An AI-Driven Product Service Recommendation Method and System

Abstract

The present invention discloses an AI-driven product service recommendation method and system, which relates to the field of intelligent healthcare. By constructing a personalized portrait, comprehensively mining product service data, and a full-process automatic learning framework, the relationship between user needs and product service characteristics can be captured more accurately, thereby providing more accurate recommendation results; by constructing a personalized portrait, the diversified needs of users are deeply explored, and personalized product service recommendations are provided according to these needs, thereby improving user satisfaction; the full-process automatic learning framework extracts feature information from the user personalized portrait and the product service data warehouse, and conducts model training and validation tuning. The multi-dimensional full-process automatic learning method enables the recommendation model to continuously learn and optimize; the problem solved is that the existing recommendation methods cannot meet the diversified needs of users. The present invention can accurately meet the diversified needs of users, improve the recommendation accuracy and user experience.

Description

Technical Field

[0001] The present invention relates to the field of intelligent healthcare, and particularly to an AI-driven product service recommendation method and system. Background Art

[0002] Currently, Patent CN113468417A discloses a product service recommendation method and an intelligent healthcare system based on intelligent healthcare big data. By medical consultation behavior, target consultation interest entries are determined, and medical service content is formed based on these entries and citation intensity data. However, the existing technology mainly relies on target consultation interest entries to form an intelligent healthcare interest service set, and cannot comprehensively cover all relevant medical service content. Because patients' needs are diverse, only considering the target consultation interest entries and their corresponding citation intensity data leads to the lack of diversity and comprehensiveness of the recommendation results, and cannot meet the diverse needs of patients.

[0003] Therefore, an AI-driven product service recommendation method and system are needed to solve the above problems. Based on user medical health data and intelligent healthcare product service data, a personalized user portrait is constructed and potential associations and patterns in medical data are mined to discover more medical service content related to patients' needs, making the recommended content more diverse and better meeting the diverse needs of patients. Also, a full-process automatic learning framework is used to update and optimize the model in real time to adapt to the changing user needs and product service characteristics, improving the efficiency of the recommendation system and the user satisfaction. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention discloses an AI-driven product service recommendation method and system, which improve the user experience, data processing efficiency and automation degree, and can provide personalized product service recommendations that better meet the needs and preferences of users, with high automation and intelligence.

[0005] The present invention adopts the following technical solutions:

[0006] An AI-driven product service recommendation method, comprising the following steps:

[0007] Construct a user personalized portrait based on an unsupervised learning model, and the unsupervised learning model identifies the user's health status, product service preferences and needs based on the obtained user medical health data;

[0008] Construct a product service data warehouse using a database management platform. The database management platform uses an association analysis algorithm to mine potential connections between different product services, and classifies product services into health status, product service preferences and needs based on the association analysis results;

[0009] A multi-dimensional AI recommendation model is constructed using a full-process automatic learning framework. The full-process automatic learning framework extracts feature information from user personalized portraits and product service data warehouses, and uses the extracted feature information for model training;

[0010] Based on the constructed multi-dimensional AI recommendation model, the matching degree between users and different types of product services is evaluated from the dimensions of users' health status, product service preferences and needs, and a personalized product service recommendation result list is generated based on the matching degree evaluation results.

[0011] Furthermore, user medical and health data and smart medical product service data are obtained through application programming interfaces (APIs) and microservice architectures, and are transmitted to a cloud server for processing and storage. The application programming interface (API) receives and updates users' personal information, medical consultation behavior data, health monitoring data and medical history records in real time by connecting with intelligent medical devices, electronic medical record systems and telemedicine platforms. The microservice architecture splits the smart medical product service data into multiple microservices according to functional modules. The microservices are deployed on the cloud server through virtualized containers, and data access and update are carried out through a service registration and discovery mechanism.

[0012] Furthermore, after obtaining user medical and health data, cleaning and preprocessing operations are performed on the medical and health data, and an autoencoder is used to extract the features of users' personal information, medical history, physical examination data, living habits and purchase history. The unsupervised learning model identifies users' health status, product service preferences and needs through self-organizing mapping, and vectorizes users according to the dimensions of health status, product service preferences and needs through a clustering algorithm to form a user personalized portrait. The user personalized portrait is updated in real time using an incremental learning method.

