Nutrient intake prescription design system based on AI large model

Abstract

The invention relates to the technical field of diet health, and discloses a nutrient intake prescription design system based on an AI large model, and the system comprises a user side which is used for collecting user health data and daily life data, and transmitting the user health data and daily life data to an intelligent nutrition scheme design platform; the intelligent nutrition scheme design platform is used for evaluating the nutrition status of the user based on an AI intelligent medical big model and outputting a personalized nutrition prescription; the nutritionist working platform is used for checking and confirming the output personalized nutrition prescription; and the e-commerce platform is used for providing nutrition commodity matching and purchasing services based on the audited and confirmed personalized nutrition prescription. The multi-source health data is automatically analyzed through the AI intelligent medical large model, the nutrition prescription is generated, the prescription generation time is greatly shortened, and the time is shortened from several hours of traditional manual service to the minute level; and meanwhile, the nutritionist resource dynamic scheduling module reduces the prescription auditing waiting time and improves the overall service efficiency.

Description

Technical Field

[0001] This invention relates to the technical field of dietary health, and more specifically, to a nutrient intake prescription design system based on an AI large model. Background Technology

[0002] Traditional nutrition services suffer from resource concentration in large cities and medical institutions, leaving rural and remote areas often without sufficient professional nutritional support. Due to the limited number of professional nutritionists and the concentration of training resources within urban healthcare systems, ordinary users often struggle to access qualified nutritionists. This leads to significant disparities in nutrition services between urban and rural areas. Current nutrition consultations rely heavily on manual services, resulting in low efficiency, especially when facing a large volume of user requests, impacting both service quality and timeliness. Nutritionists typically need to perform extensive manual data analysis, including but not limited to users' dietary records, medical examination reports, and medical records—a time-consuming process prone to human error. With an aging population and an increase in chronic diseases, the pressure on manual services under the traditional model will further intensify. In traditional nutrition services, nutritionists and health managers often cannot promptly recommend suitable nutritional products or recipes based on users' individual needs. Existing nutritional products and prescriptions are somewhat disconnected; many products have not undergone systematic medical or nutritional validation and lack personalized recommendation mechanisms. The nutritional advice many users receive does not accurately match their individual health conditions or dietary habits. Furthermore, traditional nutritional plans often require multiple doctor visits, physical examinations, and consultations, which is not only time-consuming but also cumbersome for users. For example, users need to schedule an appointment with a doctor, undergo a physical examination, obtain test reports, and then wait for a nutritionist to develop a plan based on the reports. This process usually takes several weeks or even longer, causing users to miss the optimal intervention time. Summary of the Invention

[0003] To address the aforementioned technical problems in related technologies, this invention provides a nutrient intake prescription design system based on an AI large model, which can solve the above problems.

[0004] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: A nutrient intake prescription design system based on an AI large model includes: The user terminal is used to collect user health data and daily life data, and then transmit the user health data and daily life data to the intelligent nutrition solution design platform. The intelligent nutrition solution design platform assesses users' nutritional status based on an AI-powered intelligent medical model and outputs personalized nutrition prescriptions. The nutritionist work platform reviews and confirms the personalized nutrition prescriptions output. E-commerce platforms offer matching and purchasing services for nutritional products based on verified personalized nutritional prescriptions.

[0005] Furthermore, the user terminal supports multiple data input methods, including uploading physical examination reports, connecting wearable devices, and dynamic data tracking.

[0006] Furthermore, the AI-powered intelligent medical model includes a Deepseek base model and an Agent module. The Deepseek base model generates a draft nutrition prescription based on the input user health data and daily life data. The Agent module adjusts the draft nutrition prescription through user preference adaptation to obtain the final nutrition prescription.

