Fresh milk category intelligent recommendation and automatic resuming system and method thereof

Abstract

The invention discloses a fresh milk category intelligent recommendation and automatic resuming system and method, and relates to the technical field of big data analysis and personalized services, and the system comprises a user portrait construction module which is used for collecting and storing the personal information, historical order data, drinking frequency, milk taking time and health preference of a user, and the user portrait construction module also analyzes the user behavior pattern through a machine learning algorithm to generate a dynamically updated user portrait. According to the application, precise, adaptive and family-collaborative fresh milk management is realized through multi-module collaboration, the user portrait module dynamically mines behavior rules through an improved clustering algorithm, and the health index module quantifies nutritional requirements by using a neural network, so that the scientificity of a recommendation strategy is improved; the intelligent recommendation module is combined with a multi-tower deep network and a diversity guarantee mechanism to balance personalization and exploratory performance, and the automatic resuming module optimizes a threshold value through time sequence prediction and reinforcement learning, so that the operation cost is reduced.

Description

Technical Field

[0001] This application relates to the technical field of big data analysis and personalized services, and in particular to a system and method for intelligent recommendation and automatic renewal of fresh milk products. Background Technology

[0002] Traditional fresh milk delivery services typically rely on manual ordering or fixed-cycle delivery, lacking dynamic responses to users' personalized needs and health goals. Existing systems rely heavily on static information in user profiling, making it difficult to capture the cyclical patterns of milk consumption habits and their correlation with health preferences. Recommendation strategies are often based on a single dimension of historical preferences, failing to combine real-time health indicators for nutritional matching. In family scenarios, when the needs of different members conflict, there is a lack of multi-objective optimization mechanisms, which can easily lead to inventory imbalances or decreased satisfaction. Furthermore, inventory monitoring and renewal triggers often use fixed thresholds, which cannot adapt to seasonal consumption fluctuations, resulting in stockouts or redundant inventory problems. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a smart recommendation and automatic renewal system and method for fresh milk products.

[0004] The intelligent recommendation and automatic renewal system and method for fresh milk products provided in this application adopts the following technical solution: A smart recommendation and automatic renewal system for fresh milk products, including: The user profile building module is used to collect and store users' personal information, historical order data, drinking frequency, milk collection time and health preferences. The user profile building module also uses machine learning algorithms to analyze user behavior patterns to generate dynamically updated user profiles. A health index calculation module is communicatively connected to the user profile building module. The health index calculation module regularly updates the health index based on the user's metabolic rate and activity level data and dynamically adjusts the type and amount of fresh milk in the recommendation strategy. The intelligent recommendation module is communicatively connected to the health index calculation module. The intelligent recommendation module matches fresh milk subcategories that meet the user's milk drinking habits and health needs based on the user profile and health index, and generates personalized recommendation results. An automatic renewal trigger module is communicatively connected to the intelligent recommendation module. The automatic renewal trigger module is used to monitor the user's inventory level. When the inventory is lower than a preset threshold, the automatic renewal trigger module automatically triggers a renewal order based on the user's consumption patterns and prioritizes recommending fresh milk products that are consistent with the user's historical preferences.

[0005] As a preferred technical solution of this application, the user profile construction module includes a data acquisition unit and a behavior pattern analysis unit. The data acquisition unit is used to collect users' milk collection time preferences, order change frequency and health tags in real time. The behavior pattern analysis unit has a built-in clustering algorithm. The behavior pattern analysis unit uses the clustering algorithm to identify the periodic patterns of users' milk drinking habits and associate them with health preferences to build a model.

[0006] As a preferred technical solution of this application, the health index calculation module integrates a recommendation strategy, which specifically involves updating the metabolic rate and activity level based on user input or wearable device data, mapping the health index to fresh milk nutritional parameters, and optimizing the priority of recommended product categories.

[0007] As a preferred technical solution of this application, the automatic renewal triggering module includes a consumption rate prediction unit and an elastic threshold adjustment unit. The consumption rate prediction unit predicts the user's milk consumption rate based on historical drinking frequency and inventory change trends. The elastic threshold adjustment unit dynamically adjusts the inventory threshold according to the user's seasonal or periodic demand changes.