[0013] Furthermore, the preprocessed and cleaned smart medical product service data is obtained from the database management platform, and the data is converted into a binary matrix form. The association analysis algorithm uses the frequent itemset mining method to obtain frequent itemsets that meet the minimum support threshold from the data, and generates association rules that meet the minimum confidence threshold based on the frequent itemsets. The minimum support and minimum confidence thresholds are used to screen frequent itemsets and association rules.

[0014] Furthermore, the full-process automated machine learning framework includes a feature extraction module, a model selection module, a model training module and a validation and tuning module.

[0015] Further, the feature extraction module uses a natural language processing library to screen out features that match the health status, product service preferences, and needs from the user's medical and health data and the intelligent medical product service data, and performs feature scaling, feature combination, and feature transformation on the screened features. The model selection module uses a recurrent neural network and a cross-validation method to select and evaluate the recommendation algorithm. The model training module divides the screened features into a training set, a validation set, and a test set, and uses the training set features to train the selected algorithm to obtain an initial multi-dimensional AI recommendation model. The validation and tuning module tunes the hyperparameters of the initial multi-dimensional AI recommendation model through an automated search and evaluation mechanism, and validates the tuned multi-dimensional AI recommendation model using the validation set features.

[0016] Further, in the multi-dimensional AI recommendation model, the matching degree between the user and different types of product services is evaluated from the dimensions of health status, product service preferences, and needs. The formula for the comprehensive matching degree output function between the user and different types of product services is:

[0017]

[0018] In the formula, represents the comprehensive matching degree between the user and different types of product services, represents the weight of the health status dimension, represents the weight of the product service preference dimension, represents the weight of the product service demand dimension, represents the matching degree between the user's medical and health data in the health status dimension and the product services of the health status type, represents the matching degree between the user's medical and health data in the product service preference dimension and the product services of the product service preference type, represents the matching degree between the user's medical and health data in the product service demand dimension and the product services of the product service demand type. is used to eliminate the dimensional differences between dimensions, is used to adjust the sensitivity of the comprehensive matching, is used to normalize the comprehensive matching result, represents a regulation factor used to balance the importance of each dimension.

[0019] Further, in the process of generating the personalized product service recommendation result list, a retrieval planning strategy is used to perform different types of sorting and comprehensive sorting on the personalized product service recommendation results. The retrieval planning strategy automatically expands and updates different types of product services according to the user's historical health status, product service preferences, and needs, and performs different types of sorting and comprehensive sorting on the recommended product services according to the matching degree technical results.

[0020] Furthermore, the generated list of personalized product service recommendation results is presented to the user through a user interface, and feedback from the user on the list of personalized product service recommendation results is collected. The user interface includes the name, function, price, and user reviews of the recommended products, and the multi-dimensional AI recommendation model is verified and optimized based on the user reviews.

[0021] Furthermore, an AI-driven product service recommendation system includes: a data acquisition layer, a data processing layer, a database management platform, a model training and recommendation layer, and a result sorting and display layer. The data acquisition layer obtains raw data, transmits it to the database management platform for storage, and transmits it to the data processing layer for processing. The data processing layer constructs a user personalized portrait and transmits it to the model training and recommendation layer for model training and recommendation, and then transmits it to the database management platform. The data processing layer and the database management platform are connected bidirectionally. The database management platform stores and manages data and transmits information to the model training and recommendation layer. The model training and recommendation layer generates recommendation results and transmits them to the result sorting and display layer. The result sorting and display layer displays the recommendation results and collects user feedback.

[0022] The beneficial effects of the present invention are as follows:

[0023] 1. The present invention uses the application programming interface (API) and microservices architecture to efficiently obtain user medical and health data and intelligent medical product service data from multiple data sources. This data acquisition method ensures the real-time and accuracy of the data, adapts to the increasing data volume and user needs, provides a solid foundation for subsequent analysis and recommendation, and transmits the obtained data to the cloud server for centralized processing and storage, which not only improves the data processing speed but also facilitates the unified management and maintenance of the data, reducing the risk of data loss and leakage.

