AI-based major customer nutrition case rapid evaluation system and method

Abstract

The invention relates to the technical field of medical information, and discloses an AI-based major customer nutrition case rapid evaluation system and method.The AI-based major customer nutrition case rapid evaluation method comprises the steps that crowd statistical characteristic data are processed through a meta-learning network, and a sampling basic model for specific crowds is generated; constructing a Bayesian decision network to calculate a nutrition index information value, and generating an index information cost ratio; according to the index information cost ratio, adopting a Markov decision process to construct a sequential sampling strategy, and executing adaptive index sampling; in the adaptive index sampling execution process, the information gain rate is calculated to realize dynamic termination decision, and excessive sampling is avoided; a complete nutrition evaluation result is generated by applying conditional probability inference, and a high-precision evaluation result can be obtained only by measuring a small number of indexes; according to the method, the problem of resource limitation in major customer nutrition case evaluation is solved by utilizing an information value theory and a meta-learning technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, more particularly, it relates to an AI-based large customer nutrition case rapid assessment system and method. BACKGROUND

[0002] In the nutrition management of large customer groups such as large enterprises and schools, medical institutions need to efficiently assess the nutritional status of the group to provide scientific basis for health management. Traditional nutrition assessment methods require completing all detection items for each member, usually containing dozens of indicators, resulting in high detection cost, long time consumption and resource limitation.

[0003] The existing nutrition assessment technology mainly has the following problems: it is impossible to achieve high-precision nutrition state assessment under limited detection indicators, especially lacking individual difference adaptability for different populations; the existing sampling method cannot accurately quantify the balance between index information value and sampling cost, resulting in unreasonable resource allocation; the existing sparse sampling method lacks adaptability to group differences and cannot quickly adjust the sampling strategy according to different population characteristics; in emergency situations, there is a lack of scientific decision-making method to guide resource allocation, which cannot maximize the value of limited resources.

[0004] Therefore, a large customer nutrition case rapid assessment method is needed to overcome the above technical problems, which can achieve efficient and accurate nutrition status assessment under limited resources. SUMMARY

[0005] The present application provides an AI-based large customer nutrition case rapid assessment system and method, which solves the technical problem of being unable to achieve high-precision nutrition assessment under limited resources in related technologies.

[0006] The present application provides an AI-based large customer nutrition case rapid assessment method, comprising the following steps: Using a meta-learning network to process population statistical feature data to generate a sampling base model for a specific population; Based on the sampling base model, a Bayesian decision network is constructed to calculate the information value of the nutrition indicators, and an index information cost ratio is generated, which is defined as the ratio of expected information gain to the product of economic cost and time cost; According to the index information cost ratio, a Markov decision process is used to construct a sequential sampling strategy to perform adaptive index sampling; In the process of performing adaptive index sampling, the information gain rate is calculated to realize dynamic termination decision, avoiding over-sampling; Based on the sampling results after the dynamic termination decision, the conditional probability inference is applied to generate complete nutrition assessment results and provide reliability estimation of the results.

[0007] In a preferred embodiment, the step of processing the statistical feature data of the population by the meta-learning network comprises: Collecting statistical feature data of the target population, including demographic characteristics, professional characteristics, regional characteristics, and historical nutritional status and multi-dimensional data thereof; Constructing a meta-learning network, which adopts a double-layer structure, with the outer network responsible for learning the mapping function of the population characteristics to the parameters of the inner network, and the inner network responsible for implementing the specific sampling strategy; Training the meta-learning network using a task-based meta-learning paradigm; Calculating the initial parameters of the sampling strategy for a new population by the trained meta-learning network, and generating a sampling base model suitable for the new population.

[0008] In a preferred embodiment, the step of constructing a Bayesian decision network to calculate the information value of the nutritional indicators comprises: Constructing a Bayesian network model to represent the probabilistic dependency relationship between the nutritional indicators; Learning the structure and parameters of the Bayesian network based on historical data and expert knowledge; Calculating the expected information gain of each potential sampling indicator to quantify the information contribution of the expected information gain to the nutritional status evaluation; Introducing an information cost ratio to comprehensively consider the information value, economic cost, and time cost of the indicators.

[0009] In a preferred embodiment, the step of constructing a sequential sampling strategy using a Markov decision process comprises: Constructing a Markov decision process model to formalize the sequential sampling problem as a state transition process; Implementing an improved Markov blanket information propagation algorithm to effectively utilize the conditional dependency relationship between indicators to predict the probability distribution of unmeasured indicators; Selecting the next optimal measurement indicator based on the current state and the information cost ratio; Performing measurement and updating the state.