[0007] Furthermore, the construction of the Deepseek base model includes the following steps: S100. Multi-source data preparation, including structured data, unstructured data, and medical imaging data. The structured data includes biochemical indicator data from blood test reports and laboratory reports; the unstructured data includes dietary records and exercise logs; and the medical imaging data includes body imaging data related to nutritional status. The structured data is extracted using the CRF medical entity extraction algorithm, the unstructured data is processed using the domain-adapted BERT-BiLSTM dual-channel feature extraction algorithm, and the medical imaging data has features extracted using a ResNet-50-based medical feature distillation network. S200. Based on the prepared multi-source data, the datasets are classified according to different training stages of the model, including pre-training datasets, supervised fine-tuning datasets, and reinforcement learning datasets. The pre-training datasets include medical papers, nutrition guidelines, and case database data. The supervised fine-tuning datasets include paired prescription data containing health data, nutrition prescriptions, and effects. The reinforcement learning datasets include nutritionist correction feedback data and user effect tracking data. S300, phased training and fine-tuning of the model; S310. Pre-train the model based on the pre-trained dataset to build basic medical nutrition knowledge; S320. Supervised fine-tuning of the model based on the supervised fine-tuning dataset to calibrate clinical professional output; S330: Reinforcement learning of the model is performed based on the reinforcement learning dataset, dynamically adapting to real-world feedback scenarios.

[0008] Furthermore, step S310 specifically includes: inputting medical papers, nutrition guidelines, and case database data; using unsupervised learning, the Deepseek base model learns the terminology and knowledge associations in the field of medical nutrition, and outputs an initial model with basic medical nutrition knowledge; step S320 specifically includes: inputting paired prescription data containing health data, nutrition prescriptions, and effects; using supervised learning to adjust model parameters, allowing the model to learn the logic of clinical standard prescriptions, and outputting a model that can generate a draft of a nutrition prescription that conforms to clinical standards; step S330 specifically includes: inputting the correction opinions from nutritionists during review and the effect data after user execution; using a reinforcement learning reward mechanism to optimize the model, giving positive rewards to prescriptions that are recognized by nutritionists and have good user effects, and giving negative rewards to prescriptions that need correction and have poor effects, continuously adjusting the model output weights, and outputting an optimized model that can adapt to real service scenarios.

[0009] Furthermore, the Agent module accesses user-preset personalized tags or mines preferences from historical data, builds a replacement rule base based on nutritional knowledge, adjusts prescriptions, and finally organizes and outputs the adjusted content according to a preset structure to ensure that the prescription is clear and executable.

[0010] Furthermore, the preset structure includes dietary plan content, nutritional supplement content, exercise recommendations content, and monitoring plan content; the dietary plan content clearly defines meal allocation and ingredient list, with ingredient quantities accurate to the gram and providing alternative solutions; the nutritional supplement content indicates the supplement's dosage form, dosage, administration time, and OTC / health food category; the exercise recommendations content includes calorie consumption calculation, specifying the type, intensity, and frequency of exercise; the monitoring plan content includes device reminders, determining re-examination indicators, and the cycle.

[0011] Furthermore, the nutritionist work platform includes a nutritionist management module, which includes nutritionist operation log records, service ratings, and a tiered certification system; the tiered certification system classifies nutritionists into levels based on the number of cases they handle and their assessment results.

[0012] Furthermore, the nutritionist work platform also includes a dynamic scheduling module for nutritionist resources. This module constructs a workload_score scoring model based on professional field matching degree and real-time workload. It quantifies professional level by constructing a nutritionist skill map, evaluates dynamic workload by collecting work status data in real time, and constructs a multi-objective optimization model to output allocation scheme, thereby realizing intelligent allocation of nutritionist resources.

[0013] Furthermore, the e-commerce platform includes a product management module and an inventory management module; the product management module includes e-commerce qualification review, product review, and service rating. The e-commerce qualification review verifies business licenses and production licenses, the product review is achieved through testing reports, random inspections, and expert reviews, and the service rating is determined based on sales records, user reviews, and complaint records; the inventory management module uses an LSTM hybrid model, combining historical sales data, prescription trend analysis, and seasonal factors to generate automatic replenishment suggestions and safety stock warnings.

[0014] The beneficial effects of this invention are: (1) This application automatically analyzes multi-source health data through AI intelligent medical big data model to generate nutritional prescriptions, which greatly shortens the prescription generation time from several hours in traditional manual services to minutes; at the same time, the nutritionist resource dynamic scheduling module reduces the prescription review waiting time and improves the overall service efficiency.