[0024] 2. The present invention constructs a user personalized portrait based on an unsupervised learning model, which can deeply explore the user's health status, product service preferences, and needs. This method enables the recommendation system to more accurately understand the actual needs of users, thereby providing product service recommendations that better meet user expectations. By constructing a product service data warehouse through the database management platform and using an association analysis algorithm to deeply explore the potential connections between different product services, these connections not only help discover new product service combinations but also provide users with more diversified choices. Classifying product services based on the association analysis results enables the recommendation system to more accurately match user needs with the characteristics of product services, improving the accuracy and satisfaction of recommendations.

[0025] 3. The present invention constructs a multi-dimensional AI recommendation model using a full-process automatic learning framework, making the model training, selection, and validation tuning processes more automated and intelligent, reducing the cost of manual intervention, and improving the efficiency and accuracy of recommendations. Based on the constructed multi-dimensional AI recommendation model, the matching degree between users and different types of product services is evaluated from the dimensions of users' health status, product service preferences, and demands. This matching degree evaluation method ensures the accuracy and pertinence of recommendations.

[0026] 4. When generating a personalized product service recommendation result list, the present invention uses a retrieval planning strategy to perform different types of sorting and comprehensive sorting on the recommendation results, enabling users to find the product services that best meet their needs more quickly, improving the user experience and satisfaction. The generated personalized product service recommendation result list is displayed to users through the user interaction interface, and user feedback on the recommendations is collected. This user feedback mechanism helps to promptly discover the problems and deficiencies in the recommendation system and provides strong support for the continuous optimization and improvement of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic diagram of the overall method flow of the present invention;

[0028] Figure 2 is a schematic diagram of the architecture of the full-process automatic learning framework in the present invention;

[0029] Figure 3 is a schematic diagram of the overall system architecture of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings from Figure 1 to Figure 3 . Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] An embodiment of the present invention discloses an AI-driven product service recommendation method

[0032] An AI-driven product service recommendation method, as shown in Figure 1 , includes the following steps:

[0033] Step 1. Data acquisition and transmission

[0034] Obtain users' medical and health data using the Application Programming Interface (API). Connect to intelligent medical devices (such as intelligent bracelets, intelligent sphygmomanometers, etc.) through the API to receive and update users' health monitoring data in real time, such as physiological indicators like heart rate, blood pressure, and blood sugar. Connect to the hospital's electronic medical record system through the API to obtain detailed medical data of users, including medical history records, diagnosis information, medication conditions, etc. Collaborate with the telemedicine platform to receive and update users' personal information, medical consultation behavior data, etc. in real time through the API, including users' age, gender, region, disease consultation history, etc.

[0035] Utilize the microservices architecture to obtain smart healthcare product service data. The smart healthcare product service data includes data on healthcare, medications, medical devices, and medical examinations. Split the smart healthcare product service data into multiple microservices according to functional modules, such as drug recommendation services, health management services, disease prevention services, etc. Each microservice focuses on a specific business area, providing independent functions and data. Deploy each microservice in a virtualized container (such as Docker) to achieve lightweight and portable services. This helps improve the scalability and flexibility of the services.

[0036] Deploy service registration and discovery components (such as Eureka, Nacos, etc.) on the cloud server to manage the registration, discovery, and data access of microservices. When a microservice starts, it automatically registers its service address and port with the registration center. When other microservices need to access, they obtain the service address through the registration center and conduct data access. This mechanism ensures the real-time and reliability of data access.

[0037] Use a secure network protocol (such as HTTPS) to transmit the obtained users' medical and health data and smart healthcare product service data to the cloud server. Establish a data processing system on the cloud server to perform preprocessing operations on the users' medical and health data, such as cleaning, deduplication, and format conversion, to ensure the quality and consistency of the data. At the same time, classify, summarize, and organize the smart healthcare product service data for subsequent analysis and mining. Store the processed data in a database or data warehouse on the cloud server for subsequent use in analysis and recommendation algorithms.