[0010] In a preferred embodiment, the Markov decision process model comprises: State space, representing the state of the current set of measured indicators and their observed values; Action space, representing the set of indicators that can be measured next; State transition function, representing the probability of transitioning to the next state after selecting a measurement indicator in a state; Reward function, reflecting the net benefit of measuring an indicator, defined as the information gain of the measured indicator minus the measurement cost.

[0011] In a preferred embodiment, the step of calculating the information gain rate to realize the dynamic termination decision comprises: defining the information gain rate to measure the marginal information gain brought by each additional measurement indicator; implementing the threshold-based dynamic termination algorithm, setting two termination conditions of the information gain rate threshold condition and the evaluation accuracy condition, and stopping sampling when either condition is met; building an evaluation accuracy estimation model to predict the accuracy of the evaluation result in the current state in real time; adapting the termination threshold, dynamically setting the termination condition parameters according to the urgency of the task and resource constraints.

[0012] In a preferred embodiment, the step of applying conditional probability inference to generate a complete nutritional evaluation result comprises: using a probabilistic graphical model to infer the conditional probability distribution of unmeasured indicators, based on the state when sampling is terminated, to infer the complete distribution of nutritional indicators; combining the conditional probability distribution with expert rules to generate the nutritional status evaluation result; calculating the reliability estimate of the evaluation result to quantify the uncertainty of the result; based on the evaluation result and the reliability estimate, generating personalized nutrition recommendations.

[0013] In a preferred embodiment, the reliability estimate of the evaluation result is calculated by the following formula: ; wherein, represents the reliability estimate value of the evaluation result; represents the normalized conditional entropy of the target nutritional state variable under the state , which is used to quantify the uncertainty of the nutritional status evaluation result under the current sampling state; represents the state when sampling is terminated, including all measured indicators and their observed values; the normalized conditional entropy scales the original conditional entropy to the interval, making the reliability estimate more comparable; the value range of is the larger the value, the higher the reliability, i.e., the lower the uncertainty of the evaluation result.

[0014] In a preferred embodiment, the expected information gain is calculated by the following formula: ; wherein, represents the expected information gain of the indicator ; represents the target nutritional status variable, which is the final nutritional status that needs to be assessed; Indicates the possible nutritional indicators to measure; express The entropy is used to measure the uncertainty of the target variable; Indicates that in the known under conditions The conditional entropy indicates that when the indicator Post-target variable Remaining uncertainty; Express about The mathematical expectation of The weighted average of all possible values; Indicates that through measurement indicators The expected reduction in uncertainty, i.e. the amount of information gained.

[0015] In a preferred embodiment, an AI-based rapid evaluation system for nutritional cases of key customers is used to implement an AI-based rapid evaluation method for nutritional cases of key customers according to any one of claims 1 to 9, characterized in that it includes: The crowd feature processing module is used to process crowd statistical feature data using a meta-learning network to generate a sampling basic model for a specific crowd; The information value assessment module is used to construct a Bayesian decision network to calculate the information value of nutritional indicators and generate the indicator information cost ratio; Adaptive sampling module, which is used to construct a sequential sampling strategy using Markov decision process and perform adaptive indicator sampling; Dynamic termination decision module, used to calculate the information gain rate to realize dynamic termination decision of the sampling process; The evaluation result generation module is used to apply conditional probability inference to generate complete nutritional evaluation results and provide an estimate of the reliability of the results.

[0016] The beneficial effects of the present invention are: A dynamic termination decision-making mechanism enables optimal resource allocation, reducing emergency assessment and processing time from hours to minutes. The sampling strategy based on meta-learning can automatically adapt to the characteristics of different populations, reducing the data requirements for adapting to new populations; By combining conditional probability inference and reliability estimation, it can provide highly accurate assessment results and credibility quantification even under limited sampling conditions, providing comprehensive support for medical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of an AI-based rapid evaluation method for nutritional cases of key customers of the present invention. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a method for rapid evaluation of nutritional cases of key customers based on AI, such as Figure 1 As shown, the following steps are included: Step 1: Use the meta-learning network to process the demographic data of the population and generate a sampling base model for a specific population; The specific steps include: Step 1.1: Collect statistical characteristic data of the target population, including demographic characteristics (such as age distribution and gender ratio), occupational characteristics (such as industry type and nature of work), regional characteristics, historical nutritional status and other multidimensional data.

[0020] Step 1.2, build a meta-learning network; The meta-learning network can learn the mapping relationship between different population characteristics and optimal sampling strategies from historical data.