[0008] As a preferred technical solution of this application, the output end of the user profile construction module is communicatively connected to a family profile fusion module. The family profile fusion module is used to identify and associate multiple user profiles under the same delivery address and construct a unified family profile. The family profile fusion module analyzes the health index, milk consumption preferences and real-time inventory consumption conflicts of each member, uses a multi-objective optimization algorithm to dynamically generate a family-level fresh milk recommendation combination scheme, and automatically allocates the fresh milk category delivered daily based on member priority and milk consumption time difference, in order to achieve the goal of minimizing the total family inventory cost and meeting the personalized needs of all members.

[0009] As a preferred technical solution of this application, the intelligent recommendation and automatic renewal method for fresh milk products includes the following steps: Step 1: Collect users' historical order data, milk collection time, health preferences, and real-time activity data to build user profiles; Step two: Calculate the health index based on the user's metabolic rate and activity level, and dynamically generate a nutritional requirement model; Step 3: Combining user profiles and health indices, recommend suitable fresh milk categories using a multi-dimensional matching algorithm; Step 4: Identify and associate multiple users under the same delivery address, build a family profile, and generate a family-level fresh milk combination delivery plan that meets the needs of all members based on a multi-objective optimization algorithm; Step 5: Monitor user inventory in real time. When the inventory falls below the elastic threshold based on the consumption rate prediction, automatically trigger the renewal and update the recommendation strategy synchronously.

[0010] As a preferred technical solution of this application, in step one, the construction of the user profile specifically involves extracting the user's milk collection time pattern and order interval features through time series analysis, and using a collaborative filtering algorithm to supplement the user's potential health preferences and enhance the completeness of the profile. In step two, the calculation of the health index specifically involves calibrating the nutritional requirement weights based on the user's physiological data and dynamic activity data, matching the health index with the nutritional composition table of fresh milk, and generating a personalized nutritional matching score.

[0011] As a preferred technical solution of this application, in step four, generating a family-level fresh milk combination delivery plan specifically involves analyzing the health needs and preference conflicts of each member, taking the highest overall family satisfaction and the lowest total inventory cost as optimization objectives, and calculating the optimal category combination and the delivery time series of each category. In step five, the automatic renewal is triggered by predicting the remaining inventory days based on a regression model of the user's historical consumption rate. When the predicted remaining inventory days are lower than the user's usual safety threshold, a renewal order is automatically generated and related product categories are recommended.

[0012] In summary, this application includes at least the following beneficial technical effects of a smart recommendation and automatic renewal system and method for fresh milk products: This application achieves precise, adaptive, and family-coordinated fresh milk management through multi-module collaboration. The user profiling module dynamically mines behavioral patterns through improved clustering algorithms, the health index module uses neural networks to quantify nutritional needs and improve the scientific nature of recommendation strategies, the intelligent recommendation module combines multi-tower deep networks and diversity guarantee mechanisms to balance personalization and exploration, the automatic renewal module optimizes thresholds through time-series prediction and reinforcement learning to reduce operating costs, and the family profile fusion module resolves demand conflicts through multi-objective optimization algorithms to achieve optimal allocation of family-level resources. The overall system effectively improves user satisfaction, reduces inventory waste, and reduces manual intervention costs through automation. Attached Figure Description

[0013] Figure 1 This is the overall system architecture diagram of this application; Figure 2 This is a flowchart of the intelligent recommendation and automatic renewal method for fresh milk products in this application. Detailed Implementation

[0014] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0015] See Figure 1-2 A smart recommendation and automatic renewal system for fresh milk products, including: The user profile building module collects and stores users' personal information, historical order data, drinking frequency, milk collection time, and health preferences. It also uses machine learning algorithms to analyze user behavior patterns and generate dynamically updated user profiles. The module includes a data acquisition unit and a behavior pattern analysis unit. The data acquisition unit collects users' milk collection time preferences, order change frequency, and health tags in real time. The behavior pattern analysis unit incorporates a clustering algorithm to identify periodic patterns in users' milk drinking habits and model them in relation to health preferences.