[0038] Through the above steps, users' medical and health data and smart healthcare product service data can be obtained efficiently and transmitted securely, providing a solid data foundation for subsequent product service recommendations.

[0039] Step 2: Build a user personalized profile

[0040] Clean and preprocess the user's medical and health data; check and delete duplicate records in the dataset. For missing data, handle it by filling (such as mean filling, median filling, mode filling) or deleting according to the specific situation. Use statistical methods or machine learning algorithms to detect outliers and correct or delete them according to business logic. Unify the formats of data from different sources to ensure data consistency and comparability. Normalize or standardize numerical data to eliminate dimensional differences, and encode categorical variables, such as using one-hot encoding or label encoding.

[0041] After data cleaning and preprocessing, select key features related to the user's health status, product service preferences, and needs from the user's personal information, medical history, physical examination data, lifestyle, and purchase history. Design a neural network structure with an input layer, a hidden layer, and an output layer as an autoencoder model. The input layer receives the original feature vector, the hidden layer is used for feature extraction, and the output layer reconstructs the input data. Train the autoencoder so that it can learn the low-dimensional representation (i.e., the feature vector) of the data. After training, use the output of the hidden layer as the user's feature vector.

[0042] According to the dimension and quantity of the feature vector, design the topological structure and parameters of the self-organizing mapping SOM network. Input the feature vector into the SOM network, and through iterative adjustment of the network weights, make each neuron represent a specific user group. According to the output of the SOM network, map the users to different neurons, thereby identifying the user's health status, product service preferences, and needs.

[0043] According to business requirements and data characteristics, select appropriate clustering algorithms, such as K-means, DBSCAN, etc. Input the feature vector into the clustering algorithm to obtain the clustering result. According to the clustering result, vectorize the users according to the dimensions of health status, product service preferences, and needs to form a user personalized portrait.

[0044] According to the arrival frequency and quantity of new data, design appropriate incremental learning strategies, such as online learning, batch update, etc. When receiving new medical and health data, update the autoencoder, SOM network, and clustering model according to the incremental learning strategy. Store the updated user personalized portrait in the database for subsequent use by product service recommendation algorithms, and provide an efficient portrait retrieval mechanism to quickly obtain the user's personalized portrait during product service recommendation.

[0045] Through the implementation of the above steps, an accurate and real-time user personalized portrait can be constructed based on the unsupervised learning model, providing strong support for subsequent product service recommendations. At the same time, by introducing the incremental learning method, the timeliness and accuracy of the user personalized portrait can be ensured.

[0046] Step 3: Construct a product service data warehouse

[0047] Use a database management platform to construct a product service data warehouse, and use association analysis algorithms to mine potential connections between different product services.

[0048] First, obtain the intelligent medical product service data from the database management platform, and preprocess and clean the obtained data to ensure the accuracy, integrity, and consistency of the data. This includes steps such as deduplication, filling missing values, and handling outliers, and converting the data into a binary matrix form. In this matrix, each row represents a transaction (for example, a patient's purchase record or usage record), each column represents a product service (for example, a certain medical device or medical service), and the value in the matrix indicates whether the product service appears in the transaction (1 means it appears, 0 means it does not appear).

[0049] Use frequent itemset mining algorithms, such as the Apriori algorithm or the FP-growth algorithm, to find frequent itemsets that meet the minimum support threshold from the binary matrix. By continuously iterating, candidate itemsets are gradually generated, and their support is calculated, so as to screen out frequent itemsets. This algorithm needs to scan the database multiple times, so it may be less efficient when dealing with large-scale data. By constructing an FP tree, frequent itemsets can be efficiently discovered. This algorithm does not need to generate candidate itemsets, so it is more efficient when dealing with large-scale data. According to the actual needs and the size of the dataset, set a suitable minimum support threshold. Only itemsets with a support greater than or equal to this threshold will be regarded as frequent itemsets. Run the selected algorithm to mine frequent itemsets that meet the minimum support threshold from the data. In an intelligent medical product recommendation system, the minimum support can be set to 0.1, and the minimum confidence can be set to 0.7 to ensure that the mined frequent itemsets and association rules have high significance and value. Through this setting, a group of relatively accurate and meaningful recommendation results can be obtained, improving user satisfaction.