[0021] The meta-learning network adopts a two-layer structure: The outer network is responsible for learning the mapping function from crowd characteristics to inner network parameters in is the meta-learning mapping function; Represents a population feature vector, including multi-dimensional demographic data such as age distribution, gender ratio, occupational characteristics, and regional characteristics; Represents the initial parameters of the inner network, determines the weight and bias value of the sampling strategy, and is used to guide the subsequent sampling process; The inner network is responsible for implementing the specific sampling strategy and generating sampling strategy parameters for a specific population based on population characteristics, including parameters such as sampling priority, sampling order, and termination conditions for each indicator. These parameters are personalized according to the characteristics of different populations. Step 1.3, train the meta-learning network; The task-based meta-learning paradigm is adopted, and its objective function is: ; in, is the outer network parameter, which controls how the meta-learning network generates the inner network parameters for a specific population; is the number of different population tasks used in training; is the inner network parameter optimized for the first population task, which determines the optimal sampling strategy for that specific population; is the inner network parameter optimized for the first population task, which determines the optimal sampling strategy for that specific population; is the test data of the specific population, used to evaluate the generalization performance of the sampling strategy on unseen data; is the loss function that evaluates the quality of the sampling strategy, quantifying the effectiveness of the sampling strategy; represents the index of a specific population, ranging from 1 to represents the optimization of the outer network parameter to minimize the objective function; represents the summation of the loss functions for all

[0022] Step 1.4, generate a sampling base model for the new population, according to the collected new population characteristics Calculate the initial parameters of the sampling strategy for this population through the trained meta-learning network: ; where, represents the initial parameters of the sampling strategy generated for the new population, including sampling priorities, sampling order, and termination conditions for each indicator; represents the meta-learning mapping function, which converts population characteristics into network parameters suitable for that specific population; represents the feature vector of the new population, including multi-dimensional population statistics such as age distribution, gender ratio, occupation characteristics, and regional characteristics.

[0023] Thus, a sampling base model suitable for the specific population is generated, which will guide the subsequent sampling process.

[0024] The innovation of this step lies in the use of meta-learning technology to achieve rapid adaptation of the sampling strategy to different population characteristics, solving the problem of lack of group difference adaptation ability in traditional methods. In addition, the output sampling base model contains population characteristic information, providing personalized basis for subsequent information value evaluation.

[0025] Step 2, based on the sampling base model, construct a Bayesian decision network to calculate the information value of the nutrition indicators, and generate the information cost ratio, which is defined as the ratio of expected information gain to the product of economic cost and time cost; Specifically, the following steps are included: Step 2.1, construct a Bayesian network model to represent the probabilistic dependence relationship between nutrition indicators; A Bayesian network is a directed acyclic graph in Represents a set of nodes, corresponding to each nutritional indicator, Represents a set of directed edges, corresponding to the conditional dependency between indicators.

[0026] The joint probability distribution of the Bayesian network can be expressed as: ; in, represents the joint probability distribution of all nutritional indicator variables; Respectively represent Nutritional indicator variables, such as vitamin D, calcium, iron and other specific nutritional indicators; Indicates the total number of nutritional indicators; express The parent node set of A set of indicators; Indicates that in a given Under the condition of parent node value The conditional probability distribution of ; Indicates from arrive The product operation of all conditional probabilities reflects the core feature of the Bayesian network that decomposes the joint distribution into the product of conditional probabilities; Step 2.2: Based on historical data and expert knowledge, learn the structure and parameters of the Bayesian network; Network structure learning uses scoring-based methods, such as the Bayesian Information Criterion (BIC) or the Minimum Description Length (MDL) criterion, to determine the optimal structure by evaluating the scores of different network structures. The BIC criterion evaluates network structure by balancing the model log-likelihood with a model complexity penalty, while the MDL criterion minimizes the number of bits required to describe the data based on coding theory. These scoring methods help select the structure that best reflects the true dependencies between nutritional indicators from among many possible network structures. Parameter learning uses maximum likelihood estimation or Bayesian estimation method to learn conditional probability table To accurately characterize the probability dependence relationship between various nutritional indicators, Indicates that in a given indicator The parent node Under the value conditions The conditional probability distribution of the nutritional indicators is obtained by the maximum likelihood estimation, which estimates the parameters by maximizing the likelihood function of the observed data, while the Bayesian estimation introduces a prior distribution to provide more robust estimation when the data is sparse. These learning methods ensure that the Bayesian network can accurately capture the statistical dependencies between nutritional indicators. Step 2.3, calculate the expected information gain (EIG) of each potential sampling indicator to quantify its information contribution to the nutritional status assessment: ; where, represents the expected information gain of indicator , which quantifies the information value that can be obtained by measuring this indicator; represents the target nutritional status variable, which is the final nutritional status that needs to be assessed; represents the th possible measured nutritional indicator; represents the entropy of , which measures the uncertainty of the target variable; represents the conditional entropy of given , which represents the remaining uncertainty of the target variable after observing the indicator ; represents the mathematical expectation of , which is the weighted average of all possible values of ; represents the expected uncertainty reduction by measuring the indicator , which is the amount of information obtained.