[0016] The user profile building module acquires multi-dimensional user data streams in real time through a distributed data acquisition interface, including order records received via RESTful API, time-series behavioral data uploaded by sensors, and tag information actively submitted by clients. The data standardization unit cleans and normalizes heterogeneous data to form structured user profiles. The behavior analysis engine uses an improved OPTICS clustering algorithm to perform density clustering on time-series data, automatically identify periodic behavior patterns, and establish a mapping relationship between behavioral features and preference tags through association rule mining. The profile updater dynamically adjusts feature weights based on new data within a sliding time window and uses an exponentially weighted moving average algorithm to update the user profile vector, ensuring the real-time nature and continuity of the profile.

[0017] The health index calculation module communicates with the user profile building module. The health index calculation module regularly updates the health index based on the user's metabolic rate and activity level data, and dynamically adjusts the type and quantity of fresh milk in the recommendation strategy. The health index calculation module integrates a recommendation strategy, which specifically updates the metabolic rate and activity level based on user input or wearable device data, maps the health index to the nutritional parameters of fresh milk, and optimizes the priority of recommended categories.

[0018] The health index calculation module adopts a multilayer perceptron neural network architecture. The input layer receives standardized physiological parameters and activity data, the hidden layer contains three fully connected layers that process different feature combinations, and the output layer generates a 0-1 health index through a sigmoid activation function. In this application, principal component analysis is used to reduce dimensionality and eliminate the effects of multicollinearity. When the recommendation strategy engine maps the health index to the nutritional parameter space, it uses a radial basis function neural network for nonlinear fitting to establish the correspondence between health indicators and nutritional needs. The strategy optimizer dynamically adjusts the parameters of the mapping function based on the gradient descent algorithm to minimize the difference between the recommendation result and the user's actual choice.

[0019] The intelligent recommendation module communicates with the health index calculation module. Based on user profiles and health indices, the intelligent recommendation module matches fresh milk subcategories that meet the user's milk drinking habits and health needs and generates personalized recommendation results.

[0020] The intelligent recommendation module employs a multi-tower deep neural network model, constructing user profile towers, health index towers, and product feature towers for feature cross-referencing. A comparative learning loss function ensures that products with similar user preferences and health needs are clustered in the embedding space. The ranking stage uses the LambdaMART algorithm, comprehensively considering features such as health matching degree, historical preference conformity, and inventory availability. Recommendation scores are generated through gradient boosting decision trees. The diversity guarantee mechanism controls the difference between products in the result set through the maximum marginal relevance algorithm, avoiding excessive homogenization of recommendation results.

[0021] The automatic renewal trigger module communicates with the intelligent recommendation module. It monitors user inventory levels and, when inventory falls below a preset threshold, automatically triggers renewal orders based on user consumption patterns, prioritizing fresh milk products consistent with the user's historical preferences. The automatic renewal trigger module includes a consumption rate prediction unit and a flexible threshold adjustment unit. The consumption rate prediction unit predicts the user's milk consumption rate based on historical drinking frequency and inventory change trends, while the flexible threshold adjustment unit dynamically adjusts the inventory threshold according to seasonal or periodic changes in user demand.

[0022] The automatic renewal trigger module is based on a time series forecasting architecture. It uses a Seq2Seq model with attention mechanism to analyze historical consumption data. The encoder encodes the variable-length consumption sequence into a hidden state, and the decoder outputs future multi-step consumption predictions. The elastic threshold calculation unit adopts a reinforcement learning framework to model inventory management as a Markov decision process. It learns the optimal threshold strategy through the Q-learning algorithm. The order generator uses a templated configuration system to automatically combine product categories and quantities based on the prediction results and drive the subsequent payment and logistics processes through the workflow engine.

[0023] The user profile building module's output is connected to the family profile fusion module. The family profile fusion module is used to identify and associate multiple user profiles under the same delivery address and build a unified family profile. The family profile fusion module analyzes each member's health index, milk consumption preferences, and real-time inventory consumption conflicts, and uses a multi-objective optimization algorithm to dynamically generate a family-level fresh milk recommendation combination scheme. It also automatically allocates the fresh milk category for daily delivery based on member priority and milk consumption time difference, in order to achieve the goal of minimizing the total family inventory cost and meeting the personalized needs of all members.