[0050] Use support and confidence as metrics for association rules. Support measures the frequency of rule occurrence, while confidence measures the credibility of the rule. According to the actual needs, set a suitable minimum confidence threshold. Only rules with a confidence greater than or equal to this threshold will be regarded as valid association rules. Based on the frequent itemsets, generate association rules that meet the minimum confidence threshold. These rules describe the potential connections between different product services.

[0051] According to the association rules, product services are classified into different health states. The demand of patients for future product services can be predicted.

[0052] Through the above steps, the intelligent medical product service data in the database management platform can be fully utilized, the potential connections between different product services can be mined, and the product services can be classified and optimized based on the association analysis results.

[0053] Step 4: Build a multi-dimensional AI recommendation model

[0054] The full-process automatic learning framework aims to build a multi-dimensional AI recommendation system through model training with the feature information extracted from the user's personalized profile and the product service data warehouse. The full-process automatic learning framework includes four key modules: a feature extraction module, a model selection module, a model training module, and a validation and tuning module.

[0055] The feature extraction module is used to extract features related to health status, product service preferences, demands, etc. from the user's medical and health data and the intelligent medical product service data. First, collect the user's health data (such as disease history, physical signs information, health indicators, etc.) and the relevant data of intelligent medical product services (such as usage behavior, product preferences, etc.), and through natural language processing (NLP) technology, screen out the feature information that matches the health status, product service preferences, etc. NLP technology can help extract key information (such as diagnostic reports, user comments, etc.) from structured or unstructured text data, standardize or normalize the feature data to ensure the dimensional consistency between different features, and through feature engineering technology, combine the original features to generate new and more representative features (such as user health risk assessment, usage frequency, etc.), and apply feature transformation (such as PCA principal component analysis) to reduce dimensions and improve the model efficiency.

[0056] The model selection module is used to select an appropriate recommendation algorithm and evaluate its effect. According to the data characteristics (such as sequence data, time-series behavior, etc.), select a suitable recommendation algorithm. Here, the recurrent neural network (RNN) is considered a model suitable for processing time-series data and recommendation systems, especially suitable for the changing trends of user health data and product services. Through cross-validation methods (such as k-fold cross-validation), evaluate the performance of different algorithms and select the optimal model. Cross-validation helps to avoid overfitting and ensure the generalization ability of the model on different datasets.

[0057] The model training module is used to train a model using a training set to obtain an initial multi-dimensional AI recommendation model. First, the extracted feature data is divided into three subsets: a training set, a validation set, and a test set. Generally, 70% of the data is used as the training set, 15% as the validation set, and 15% as the test set. The selected algorithm (such as RNN) is trained using the training set features to obtain an initial recommendation model. The training process includes preliminary adjustment of the model parameters to minimize the prediction error, and the model is optimized through loss functions (such as mean squared error, cross-entropy, etc.) to ensure that the model can accurately predict user needs and recommend relevant products and services.

[0058] The validation and tuning module is used to validate and tune the initial model to improve the model's performance and recommendation accuracy. The hyperparameters of the model (such as learning rate, number of hidden layer units, etc.) are adjusted through automated search (such as grid search or random search) to optimize the model performance. The tuned model is validated using the validation set to evaluate its performance on unseen data, ensuring that the model does not overfit. The performance of the tuned multi-dimensional AI recommendation model is evaluated using common evaluation metrics (such as accuracy, recall, F1-score, etc.), and the hyperparameters are further adjusted to improve the prediction effect. According to the model validation results, it may be necessary to return to the feature extraction or model training stage to further optimize the feature selection and algorithm parameters to improve the recommendation accuracy and stability of the model.