[0027] Step 2.4, introduce the information cost ratio (ICR) to consider the information value, economic cost, and time cost of the indicator comprehensively: ; where, represents the information cost ratio of indicator , which quantifies the information value provided by this indicator under unit cost; represents the expected information gain of indicator , which measures the amount of information that can be obtained by measuring this indicator; represents the economic cost of measuring indicator ; represents the time cost of measuring indicator ; represents the comprehensive cost of measuring indicator , which is the product of the economic cost and the time cost.

[0028] The innovation of this step is to introduce the information value theory to quantify the balance between the value and cost of the indicator, solving the problem of resource allocation that traditional methods cannot solve. Therefore, the output information cost ratio of the indicator provides a scientific decision basis for subsequent adaptive sampling.

[0029] Step 3. Constructing a sequential sampling strategy using Markov Decision Process according to the index information cost ratio, and performing adaptive index sampling; Specifically, the following steps are included: Step 3.1. Constructing a Markov Decision Process (MDP) model to formalize the sequential sampling problem as a state transition process; State space The state representing the current set of measured indices and their observations is denoted as: ; where, is the set of measured index indices at time ; denotes the system state at time point ; denotes the th nutritional index variable; denotes the actual observation value of index ; denotes the index set ; ; Action space The set of indices that can be measured next is denoted as: ; where, denotes the set of actions available in state ; denotes the nutritional index that can be measured; denotes the index set ; ; State transition function denotes the probability of transitioning to state after choosing to measure index in state , where, denotes the probability function; denotes the state at the next time point; denotes the current state; denotes the index chosen to measure at time point ; this function describes the dynamic characteristics of how the system state changes with action selection; Reward function reflects the net benefit of measuring index , defined as the information gain of that index minus the measurement cost, where, represents the reward function; represents the current state; represents the selected action (measured indicator); represents the new state after performing the action; this function quantifies the net value obtained by selecting a particular indicator for measurement in the current state, to guide the action selection in the decision-making process; Step 3.2, implement the improved Markov blanket information propagation algorithm to effectively utilize the conditional dependencies between indicators to predict the probability distribution of unmeasured indicators.

[0030] For each indicator define its Markov blanket including its parent nodes, child nodes, and other parent nodes of child nodes; According to the principle of conditional independence, is conditionally independent of other nodes in the network given its Markov blanket, i.e., ; where represents the nutritional indicator variable currently under consideration; represents the set of all indicator variables except ; represents the conditional probability distribution of given all other indicators; represents the Markov blanket of , containing its parent nodes, child nodes, and other parent nodes of child nodes; represents the conditional probability distribution of given only the indicators in its Markov blanket; this equation shows that only the indicator values in the Markov blanket need to be known to completely determine the conditional probability of without considering other indicators in the network; Using this property, when some indicators are measured, the conditional probability distribution of unmeasured indicators can be efficiently updated; this means that without collecting all indicator data, only by obtaining the measurement values of key indicators, accurate inference of other unmeasured indicators can be made, thereby greatly improving the sampling efficiency; Step 3.3, based on the current state and the information cost ratio, select the next optimal measurement indicator: ; where, represents the optimal measurement indicator selected at time point ; represents the parameter that maximizes the following expression represents the action belongs to the current state , i.e., the set of indicators that have not been measured; refers to the conditional information cost ratio of the indicator under the current state , and the calculation formula is: ; wherein, represents the conditional information cost ratio of the indicator under the current state ; represents the conditional expected information gain of measuring the indicator under the condition of knowing the state ; represents the economic cost of measuring the indicator ; represents the time cost of measuring the indicator ; represents the comprehensive cost of measuring the indicator .

[0031] and represents the conditional expected information gain of measuring the indicator under the condition of knowing the state ; wherein, represents the conditional expected information gain of measuring the indicator under the condition of knowing the state , represents the conditional entropy of the target nutritional status variable under the current state , which measures the uncertainty of the nutritional status assessment under the current state; represents the conditional entropy of the target variable after the current state and the indicator are measured; represents the conditional entropy expectation about the possible values of the indicator ; the entire formula quantifies the expected reducible uncertainty of measuring the indicator under the current state.

[0032] Step 3.4, perform measurement and update state, repeat step 3.2 and step 3.3 until the termination condition is met.