[0024] The family profile fusion module adopts a multi-agent collaborative optimization framework, modeling family members as independent agents. It resolves demand conflicts through a distributed constraint optimization algorithm, formalizes the resource allocation problem into a mixed-integer linear programming model, and optimizes both total satisfaction and operating costs in the objective function. The combination solution generator uses the NSGA-II multi-objective evolutionary algorithm to solve for the Pareto optimal solution set and selects the final solution through fuzzy comprehensive evaluation. The delivery scheduler is based on a spatiotemporal conflict detection algorithm to allocate appropriate delivery time windows to different members, ensuring the dynamic balance of family-level inventory.

[0025] The data standardization unit processes heterogeneous data according to predefined business rules: for order records, features such as product SKU, fat content, and volume are extracted and transformed into multi-dimensional vectors; for milk collection timestamps uploaded by sensors, they are aggregated into frequency distribution histograms at the daily / weekly granularity; and for tags submitted by clients, they are mapped to a unified health tag system. The improved OPTICS clustering algorithm used by the behavior analysis engine is improved by introducing dynamic time warping distance as a metric for density calculation to better handle the elastic period of user milk collection time series. After clustering, the algorithm automatically identifies the period length and phase shift and uses the FP-Growth association rule algorithm to mine strong rules such as "80% of users who collect milk every Wednesday and Friday also prefer high-protein fresh milk," thereby establishing a mapping between behavior and preference. In the exponentially weighted moving average algorithm used by the profile updater, the weight factor λ of recent data is dynamically set according to data freshness. For example, λ=0.7 for data within 7 days and λ=0.3 for data from 7 to 30 days, ensuring that the profile can quickly respond to the latest changes without excessive fluctuations.

[0026] The standardized physiological parameters received by the input layer of the multilayer perceptron in the health index calculation module include basal metabolic rate, daily average steps, and resting heart rate. The three fully connected layers of the hidden layer are set to 64, 32, and 16 nodes respectively, and the ReLU activation function is used for nonlinear transformation to capture the combined effect of features of different granularities. When using principal component analysis to eliminate multicollinearity, the cumulative contribution rate threshold for preserving variance is set to 95% to determine the number of principal components. When the recommendation strategy engine uses radial basis function neural network for nonlinear fitting, the basis function of its hidden layer is a Gaussian function, and the center point is learned from historical health index data through K-means clustering. The linear combination weights of the output layer map the health index to specific nutritional parameters. The strategy optimizer uses the stochastic gradient descent algorithm to dynamically adjust the output weights of the RBFN to minimize the cosine similarity loss between the user's final selected product and the recommended product on the nutritional parameter vector.

[0027] In the multi-tower deep neural network of the intelligent recommendation module, the input to the user profile tower is the user profile vector after standardization and embedding processing; the input to the health index tower is the health index and its derived features; and the input to the product feature tower is the embedding vector of attributes such as nutritional components, taste, and brand of fresh milk products. The feature cross-interaction stage uses the classic attention mechanism to calculate the comprehensive matching degree of user-product pairs. In the ranking stage, the gradient boosting decision tree model used by the LambdaMART algorithm has its input features engineered to include health matching degree, historical preference conformity, real-time inventory availability, and product freshness. To ensure diversity, the maximum marginal relevance algorithm, when selecting recommended products in each iteration, not only considers the relevance score with the user, but also calculates the average cosine distance between the product and the selected product set in the feature space of category, taste, and brand. The next product is selected by setting a weighted sum of relevance weight α=0.7 and diversity weight β=0.3 to avoid too many homogeneous products in the list.

