector and the efficacy description of each main product, filters out the main products that have a clear correlation with the user's current health status, extracts the correlation basis, and obtains the product health correlation analysis results. The product health correlation analysis results include the user's specific symptom matching points and efficacy correspondence for each main product.

[0050] After obtaining the product health correlation analysis results, the AI ​​big data model uses these results and the current user's personalized health product recommendation list as input. Based on the correlation between the user's specific health problems and the corresponding product efficacy, it generates personalized sales scripts and sales guidance materials for sales personnel. The sales scripts use the user's specific health test results as a starting point, explaining the correspondence between the user's symptoms and the product's conditioning effects one by one, forming targeted product recommendations. The sales guidance materials include the product recommendation order, the corresponding user health correlation explanation for each product, and the reasons for the recommendation. Sales personnel can make professional product recommendations to the current user based on the personalized sales scripts and sales guidance materials, without needing professional medical or traditional Chinese medicine background knowledge, thereby reducing the store's reliance on the professional quality of sales personnel and reducing the store training costs for health product manufacturers.

[0051] The above describes the health information management method based on an AI large model in the embodiments of the present invention. The following describes the health information management system based on an AI large model in the embodiments of the present invention. Please refer to [link / reference]. Figure 6 One embodiment of the health information management system based on an AI large model in this invention includes: Module 1 is used to perform feature fusion encoding on the health screening information collected by the health screening equipment to obtain a fused feature vector; Health profile encoding module 2 is used to input the fused feature vector into the AI ​​big model for health profile encoding, to obtain user health profile vector, and to perform clustering and classification on the user health profile vector to obtain population classification categories; The filtering module 3 is used to filter candidate health products based on user health profile vectors and population classification categories, and obtain a recommended list of health products through a product matching scoring model. The distribution module 4 is used to generate a physiotherapy treatment plan configuration package based on the user's health profile vector and distribute the physiotherapy treatment plan configuration package to the surrounding physiotherapy equipment to complete remote preset.

[0052] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0053] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A health information management method based on an AI large-scale model, characterized in that, include: The health screening information collected by the health screening equipment is fused and encoded to obtain a fused feature vector. The fused feature vector is input into the AI ​​large model for health profile encoding to obtain the user health profile vector, and the user health profile vector is clustered and classified to obtain the population classification category. Based on the user health profile vector and the population classification, a health product recommendation list is obtained by screening candidate health products through a conditioning product matching and scoring model. Based on the user's health profile vector, a physiotherapy treatment plan configuration package is generated and distributed to surrounding physiotherapy equipment to complete remote preset.

2. The health information management method based on an AI large model according to claim 1, characterized in that, The process of performing feature fusion encoding on the health screening information collected by the health screening device to obtain a fused feature vector includes: When the health screening device includes a hand diagnostic instrument, a face diagnostic instrument, and a body index detection device, image features are extracted from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument to obtain image feature vectors; the physiological indicators collected by the body index detection device are standardized to obtain physiological indicator vectors; the image feature vectors and the physiological indicator vectors are fused and spliced ​​together by cross-modal attention to obtain fused feature vectors. When the health screening device includes only a hand diagnostic instrument and a face diagnostic instrument, image features are extracted from the hand diagnostic images collected by the hand diagnostic instrument and the face diagnostic images collected by the face diagnostic instrument to obtain image feature vectors; the image feature vectors are then used as fusion feature vectors.

3. The health information management method based on an AI large model according to claim 2, characterized in that, The process involves inputting the fused feature vector into an AI large-scale model for health profile encoding to obtain a user health profile vector, and then performing clustering and classification on the user health profile vector to obtain population classification categories, including: The fused feature vector is input into the AI ​​large model for health status semantic encoding to obtain the user health profile vector; Based on the user health profile vector, the Gaussian mixture model's component weights, mean vector, and covariance matrix are iteratively solved using the expectation-maximization algorithm, and the user is assigned to a population category, thus obtaining the population classification category.

4. The health information management method based on an AI large model according to claim 3, characterized in that, Also includes: Based on the user's testing location information, a set of candidate pharmacists matching the user's testing location is selected from the registered pharmacists. Each pharmacist in the candidate pharmacist set is scored and calculated according to the pharmacist comprehensive scoring model to obtain the comprehensive score of each candidate pharmacist. The comprehensive score ranking results are pushed to the user's front end, and the user can select and bind a family health consultant pharmacist. The user's health profile vector and subsequent health monitoring updates are continuously synchronized to the workbench of the family health consultant pharmacist. The family health consultant pharmacist then provides online health consultation services to the user based on the user's health profile vector. The consultation service data of the family health consultant pharmacist is fed back to the pharmacist comprehensive scoring model to complete the iterative update of the pharmacist's score.