[0059] In the feature extraction module, data cleaning is a crucial step, which directly affects the effects of subsequent feature scaling, feature combination, and feature transformation. For different types of missing values and outliers, we select different filling strategies and detection methods, as follows:

[0060] Missing value filling strategies: 1. Mean filling: When the missing values are few and evenly distributed, mean filling can be used because it does not significantly change the overall distribution of the data. This method is applicable to numerical features, especially those without obvious trends or periodicity. 2. Median filling: For data with outliers or skewed distributions, median filling is more robust than mean filling because it is not affected by extreme values. 3. Mode filling: For categorical features, if the missing values are not many and the frequency of a certain category is significantly higher than other categories, mode filling can be used. 4. Interpolation: For time series data, if the missing values appear continuously and the quantity is not large, interpolation methods (such as linear interpolation, polynomial interpolation) can be used to estimate the missing values. 5. Regression prediction: For complex missing data patterns, machine learning models (such as linear regression, decision trees, etc.) can be used to predict the missing values.

[0061] Outlier Detection Methods: 1. Detection based on statistical distribution: Such as the 3σ principle or the Z-score method, which is applicable to data with normal distribution or approximately normal distribution. By calculating the multiple of the standard deviation between each data point and the mean, it is determined whether it belongs to an outlier. 2. Detection based on box plot: Applicable to data with any distribution. By drawing a box plot, the quartiles and outlier range of the data can be visually identified. 3. Detection methods based on machine learning: Such as IsolationForest or Local Outlier Factor (LOF). These methods do not rely on the distribution assumption of the data and can more flexibly detect outliers in complex datasets.

[0062] In the model selection module, we choose the Recurrent Neural Network (RNN) as one of the basic models of the recommendation algorithm, mainly based on its ability to process sequential data. However, RNN also has some limitations, such as the long-term dependence problem (i.e., it is difficult to capture long-distance temporal dependence relationships) and the gradient vanishing / exploding problem during training. Therefore, when choosing RNN, we need to weigh its advantages and disadvantages. RNN can naturally process sequential data, such as time series analysis, natural language processing and other tasks. It uses recurrent connections to take the output of the previous moment as the input of the next moment, so as to capture the temporal dependence in the data. RNN is prone to the long-term dependence problem when dealing with long sequences, resulting in the model being difficult to learn long-distance dependence relationships. In addition, the training process of RNN is easily affected by gradient vanishing or gradient explosion, resulting in the model being difficult to converge or the training being unstable.

[0063] Compared with RNN, the Transformer model has stronger capabilities and higher efficiency in processing sequential data. Transformer captures the dependence relationships in the data through the Self-Attention mechanism and is not limited by the sequence length. In addition, the Transformer model can perform parallel computing, greatly improving the training speed. However, the Transformer model also has some challenges, such as higher model complexity, requiring more data and computing resources, etc.

[0064] As shown in the Figure 2 attachment, the output end of the feature extraction module is connected to the input end of the model selection module, the output end of the model selection module is connected to the input end of the model training module, the output end of the model selection module is connected to the input end of the validation and tuning module, and the output end of the validation and tuning module is connected to the input end of the model training module;

[0065] Through a full-process automatic learning framework, combined with key steps such as feature extraction, model selection, training, and optimization, it is possible to automatically build a multi-dimensional AI recommendation model to accurately provide users with personalized healthcare recommendation services. Through continuous verification and optimization, the accuracy and adaptability of the recommendation system are continuously improved, and finally, personalized product and service matching is achieved to meet the health needs of users.

[0066] Step 5: Generate a list of personalized product and service recommendation results

[0067] Collect health status data, product and service preference data, and demand data from the user's personalized profile, and organize a product and service dataset containing health status types, product and service preference types, and product and service demand types. Use machine learning or deep learning techniques, combined with user data and product and service data, to train a multi-dimensional AI recommendation model.

[0068] Use the trained multi-dimensional AI recommendation model to calculate the matching degree between the user and different types of products and services from three dimensions: health status, product and service preferences, and demand. By comparing the user's historical behavior and product and service evaluations, verify the accuracy and reliability of the matching degree calculation. The retrieval planning strategy automatically expands and updates health-related products and services, such as nutritional products and sports equipment, based on the user's historical health status data, automatically expands and updates the user-preferred products and services, such as brands and types, based on the user's historical purchase behavior and evaluations, automatically expands and updates the products and services that meet the needs, such as seasonal products and regular maintenance services, based on the user's current and future needs, and generates a recommendation candidate set with a higher matching degree with the user according to the automatically expanded and updated products and services.