[0033] The innovation of this step is to formalize the sequential sampling problem as a Markov decision process, and to realize efficient information dissemination combined with an improved Markov blanket algorithm, ensuring that the indicator with the highest information value is selected for measurement at each step. In addition, through this adaptive sampling strategy, the problem of low efficiency of traditional fixed sampling schemes is solved.

[0034] Step 4, in the process of adaptive index sampling, the information gain ratio is calculated to achieve a dynamic termination decision to avoid over-sampling; Specifically, the following steps are included: Step 4.1, define the information gain ratio (IGR) to measure the marginal information gain brought by each new measurement index: ; Wherein, represents the information gain ratio of the th step, which is used to measure the relative information gain brought by the current measurement index compared with the measurement index of the last step; and respectively represent the measurement indexes of the th and th steps; and respectively represent the states before measuring the indexes and , which contain all the measured indexes and their values at the time points represents the conditional expected information gain brought by the measurement index under the state , which quantifies the amount of information provided by the newly added measurement index in the current step; represents the conditional expected information gain brought by the measurement index under the state , which quantifies the amount of information provided by the measurement index in the last step; the whole ratio reflects the trend of information acquisition efficiency, and when the ratio is less than a certain threshold, it indicates that the marginal benefit of continuing sampling is decreasing, and the sampling process can be terminated; Step 4.2, implement a dynamic termination algorithm based on a threshold, set two termination conditions, and stop sampling when either condition is met.

[0035] The two conditions are: Information gain ratio threshold condition: when , it indicates that the information gain brought by the new measurement index is not economical enough compared with the cost, and the sampling should be stopped, wherein represents the information gain ratio of the th step, which is used to measure the relative information gain brought by the current measurement index compared with the measurement index of the last step; is a pre-set information gain ratio threshold, which represents the lowest acceptable information gain ratio, and a value lower than this means that the marginal benefit of continuing sampling is too low; Evaluation accuracy condition: when the evaluation accuracy estimate value under the current state has reached the pre-set target Time, i.e. sampling can be terminated; wherein, represents the system state at time point , containing the measured indicator set and its observations; represents the estimated evaluation accuracy under the current state , quantifying the accuracy degree of the nutritional evaluation based on the measured indicators; is a preset accuracy threshold, representing the minimum accuracy standard required for evaluation, reaching this value indicates that the evaluation result is reliable enough; Step 4.3, build an evaluation accuracy estimation model to predict the accuracy of the evaluation result in the current state in real time: ; wherein, represents the evaluation accuracy estimation value under the current state , used to predict the accuracy degree of the evaluation result based on the measured indicator set; is the evaluation accuracy estimation function, a mathematical model that maps the current state to the accuracy estimation value; represents the system state at time point , containing all measured indicators and their observations; is a set of parameters of the function, including weights, biases, and other model parameters, which can be obtained by training historical data, used to adjust the model to adapt to the characteristics of different populations and evaluation tasks; Step 4.4, adaptively adjust the termination threshold, dynamically set and according to the urgency of the task and resource constraints; In emergency situations (such as suspected food safety issues), increase and decrease to terminate sampling earlier to quickly obtain results; In resource-abundant and high-accuracy situations, decrease and increase to conduct more comprehensive sampling to improve evaluation accuracy.

[0036] The innovation of this step lies in the introduction of information gain rate concept to realize dynamic termination decision of the sampling process, avoiding resource waste caused by over-sampling, while ensuring that the evaluation result reaches the required accuracy. Therefore, this dynamic termination method is particularly suitable for nutritional evaluation in resource-constrained and emergency situations, and can achieve the optimal balance between resource input and evaluation accuracy.