[0028] The Seq2Seq prediction model of the automatic renewal trigger module uses a bidirectional LSTM encoder to capture the dependencies between historical consumption sequences, and a unidirectional LSTM decoder with an additive attention mechanism to dynamically focus on different parts of the input sequence during decoding, outputting a predicted daily consumption sequence for the next 7 days. The elastic threshold adjustment unit models inventory management as a Markov decision process: the state space is defined as (current inventory level, predicted consumption sequence, seasonal indicator), the action space is the threshold adjustment range (e.g., -1, 0, +1), and the reward function combines the reward for avoiding stockouts (inventory > 0) and the deduction... The penalty for low redundant inventory (inventory holding cost) is addressed by using the Q-learning algorithm, where the system learns the optimal threshold strategy online. In the family profile fusion module, the objective function of the mixed-integer linear programming model for resource allocation is specifically defined as: maximizing the sum of satisfaction among all members while minimizing the total delivery cost. Constraints include the minimum daily consumption of each member, the upper limit of the total family inventory, and non-overlapping delivery time windows. The NSGA-II algorithm is used to solve for the Pareto fronts of the two objectives of satisfaction and cost. The final solution is selected from the non-dominated solution set by the fuzzy comprehensive evaluation method based on preset preference weights.

[0029] The intelligent recommendation and automatic renewal method for fresh milk products includes the following steps: Step 1: Collect users' historical order data, milk collection time, health preferences, and real-time activity data to build user profiles. In Step 1, the construction of user profiles specifically involves extracting the patterns of users' milk collection time and order interval characteristics through time series analysis, and using collaborative filtering algorithms to supplement users' potential health preferences to enhance the completeness of the profiles. Step 2: Calculate the health index based on the user's metabolic rate and activity level, and dynamically generate a nutritional requirement model. In Step 2, the calculation of the health index specifically involves calibrating the nutritional requirement weights based on the user's physiological data and dynamic activity data, matching the health index with the nutritional composition table of fresh milk, and generating a personalized nutritional matching score. Step 3: Combining user profiles and health indices, recommend suitable fresh milk categories using a multi-dimensional matching algorithm; Step 4: Identify and associate multiple users under the same delivery address, construct a family profile, and generate a family-level fresh milk combination delivery plan that meets the needs of all members based on a multi-objective optimization algorithm. In Step 4, generating a family-level fresh milk combination delivery plan specifically involves analyzing the health needs and preference conflicts of each member, taking the highest overall family satisfaction and the lowest total inventory cost as optimization objectives, and calculating the optimal category combination and the delivery time series of each category. Step 5: Monitor user inventory in real time. When the inventory is lower than the elastic threshold predicted based on the consumption rate, automatically trigger the renewal and update the recommendation strategy. In step 5, the automatic renewal trigger condition is that the remaining inventory days are predicted based on the regression model of the user's historical consumption rate. When the predicted number of days is lower than the user's usual safety threshold, a renewal order is automatically generated and related categories are recommended.

[0030] This application achieves precise, adaptive, and family-coordinated fresh milk management through multi-module collaboration. The user profiling module dynamically mines behavioral patterns through improved clustering algorithms, the health index module uses neural networks to quantify nutritional needs and improve the scientific nature of recommendation strategies, the intelligent recommendation module combines multi-tower deep networks and diversity guarantee mechanisms to balance personalization and exploration, the automatic renewal module optimizes thresholds through time-series prediction and reinforcement learning to reduce operating costs, and the family profile fusion module resolves demand conflicts through multi-objective optimization algorithms to achieve optimal allocation of family-level resources. The overall system effectively improves user satisfaction, reduces inventory waste, and reduces manual intervention costs through automation.

[0031] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A system for intelligent recommendation and automatic renewal of fresh milk categories, characterized in that: The application relates to a milk recommendation system, which comprises the following modules: a user portrait construction module, a health index calculation module, an intelligent recommendation module and an automatic renewal triggering module. The user portrait construction module is used for collecting and storing personal information, historical order data, drinking frequency, milk taking time and health preferences of a user, and the user portrait construction module further generates a dynamically updated user portrait by analyzing user behavior patterns through a machine learning algorithm. The health index calculation module is in communication connection with the user portrait construction module, the health index calculation module regularly updates a health index according to metabolism rate and activity level data of the user and dynamically adjusts fresh milk types and quantities in a recommendation strategy. The intelligent recommendation module is in communication connection with the health index calculation module, the intelligent recommendation module matches fresh milk sub-categories that meet the milk drinking habits and health needs of the user based on the user portrait and the health index and generates an individualized recommendation result. The automatic renewal triggering module is in communication connection with the intelligent recommendation module, the automatic renewal triggering module is used for monitoring a user inventory, and when the inventory is lower than a preset threshold, the automatic renewal triggering module automatically triggers a renewal order and preferentially recommends fresh milk categories consistent with historical preferences of the user in combination with consumption rules of the user.