[0015] (2) The model in this application integrates structured, unstructured and medical imaging multi-source data through multi-stage training and reinforcement learning, and combines user preference adaptation to generate prescriptions that are more in line with the individual health needs of users; the introduction of labeled data from tertiary hospitals and review by nutritionists further ensures the professionalism and accuracy of prescriptions.

[0016] (3) The dynamic scheduling module for nutritionist resources in this application achieves a balance between professional field matching and workload, avoids some nutritionists from being overloaded and some nutritionist resources from being idle, and improves the utilization rate of nutritionist resources.

[0017] (4) Users of this application do not need to go to the hospital. They can complete data upload, prescription acquisition and product selection through the APP. The process is convenient. Personalized prescriptions and accurate product recommendations meet the diverse health needs of users. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] The present invention will now be described in further detail with reference to the accompanying drawings.

[0020] Figure 1 This is an architecture diagram of a nutrient intake prescription design system based on an AI large model, as described in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the process of generating a nutritional prescription according to an embodiment of the present invention; Figure 3This is a flowchart of the tiered certification system assessment process described in this embodiment of the invention; Figure 4 This is a flowchart of the inventory management module described in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0022] like Figures 1-4 As shown, this invention discloses a nutrient intake prescription design system based on an AI big data model, comprising: a user terminal for collecting user health data and daily life data, and transmitting the user health data and daily life data to an intelligent nutrition solution design platform; an intelligent nutrition solution design platform for evaluating the user's nutritional status based on an AI intelligent medical big data model and outputting a personalized nutrition prescription; a nutritionist work platform for reviewing and confirming the output personalized nutrition prescription; and an e-commerce platform for providing nutritional product matching and selection services based on the reviewed and confirmed personalized nutrition prescription.

[0023] Example 1: The user terminal in this application is used to collect user health data and daily life data, including: Data uploaded for physical examination reports: such as photo OCR recognition (supports lab reports / physical examination reports), direct PDF file upload (interfacing with hospital / physical examination institution API), and manual supplementary input (secondary confirmation of key indicators).

[0024] Data access for wearable devices: such as health devices (Apple Watch / Huawei Band, etc.), smart body fat scales (brands such as Yunmai / Youpin, etc.), and blood glucose meters (medical-grade devices such as Medtronic / Abbott, etc.).

[0025] Dynamic tracking data: such as diet records (photo recognition + manual correction), exercise data (steps / calories synchronized), and vital sign monitoring (daily weight / sleep quality).

[0026] Example 2: This application's AI-powered intelligent medical model adopts a layered architecture of "basic capability foundation + personalized adaptation agent," including the Deepseek foundation model and the Agent module. The Deepseek foundation model, as the model's "knowledge core," is responsible for integrating general knowledge in the fields of medicine and nutrition, processing the basic logic of multi-source data, and undertaking the core function of generating initial prescription drafts. The core reason for choosing the Deepseek foundation model is its text understanding and logical reasoning capabilities in professional fields, adapting to the complex analytical needs of medical and nutritional data. The Agent module, as the model's "personalization regulator," is responsible for precisely adjusting the initial draft output by the foundation model based on the user's specific needs (such as allergies or dietary preferences), ultimately generating a prescription that matches the user's individual habits, solving the pain point of the "one-size-fits-all" approach of general models.

[0027] Example 3: The construction of the Deepseek pedestal model in this application includes the following steps: (1) Preparation of multi-source data, including structured data, unstructured data and medical imaging data, as shown in the table below: The structured data employs a CRF (Conditional Random Field) medical entity extraction algorithm to accurately extract core information such as biochemical indicator names, values, and units (e.g., extracting "fasting blood glucose: 6.8" from "fasting blood glucose 6.8 mmol / L"), avoiding interference from redundant information.

[0028] The unstructured data uses the domain-adapted BERT-BiLSTM dual-channel feature extraction algorithm. For example, it can be used to identify potential nutritional problems such as "high carbohydrate intake and insufficient dietary fiber" from the example of "eating three meals of rice a day and not liking to eat vegetables".

[0029] The medical image data uses a ResNet-50-based medical feature distillation network to extract key nutrition-related features from the images (such as visual features related to fat distribution and muscle mass), and compresses the data dimensions to adapt to the model's computational needs.