[0037] Step 5, based on the sampling results after dynamic termination decision, apply conditional probability inference to generate complete nutritional evaluation results, and provide result reliability estimation; Specifically comprising the following steps: Step 5.1, inferring the conditional probability distribution of unmeasured indicators using probabilistic graphical models, based on the state at the time of termination of sampling inferring the complete nutritional indicator distribution; The conditional probability distribution can be expressed as: ; wherein, represents the joint probability distribution of all nutritional indicators under the condition that the state at the time of termination of sampling ; respectively represent the first nutritional indicator variable; represents the total number of nutritional indicators; represents the set of measured indicators, i.e., the set of nutritional indicator variables that have been measured at the time of termination of sampling; represents the set of unmeasured indicators, i.e., the set of nutritional indicator variables that have not been measured at the time of termination of sampling; represents the system state at the time of termination of sampling, containing the specific observation values of the measured indicators; is the observation distribution of the measured indicators, representing the probability distribution of the measured indicators given the state ; is the conditional probability distribution of the unmeasured indicators, representing the probability distribution of the unmeasured indicators given the measured indicator values; the entire formula shows that the joint distribution of the complete indicators can be decomposed into the product form of the measured part and the inferred part based on the measured values; Step 5.2, combining the conditional probability distribution with expert rules to generate the nutritional status assessment result; The nutritional status assessment result can be expressed as a function of the measured indicators and the inferred indicators: ; wherein, represents the final nutritional status assessment result, which is the output of the assessment function; represents the nutritional status assessment function, which is used to map the measured indicators and the inferred indicators to the final assessment result; represents the set of measured indicators, containing all the indicators actually measured during the sampling process and their values; is the inferred value of the unmeasured indicators, representing the estimated value of the unmeasured indicators calculated through the conditional probability model; Step 5.3, calculating the reliability estimate of the assessment result to quantify the uncertainty of the result; The result reliability score is defined as: ; wherein, Represents the reliability score of the evaluation result, and its value range is A larger value indicates a higher reliability of the evaluation results; Is in state Down The normalized conditional entropy is used to quantify the uncertainty of the evaluation results; It represents the nutritional status assessment result, which is the final assessment result generated based on measured and inferred indicators; Represents the system state at the end of sampling, including the observation values ​​of all measured indicators; the normalization process ensures that the value of conditional entropy is mapped to interval, so that the reliability score Also located The lower the uncertainty of the evaluation result, the closer the normalized conditional entropy is to 0, and the reliability score is The closer to 1; Step 5.4: Generate personalized nutrition recommendations based on the assessment results and reliability estimates; Regarding the evaluation results Combined with the expert knowledge base, the various nutritional indicators in the database can be used to identify abnormal indicators and their potential causes; Determine intervention priorities based on the severity and reliability estimates of abnormal indicators; Generate targeted nutritional improvement recommendations and determine the confidence level of the recommendations based on reliability estimates.

[0038] The innovation of this step lies in combining conditional probabilistic inference with expert rules, achieving an efficient mapping from limited sampling to complete assessment. Simultaneously, the uncertainty of the results is quantified through reliability estimation, providing more comprehensive information support for medical decision-making. Furthermore, this combined approach of probabilistic inference and uncertainty quantification addresses the difficulty of traditional assessment methods in ensuring reliability under limited data conditions.

[0039] Application examples of this implementation: According to the embodiments of the present application, the application of the present method is described in detail below through an actual scenario of nutritional status assessment of enterprise employees.

[0040] A technology company needed to conduct nutritional assessments for its 2,000 employees to support its corporate health management program. Traditional methods required each employee to complete 35 nutritional indicators, totaling 70,000 tests, taking an estimated three weeks and costing approximately 1.4 million yuan. The company sought a more efficient way to complete this assessment within limited resources. Implementation process example:

[0041] Meta-learning network application stage: First, collect the demographic characteristics of the employees of this enterprise, including: average age 32.5 years old, male to female ratio 1.6:1, 85% in R&D or technical positions, 60% with a sedentary habit, 12% with irregular work and rest. These characteristic data are represented as a population feature vector ; Through the pre-trained meta-learning network, the initial sampling strategy parameters are generated according to the population characteristics: These parameters guide the initial sampling preferences, such as giving higher initial weights to indicators related to common neck problems and vision-related nutrients for IT practitioners. Among them, represents the initial sampling strategy parameter vector, which contains the sampling priority weights of each nutrient indicator; represents the meta-learning network function, which is used to map the population characteristics to the optimal sampling strategy parameters; represents the population feature vector, which contains the statistical characteristics of the target population such as age, gender ratio, occupation distribution, etc.

[0042] Bayesian network modeling stage: A Bayesian network containing 35 nutrient indicator nodes is constructed, and the edges between nodes represent the dependency relationship between indicators. For example, there is a strong dependency relationship between vitamin D and calcium indicators, and a directed edge from vitamin D to calcium absorption is added in the network.

[0043] For this group of employees, the information cost ratio of each indicator is calculated.

[0044] For example, the EIG value of vitamin B12 is 0.42, EIG represents the expected information gain obtained by measuring this indicator, the detection cost is 80 yuan, and the time cost is 15 minutes, and the ICR is calculated as: ICR represents the information acquisition efficiency per unit of economic and time cost.

[0045] The EIG value of vitamin D is 0.38, EIG represents the expected information gain obtained by measuring the vitamin D indicator, the detection cost is 60 yuan, and the time cost is 10 minutes, and the ICR is calculated as: ICR represents the information acquisition efficiency per unit of economic and time cost.

[0046] According to the calculation results, the system determines the initial priority measurement indicator set.

[0047] Adaptive sampling execution stage: The system starts from the initial state Then based on the Markov decision process model, the optimal next measurement indicator is selected in turn.