2. The fresh milk category intelligent recommendation and automatic renewal system according to claim 1, characterized in that: The user portrait construction module comprises a data collection unit and a behavior pattern analysis unit, the data collection unit is used for collecting user milk taking time preferences, order replacement frequency and health labels in real time, the behavior pattern analysis unit is internally provided with a clustering algorithm, and the behavior pattern analysis unit identifies periodic rules of milk drinking habits of the user and models the health preferences in association.

3. The fresh milk category intelligent recommendation and automatic renewal system according to claim 1, characterized in that: The health index calculation module internally integrates a recommendation strategy, the recommendation strategy is specifically as follows: metabolism rate and activity level are updated according to user input or wearable device data, a health index is mapped to fresh milk nutrition parameters, and a recommended category priority is optimized.

4. The fresh milk category intelligent recommendation and automatic renewal system according to claim 1, characterized in that: The automatic renewal triggering module comprises a consumption speed prediction unit and an elastic threshold adjustment unit, the consumption speed prediction unit predicts the milk consumption speed of the user based on historical drinking frequency and inventory change trends, and the elastic threshold adjustment unit dynamically adjusts the inventory threshold according to seasonal or periodic demand changes of the user.

5. The fresh milk category intelligent recommendation and automatic renewal system according to claim 1, characterized in that: An output end of the user portrait construction module is in communication connection with a family portrait fusion module, the family portrait fusion module is used for identifying and associating multiple user portraits under a same delivery address and constructing a unified family portrait, the family portrait fusion module analyzes health indexes, milk drinking preferences and real-time inventory consumption conflicts of each member, dynamically generates a family-level fresh milk recommendation combination scheme through a multi-objective optimization algorithm and automatically allocates fresh milk categories for daily delivery based on member priorities and milk drinking time differences, so that the goal of minimizing the total inventory cost of a family and meeting individualized needs of all members is achieved.

6. The method of claim 1, wherein the method is applied to the system of any one of claims 1-5. The application further discloses a milk recommendation method, which comprises the following steps: Step one: collecting historical order data, milk taking time, health preferences and real-time activity data of a user to construct a user portrait; Step two: calculating a health index according to metabolism rate and activity level of the user to dynamically generate a nutrition demand model; Step three: combining the user portrait and the health index to recommend suitable fresh milk categories through a multi-dimensional matching algorithm. Step four, identify and associate multiple users under the same delivery address, build family portrait, generate family-level fresh milk combination delivery scheme based on multi-objective optimization algorithm to meet the needs of all members; Step five, real-time monitoring of user inventory, when the inventory is lower than the elastic threshold predicted based on consumption speed, automatically trigger the renewal and update the recommended strategy synchronously.

7. The method of claim 6, wherein the method further comprises: In step one, the construction of user portrait is specifically, through time series analysis to extract the user's milk taking time regularity and order interval characteristics, use collaborative filtering algorithm to supplement the user's potential health preference, enhance the portrait integrity; In step two, the calculation of health index is specifically, according to the user's physiological data and dynamic activity data to calibrate the nutritional demand weight, match the health index with the fresh milk nutrient composition table, and generate the personalized nutrition matching degree score.

8. The method of claim 6, wherein the method further comprises: In step four, the generation of family-level fresh milk combination delivery scheme is specifically, analyzing the conflict between the health needs and preferences of each member, taking the highest total satisfaction and the lowest total inventory cost as the optimization goal, calculating the optimal category combination and the delivery time sequence of each category; In step five, the triggering condition of automatic renewal is, based on the regression model of user historical consumption speed to predict the remaining inventory available days, when the predicted available days is lower than the safety threshold of user habit, automatically generate the renewal order and recommend the associated categories.

Images