[0030] (2) Based on the prepared multi-source data, the datasets are classified according to different training stages of the model, as shown in the table below: (3) The model is trained and fine-tuned in stages, as follows: (3-1) First stage: Pre-training - Building basic medical nutrition knowledge.

[0031] Input: 5 million medical papers, nutrition guidelines, and case databases.

[0032] Process: Through unsupervised learning, the Deepseek base model learns the terminology and knowledge relationships in the field of medical nutrition (such as "iron deficiency → decreased Hb → need to supplement iron + vitamin C to promote absorption").

[0033] Output: An "initial model" with basic medical nutrition knowledge, which can help understand the logical relationship between health data and nutritional needs.

[0034] (3-2) Second stage: Supervision and fine-tuning - calibrating clinical professional output.

[0035] Input: 200,000 pairs of labeled "health data-nutritional prescription-effect" data from top-tier hospitals (e.g., "Hb 9.8g / dL → recommended duck blood + ferrous sulfate → Hb rose to 11g / dL after 3 months").

[0036] Process: The model parameters are adjusted through supervised learning, allowing the model to learn the "logic of clinical standard prescriptions"—for example, for iron deficiency anemia, it is necessary to match "iron supplementation through food + supplements + monitoring period".

[0037] Output: The model can generate a "prescription draft" that conforms to clinical standards. For example, when iron deficiency data is input, it can output a draft that includes dietary, supplement, and monitoring recommendations.

[0038] (3-3) Third stage: reinforcement learning - dynamically adapting to real-world scenario feedback.

[0039] Input: Two types of real-time feedback data—① Correction opinions during the nutritionist's review (e.g., "Seafood protein in the prescription needs to be replaced with pea protein, the user is allergic"); ② Effect data after the user's implementation (e.g., "After supplementing with iron according to the prescription, ferritin rose to 25 ng / mL").

[0040] Process: The model is optimized based on a "reward mechanism" of reinforcement learning—positive rewards are given to prescriptions that are "approved by nutritionists and have good effects on users," while negative rewards are given to prescriptions that "need to be corrected and have poor effects," and the output weights of the model are continuously adjusted.

[0041] Output: An "optimized model" that can adapt to real-world service scenarios, significantly improving prescription accuracy and user suitability.

[0042] Example 4: The "general prescription draft" generated by the base model in this application needs to be processed by the Agent module to finally output a prescription tailored to the individual needs of the user. The specific process includes the following: (1) User preference identification: The Agent module accesses the user's preset personalized tags (such as "seafood allergy", "vegetarian" and "lactose intolerance"), or mines preferences from historical data (such as "repeatedly rejecting milk-related suggestions → suspected lactose intolerance").

[0043] (2) Prescription adjustment rules: Based on nutritional knowledge, a "replacement rule base" is constructed. For example: vegetarians / seafood allergy sufferers: replace "seafood protein" with "pea protein"; lactose intolerant sufferers: replace "milk" with "lactose-free milk / soy milk".

[0044] (3) Prescription structure standardization: The Agent module organizes the adjusted content according to the preset structure to ensure that the prescription is clear and executable, as shown in the table below: .

[0045] (4) The model construction in this application is not completed in one go. It relies on "full-process service data" to form an iterative closed loop to ensure long-term effectiveness: (4-1) Data input closed loop: user data collection → model generates prescription → nutritionist review → user execution → effect feedback, and the data in the whole process continuously flows back to the reinforcement learning stage.

[0046] (4-2) Model update mechanism: Regularly (e.g., monthly) fine-tune the model based on new feedback data, or supplement specific training data for new scenarios (e.g., "pregnancy nutrition" and "elderly diabetes nutrition") to expand the applicability of the model.

[0047] (4-3) Quality monitoring: Monitor the model's performance through indicators such as "nutritionist approval rate" and "user effect achievement rate". When the indicators are below the threshold (e.g., approval rate < 85%), trigger emergency fine-tuning and data quality verification.