[0048] For example, after measuring Basal Metabolic Rate (BMR), Hemoglobin (Hb), and Total Cholesterol (TC), the current state is .

[0049] The system calculates the conditional information cost ratio of the remaining indicators and finds that the ICR value of Vitamin D in the current state is the highest, 0.00085, so it selects Vitamin D as the 4th measurement indicator.

[0050] Dynamic termination decision phase: as the sampling progresses, the system continuously monitors the change in Information Gain Rate (IGR). After completing the measurement of the 14th indicator, it is calculated that is lower than the preset threshold .

[0051] At the same time, the current evaluation accuracy estimate has exceeded the target accuracy satisfies one of the two termination conditions, the system decides to terminate the sampling process.

[0052] Evaluation result generation phase: based on the 14 measured indicators, the system uses the conditional probability model to infer the values of the remaining 21 unmeasured indicators. For example, based on the measured iron, hemoglobin, and transferrin values, the system infers the distribution of ferritin levels and takes its conditional expectation as the estimate.

[0053] For each employee, the system generates a complete evaluation result of 35 indicators and calculates the result reliability score. For example, for an employee, the system calculates the reliability score indicates that the evaluation result has a high degree of credibility.

[0054] Based on the evaluation results, the system identifies the common nutritional problems in the employee group of the enterprise, such as Vitamin D deficiency in 68% and magnesium intake deficiency in 52%, and generates personalized nutrition improvement suggestions. Technical effect verification:

[0055] Sampling efficiency improvement effect: after applying this method, the actual effect data of employee nutrition evaluation is as follows: On average, only 13.4 indicators are measured per person (38.3% of the total indicators), the evaluation accuracy reaches 92.7%, the evaluation time is shortened from the original 3 weeks to 6.5 days (65.2% reduction), and the total cost is reduced to 584,000 yuan (58.3% savings).

[0056] Population adaptability improvement effect: When this method is applied to employees in different departments, it shows good adaptability. The system can automatically adjust the sampling strategy according to the characteristics of employees in different departments, as follows: Comparison of sampling strategies between R&D department and sales department: R&D department: prefer to select vision-related nutrition indicators and sedentary-related metabolic indicators; Sales department: prefer to select energy metabolism indicators and immunity-related indicators; Adapt to the data volume required by the new department: only 15 samples are required (85% less than the traditional method).

[0057] Emergency response effect: In a suspected food safety incident, the system can complete the preliminary evaluation of 100 employees who may have been exposed to the problem food within 30 minutes, with an average of only 8 indicators per person, and accurately identify 12 employees who need further medical intervention. Before the application of this method, it takes 4 to 5 hours to complete the preliminary screening in similar situations.

[0058] As can be seen from the practical application examples, the AI-based large customer nutrition case rapid evaluation method provided by the present application exhibits significant technical effects in actual enterprise health management scenarios, not only greatly improving the evaluation efficiency and reducing resource consumption, but also improving the flexibility and adaptability of the evaluation, especially in resource-limited and emergency situations. The embodiments of the present application have been described above, but the embodiments of the present application are not limited to the specific embodiments described above, which are only illustrative and not limiting. Those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection scope of the embodiments.

Claims

1. A rapid evaluation method for nutritional cases of key customers based on AI, characterized by: The following steps are involved: Use meta-learning networks to process demographic data and generate sampling base models for specific populations; Based on the sampling basic model, a Bayesian decision network was constructed to calculate the information value of nutritional indicators and generate the indicator information cost ratio, which is defined as the ratio of expected information gain to the product of economic cost and time cost. According to the indicator information cost ratio, a sequential sampling strategy is constructed using Markov decision process to perform adaptive indicator sampling; During the adaptive index sampling process, the information gain rate is calculated to achieve dynamic termination decision and avoid oversampling; Based on the sampling results after the dynamic termination decision, conditional probability inference is applied to generate complete nutritional assessment results and provide an estimate of the reliability of the results.

2. The AI-based rapid assessment method for nutritional cases of key customers according to claim 1 is characterized in that: The step of processing the demographic feature data using the meta-learning network includes: Collect statistical data of the target population, including demographic characteristics, occupational characteristics, regional characteristics, historical nutritional status and multi-dimensional data; Construct a meta-learning network. The meta-learning network adopts a two-layer structure. The outer network is responsible for learning the mapping function from crowd characteristics to inner network parameters, and the inner network is responsible for implementing the specific sampling strategy. Train the meta-learning network using a task-based meta-learning paradigm; The initial parameters of the sampling strategy for the new population are calculated through the trained meta-learning network to generate a basic sampling model suitable for the new population.