[0069] According to the priorities of the health status, product and service preference, and demand dimensions, sort the products and services in the recommendation candidate set by type. For example, for users with poor health status, give priority to recommending health-related products and services.

[0070] According to the matching degree calculation results and weight allocation, calculate the comprehensive matching degree of each recommendation candidate. Use a suitable sorting algorithm (such as quicksort, mergesort, etc.) to sort the recommendation candidates according to the comprehensive matching degree, and generate a list of personalized product and service recommendation results based on the sorted recommendation candidates.

[0071] Use evaluation metrics such as accuracy, recall rate, and F1 score to evaluate the recommendation model. According to the evaluation results and user feedback, regularly update the multi-dimensional AI recommendation model and the retrieval planning strategy to improve the accuracy of the recommendation system and user satisfaction.

[0072] The multi-dimensional AI recommendation model evaluates the matching degree between users and different types of product services from the dimensions of health status, product service preferences, and needs. The datasets of different types of product services are , representing the product services of the health status type, representing the product services of the product service preference type, representing the product services of the product service need type. The datasets of user personalized portraits are , representing the user's medical and health data in the health status dimension, representing the user's medical and health data in the product service preference dimension, representing the user's medical and health data in the product service need dimension. The calculation formula for the matching degree between users and different types of product services is:

[0073] (1)

[0074] In formula (1), represents the matching degree between the user's medical and health data in the health status dimension and the product services of the health status type, represents the matching degree between the user's medical and health data in the product service preference dimension and the product services of the product service preference type, represents the matching degree between the user's medical and health data in the product service need dimension and the product services of the product service need type, represents the relationship factor between the user's medical and health data in the health status dimension and the product services of the health status type, used to represent the correlation between the user's medical and health data in the product service preference dimension and the product services of the product service preference type. The formula for the comprehensive matching degree output function between users and different types of product services is:

[0075] (2)

[0076] In formula (2), represents the comprehensive matching degree between users and different types of product services, represents the weight of the health status dimension, represents the weight of the product service preference dimension, represents the weight of the product service need dimension, is used to eliminate the dimensional differences between dimensions, is used to adjust the sensitivity of the comprehensive matching, is used to normalize the comprehensive matching results, represents the adjustment factor, used to balance the importance of each dimension.

[0077] In a multi-dimensional AI recommendation model, the setting of weights is crucial for the calculation of the comprehensive matching degree. The weight optimization algorithm or strategy can be obtained through the following methods: 1. Weight allocation based on historical data. In the initial stage, the weights of each dimension can be initially set according to the influence degree of each dimension in historical data on users' purchase or use of product services. For example, if the data in the health status dimension shows the greatest influence on users' selection of product services in historical data, a higher weight can be given to this dimension. 2. Dynamic adjustment of weights based on A / B testing. In practical applications, the effects of different weight allocation schemes can be verified through A / B testing. Randomly divide the user group into two groups, apply different weight allocation schemes respectively, and then compare indicators such as the conversion rate and satisfaction of users' purchase or use of product services in the two groups. According to the results of the A / B testing, dynamically adjust the weights of each dimension to achieve the optimal comprehensive matching degree. 3. Optimize weights using machine learning algorithms. Machine learning algorithms (such as gradient descent, stochastic gradient descent, Adam, etc.) can be used to optimize weights. Take the comprehensive matching degree as the objective function, and continuously adjust the weights through iterative optimization algorithms to make the objective function reach the optimal value.