[0048] Example 5: The nutritionist work platform described in this application includes a nutritionist management module, which comprises nutritionist operation logs, service ratings, and a tiered certification system. The operation logs record nutritionists' prescription review records, communication with users, and user evaluations of nutritionists in real time, facilitating traceability and management. Service ratings assess the quality of nutritionist services based on indicators such as review accuracy, user satisfaction, and response speed, providing a reference for user selection. The tiered certification system has four certification levels: intern, junior, senior, and expert. Intern nutritionists must complete basic training, junior nutritionists must handle 100 cases, senior nutritionists must handle 500 cases and pass an assessment, and expert nutritionists must handle 1000 cases and submit relevant professional papers to ensure the professional level of nutritionists.

[0049] Example 6: The nutritionist work platform described in this application also includes a dynamic scheduling module for nutritionist resources. This module constructs a workload_score scoring model based on professional field matching degree and real-time workload. It quantifies professional level by constructing a nutritionist skill map, evaluates dynamic workload by collecting work status data in real time, and constructs a multi-objective optimization model to output allocation scheme, thereby realizing intelligent allocation of nutritionist resources.

[0050] The efficient allocation of nutritionist resources is achieved based on the workload score model. The specific process includes: (1) Construct a skill map of nutritionists: Quantify the professional level of each nutritionist in the fields of diabetes management, pregnancy nutrition, and child nutrition (e.g., "Diabetes management professional level 4").

[0051] (2) Dynamic load assessment: Real-time collection of work status data such as the current workload, number of prescriptions to be reviewed, and processing time of nutritionists to assess their workload.

[0052] (3) Intelligent allocation decision: Construct a multi-objective optimization model that comprehensively considers the matching degree of professional fields and real-time workload. Through components such as allocation strategy engine, real-time feature calculation, and Kafka message queue, output the optimal prescription allocation scheme to ensure that user prescriptions can be quickly matched with suitable nutritionists.

[0053] Example 7: The product management module in this application includes the following functions: Qualification verification: Verify the qualifications of e-commerce companies that join the platform, including their business licenses and production licenses, to ensure that they operate in compliance with regulations.

[0054] Product review: The quality of nutritional products is reviewed through methods such as checking test reports, regular spot checks, and expert evaluation to prevent unqualified products from being put on the shelves.

[0055] Service rating: Based on the e-commerce platform's sales records, user reviews, and complaint records, the e-commerce service is rated, and the rating results are made public to users.

[0056] Product listing management: Standardize the product listing process, requiring products to label information such as ingredients, target population, and methods of consumption to ensure they match prescription needs.

[0057] The inventory management module in this application includes the following functions: it uses an LSTM hybrid model to manage the inventory of nutritional products. The model inputs include historical sales data of products, prescription trend analysis data (such as the increase in prescription recommendations for a certain type of nutritional supplement), and seasonal factors (such as the increased demand for vitamin D in winter). The outputs are automatic replenishment suggestions and safety stock warnings to avoid product shortages or overstocking.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A nutrient intake prescription design system based on an AI large model, characterized in that, include: The user terminal is used to collect user health data and daily life data, and then transmit the user health data and daily life data to the intelligent nutrition solution design platform. The intelligent nutrition solution design platform assesses users' nutritional status based on an AI-powered intelligent medical model and outputs personalized nutrition prescriptions. The nutritionist work platform reviews and confirms the personalized nutrition prescriptions output. E-commerce platforms offer matching and purchasing services for nutritional products based on verified personalized nutritional prescriptions.

2. The nutrient intake prescription design system based on an AI large model according to claim 1, characterized in that, The user terminal supports multiple data input methods, including uploading physical examination reports, connecting wearable devices, and dynamic data tracking.

3. The nutrient intake prescription design system based on an AI large model according to claim 1, characterized in that, The AI-powered intelligent healthcare model includes the Deepseek base model and the Agent module. The Deepseek base model generates a draft nutrition prescription based on the input user health data and daily life data. The Agent module adjusts the draft nutrition prescription through user preference adaptation to obtain the final nutrition prescription.