5. The health information management method based on an AI large model according to claim 3, characterized in that, Based on the user health profile vector, the Gaussian mixture model's component weights, mean vector, and covariance matrix are iteratively solved using the expectation-maximization algorithm to assign users to their respective population categories, resulting in population classification categories, including: Based on the weights of each component, the mean vector, and the covariance matrix, the first posterior probability of each component in the Gaussian mixture model to which the user health profile vector belongs is calculated. The weights, mean vectors, and covariance matrices of each component of the Gaussian mixture model are updated in a weighted manner based on the first posterior probability until the log-likelihood function converges, thus obtaining the parameters of the Gaussian mixture model. Based on the parameters of the Gaussian mixture model, the second posterior probability of each component to which the user health profile vector belongs is calculated, and the population category corresponding to the component with the largest second posterior probability is taken as the user's population classification category.

6. The health information management method based on an AI large model according to claim 5, characterized in that, The process involves filtering candidate health products based on the user health profile vector and the population classification, using a product matching and scoring model to obtain a recommended list of health products, including: Based on the user health profile vector, a weighted score is calculated on the compatibility between each candidate health product and the user's current symptoms using a product matching scoring model, resulting in a compatibility score for each candidate health product. Based on the aforementioned population classification, user preferences are inferred from the historical interaction data of similar populations through collaborative filtering matrix decomposition to obtain user preference scores. The user preference scores are then combined with the conditioning suitability scores for weighted filtering to obtain a list of recommended health products.

7. The health information management method based on an AI large model according to claim 6, characterized in that, Based on the population classification, user preferences are inferred from historical interaction data of similar populations through collaborative filtering matrix decomposition to obtain user preference scores. These user preference scores are then jointly weighted and filtered with the conditioning suitability scores to obtain a list of recommended health products, including: Based on the aforementioned user segmentation categories, user product interaction records of similar user groups are extracted from historical interaction data. The user latent vector of each user is initialized with the mean of the latent vectors of similar user groups to obtain the initialized user latent vector. The initial user latent vector and the product latent vector of each candidate health product are solved iteratively through matrix factorization objective function to obtain the user preference score of each candidate health product. The user preference score and the conditioning suitability score are weighted and summed to obtain a comprehensive score. The comprehensive score is then filtered and sorted to obtain a list of recommended health products.

8. The health information management method based on an AI large model according to claim 7, characterized in that, The process of generating a physiotherapy treatment plan configuration package based on the user's health profile vector and distributing the physiotherapy treatment plan configuration package to surrounding physiotherapy equipment to complete remote preset includes: Based on the user health profile vector, retrieve treatment plan templates from the treatment plan template library that match the user's population category and the type of target physiotherapy equipment; By combining the user health profile vector, each adjustable parameter in the conditioning plan template is assigned a personalized value to obtain a physiotherapy conditioning plan configuration package, and the physiotherapy conditioning plan configuration package is sent to the surrounding physiotherapy equipment through the device communication protocol to complete remote preset.

9. The health information management method based on an AI large model according to claim 8, characterized in that, After the physiotherapy treatment plan configuration package is distributed to the surrounding physiotherapy equipment to complete the remote preset, it also includes: The product correction records in the audit feedback signal are analyzed to extract the product identifier that was replaced and the product identifier that was replaced. Combined with the corresponding symptom indication results in the user health profile vector, the direction of the deviation between the symptom category involved in this correction and the candidate health products is determined to obtain the target update signal. Based on the target update signal, the adaptation score coefficient between the replaced product and the corresponding symptom in the conditioning product matching score model is updated by negative gradient, and the adaptation score coefficient between the replaced product and the corresponding symptom is updated by positive gradient, thus completing the iterative update of the conditioning product matching score model.

10. A health information management system based on an AI large-scale model, characterized in that, A method for implementing health information management based on an AI large model as described in any one of claims 1-9, comprising: The module is used to perform feature fusion encoding on the health screening information collected by the health screening equipment to obtain a fused feature vector. The health profile encoding module is used to input the fused feature vector into the AI ​​big model for health profile encoding to obtain the user health profile vector, and to perform clustering and classification on the user health profile vector to obtain the population classification category. The filtering module is used to filter candidate health products based on the user health profile vector and the population classification, and obtain a recommended list of health products through a conditioning product matching scoring model. The distribution module is used to generate a physiotherapy treatment plan configuration package based on the user health profile vector and distribute the physiotherapy treatment plan configuration package to the surrounding physiotherapy equipment to complete remote preset.

Images