3. The AI-based rapid assessment method for nutritional cases of key customers according to claim 1 is characterized in that: The step of constructing a Bayesian decision network to calculate the nutritional index information value includes: A Bayesian network model was constructed to represent the probabilistic dependency relationship among nutritional indicators; Learning the structure and parameters of Bayesian networks based on historical data and expert knowledge; The expected information gain of each potential sampling indicator was calculated to quantify the information contribution of the expected information gain to nutritional status assessment; The information cost ratio is introduced to comprehensively consider the information value, economic cost and time cost of the indicator.

4. The AI-based rapid evaluation method for nutritional cases of key customers according to claim 1 is characterized in that: The steps of constructing a sequential sampling strategy using a Markov decision process include: Construct a Markov decision process model and formalize the sequential sampling problem into a state transition process; Implement an improved Markov blanket information propagation algorithm to effectively use the conditional dependencies between indicators to predict the probability distribution of unmeasured indicators; Based on the current state and information cost ratio, select the next optimal measurement indicator; Perform the measurement and update the status.

5. The AI-based rapid evaluation method for nutritional cases of key customers according to claim 4 is characterized in that: The Markov decision process model includes: The state space represents the state of the current set of measured indicators and their observations; Action space, which represents the set of indicators that may be measured in the next step; The state transition function represents the probability of transitioning to the next state after selecting a measurement indicator in one state; The reward function reflects the net benefit brought by the measurement indicator and is defined as the information gain of the measurement indicator minus the measurement cost.

6. The AI-based rapid evaluation method for nutritional cases of key customers according to claim 1 is characterized in that: The step of calculating the information gain rate to realize dynamic termination decision includes: Define the information gain rate to measure the marginal information gain brought by each new measurement indicator; Implement a threshold-based dynamic termination algorithm, set two termination conditions: information gain rate threshold condition and evaluation accuracy condition, and stop sampling when either condition is met; Build an evaluation accuracy estimation model to predict the accuracy of the evaluation results in real time under the current state; Adaptively adjust the termination threshold and dynamically set the termination condition parameters based on the urgency of the task and resource constraints.

7. The AI-based rapid evaluation method for nutritional cases of key customers according to claim 1 is characterized in that: The step of applying conditional probability inference to generate a complete nutritional assessment result includes: The probabilistic graphical model is used to infer the conditional probability distribution of unmeasured indicators and the complete distribution of nutritional indicators is inferred based on the state when sampling is terminated. Combine conditional probability distribution with expert rules to generate nutritional status assessment results; Calculate reliability estimates of assessment results and quantify the uncertainty of the results; Generate personalized nutritional recommendations based on the assessment results and reliability estimates.

8. The AI-based rapid assessment method for nutritional cases of key customers according to claim 7 is characterized in that: The reliability of the evaluation results is estimated by the following formula: ； in, Represents the reliability estimate of the assessment result; Indicates that the status Target nutritional status variable The normalized conditional entropy is used to quantify the uncertainty of the nutritional status assessment results under the current sampling state; Indicates the state when sampling is terminated, including all measured indicators and their observed values; Normalized conditional entropy scales the original conditional entropy to interval, making the reliability estimates more comparable; The value range is A larger value indicates higher reliability, that is, lower uncertainty in the evaluation results.

9. The AI-based rapid evaluation method for nutritional cases of key customers according to claim 1 is characterized in that: The expected information gain is calculated by the following formula: ； in, Indicator The expected information gain of represents the target nutritional status variable, which is the final nutritional status that needs to be assessed; Indicates the possible nutritional indicators to measure; express The entropy is used to measure the uncertainty of the target variable; Indicates that in the known under conditions The conditional entropy indicates that when the indicator Post-target variable Remaining uncertainty; Indicates about The mathematical expectation of The weighted average of all possible values; Indicates that through measurement indicators The expected reduction in uncertainty, i.e. the amount of information gained.

10. An AI-based rapid evaluation system for nutritional cases of key customers, used to implement an AI-based rapid evaluation method for nutritional cases of key customers according to any one of claims 1 to 9, characterized in that: include: The crowd feature processing module is used to process crowd statistical feature data using a meta-learning network to generate a sampling basic model for a specific crowd; The information value assessment module is used to construct a Bayesian decision network to calculate the information value of nutritional indicators and generate the indicator information cost ratio; Adaptive sampling module, which is used to construct a sequential sampling strategy using Markov decision process and perform adaptive indicator sampling; Dynamic termination decision module, used to calculate the information gain rate to realize dynamic termination decision of the sampling process; The evaluation result generation module is used to apply conditional probability inference to generate complete nutritional evaluation results and provide an estimate of the reliability of the results.

Images