4. The nutrient intake prescription design system based on an AI large model according to claim 3, characterized in that, The construction of the Deepseek base model includes the following steps: S100. Multi-source data preparation, including structured data, unstructured data, and medical imaging data. The structured data includes biochemical indicator data from blood test reports and laboratory reports; the unstructured data includes dietary records and exercise logs; and the medical imaging data includes body imaging data related to nutritional status. The structured data is extracted using the CRF medical entity extraction algorithm, the unstructured data is processed using the domain-adapted BERT-BiLSTM dual-channel feature extraction algorithm, and the medical imaging data has features extracted using a ResNet-50-based medical feature distillation network. S200. Based on the prepared multi-source data, the datasets are classified according to different training stages of the model, including pre-training datasets, supervised fine-tuning datasets, and reinforcement learning datasets. The pre-training datasets include medical papers, nutrition guidelines, and case database data. The supervised fine-tuning datasets include paired prescription data containing health data, nutrition prescriptions, and effects. The reinforcement learning datasets include nutritionist correction feedback data and user effect tracking data. S300, phased training and fine-tuning of the model; S310. Pre-train the model based on the pre-trained dataset to build basic medical nutrition knowledge; S320. Supervised fine-tuning of the model based on the supervised fine-tuning dataset to calibrate clinical professional output; S330: Reinforcement learning of the model is performed based on the reinforcement learning dataset, dynamically adapting to real-world feedback scenarios.

5. The nutrient intake prescription design system based on an AI large model according to claim 4, characterized in that, Step S310 specifically includes: inputting medical papers, nutrition guidelines, and case database data; using unsupervised learning, the Deepseek base model learns the terminology and knowledge relationships in the field of medical nutrition, and outputs an initial model with basic medical nutrition knowledge; Step S320 specifically includes: inputting paired prescription data containing health data, nutrition prescriptions, and effects; using supervised learning, adjusting model parameters to allow the model to learn the logic of clinical standard prescriptions, and outputting a model that can generate a draft of a nutrition prescription that conforms to clinical standards; Step S330 specifically includes: inputting the correction opinions from nutritionists during review and the effect data after user execution; using a reinforcement learning reward mechanism to optimize the model, giving positive rewards to prescriptions that are recognized by nutritionists and have good user effects, and giving negative rewards to prescriptions that need correction and have poor effects, continuously adjusting the model output weights, and outputting an optimized model that can adapt to real service scenarios.

6. The nutrient intake prescription design system based on an AI large model according to claim 3, characterized in that, The Agent module accesses user-preset personalized tags or mines preferences from historical data, builds a replacement rule base based on nutritional knowledge, adjusts prescriptions, and finally organizes and outputs the adjusted content according to a preset structure to ensure that the prescription is clear and executable.

7. A nutrient intake prescription design system based on an AI large model according to claim 6, characterized in that, The preset structure includes dietary plan content, nutritional supplement content, exercise suggestion content, and monitoring plan content; The dietary plan clearly defines the meal allocation and ingredient list, with ingredient quantities accurate to the gram and providing alternative options. The nutritional supplement information includes the supplement's dosage form, dosage, administration time, and OTC / health food category; the exercise recommendations include calorie consumption calculations, specifying the type, intensity, and frequency of exercise; and the monitoring plan includes device reminders, determination of re-examination indicators, and cycles.

8. The nutrient intake prescription design system based on an AI large model according to claim 1, characterized in that, The nutritionist work platform includes a nutritionist management module, which includes nutritionist operation logs, service ratings, and a tiered certification system. The tiered certification system is divided into levels based on the number of cases handled by the nutritionist and the assessment results.

9. A nutrient intake prescription design system based on an AI large model according to claim 1, characterized in that, The nutritionist work platform also includes a dynamic scheduling module for nutritionist resources. This module constructs a workload_score scoring model based on professional field matching degree and real-time workload. It quantifies professional level by constructing a nutritionist skill map, evaluates dynamic workload by collecting work status data in real time, and constructs a multi-objective optimization model to output allocation scheme, thereby realizing intelligent allocation of nutritionist resources.

10. A nutrient intake prescription design system based on an AI large model according to claim 1, characterized in that, The e-commerce platform includes a product management module and an inventory management module. The product management module includes e-commerce qualification review, product review, and service rating. The e-commerce qualification review verifies business licenses and production licenses. The product review is achieved through testing reports, random inspections, and expert reviews. The service rating is determined based on sales records, user reviews, and complaint records. The inventory management module uses an LSTM hybrid model, combining historical sales data, prescription trend analysis, and seasonal factors to generate automatic replenishment suggestions and safety stock warnings.

Images