[0078] The cold start problem is a common problem in recommendation systems, especially when new users or new product services are newly added to the system. Solutions to the cold start problem include the following: 1. A hybrid model combining collaborative filtering and content recommendation. When a new user joins, the content-based recommendation method can be used to make preliminary personalized recommendations based on the user's basic information and preferences provided during registration. At the same time, as the accumulation of user behavior, the collaborative filtering-based recommendation method is gradually introduced, and similar content liked by similar users is recommended by analyzing the similarity between users. Integrate the methods of collaborative filtering and content recommendation to form a hybrid recommendation system to provide more accurate and diverse recommendation services. 2. Utilize user basic information and registration preferences. When a new user registers, collect the user's basic information and preferences, such as age, gender, occupation, hobbies, etc. According to this information, use rule-based recommendation methods or content-based recommendation methods to recommend some product services that the user may be interested in. 3. Guide new users to perform initial behaviors. By designing some guiding tasks or questionnaires, guide new users to perform initial behaviors, such as filling out interest questionnaires, watching guides, etc. These behaviors can help the system understand the users' interests and needs faster, so as to provide more accurate recommendation services for users. 4. Use popular product services for recommendation. When a new user joins, some popular product services or popular content can be recommended to the user. These popular product services usually have high user satisfaction and conversion rates, and can help new users find interesting content faster.

[0079] By optimizing strategies such as weight setting and introducing a hybrid recommendation model, the accuracy and reliability of the multi-dimensional AI recommendation model can be improved. At the same time, by leveraging users' basic information and registration preferences, guiding new users to perform initial behaviors, and using popular product services for recommendations, the cold start problem can be effectively solved.

[0080] Step Six: Display the recommendation results through the user interface and collect feedback

[0081] Design an intuitive and user-friendly user interface, such as a health management APP, web page, etc., to display the list of personalized product service recommendation results. The interface design can refer to modern UI / UX design principles to ensure a smooth user experience. Intuitively display the generation process and basis of the recommendation results through forms such as animations and charts. Provide personalized notification functions such as regular health advice and preferential service recommendations, and reach users through push notifications, emails, etc.

[0082] Adopt a simple and clear layout to ensure that users can quickly find the information they need. A navigation bar can be set at the top of the home page, including module links such as "Recommended Products", "Health News", "Personal Center", etc. The central area of the home page displays the list of personalized product service recommendation results. Each recommendation item should include the product / service name, picture, short description, price / discount information, and a "View Details" button. The recommendation items are sorted according to the matching degree calculated by the algorithm, and the item that best meets the user's needs is placed at the top. Users can browse more recommendation items by swiping the screen, and click the "View Details" button to enter the product / service details page. The top of the details page displays the high-definition picture and name of the product / service, and below are the detailed description, price / discount information, user reviews, purchase / favorite buttons, etc. Users can view more product information on the details page, such as usage instructions, ingredient list (for health products), user reviews, etc. Clicking the "Buy" button can directly jump to the purchase page, and clicking the "Favorite" button can add the product / service to the personal favorites folder. The personal center page includes modules such as user basic information, favorites, purchase records, health data, settings, etc. On the personal center page, users can set preferences for regular health advice or preferential service push. The system regularly pushes personalized health advice and preferential service information to users according to their health status, product service preferences, and needs. The push methods can include in-APP notifications, text messages, emails, etc.

[0083] After the user logs in, the system automatically generates a list of personalized product service recommendation results based on the user's health status, product service preferences, and needs. The user browses the recommended items by swiping the screen and clicks "View Details" to enter the details page to learn about the detailed information of the product / service and decide whether to purchase or collect it. After the user browses the recommendation results or uses the product / service, they can enter the feedback page through the personal center or other entrances. The system provides options for satisfaction scoring, comment content input area, and operation behavior statistics (such as clicks, purchases, etc.), and the user can fill in the feedback content according to the actual situation.

[0084] Design a feedback classification system to classify user feedback into three categories: satisfaction scoring, comment content analysis, and operation behavior statistics. Satisfaction scoring is used to measure the user's satisfaction with the recommendation results; comment content analysis is used to mine the opinions and suggestions of the user on the recommendation results; operation behavior statistics is used to record the click, purchase, and other behavior data of the user on the recommendation results. Collect user feedback data in real time and conduct in-depth analysis. According to the analysis results, continuously optimize the recommendation algorithm and improve the recommendation effect.

[0085] The user can give a satisfaction score to each recommendation result, such as a 5-star system, a 10-point system, etc. The scoring results are used to measure the accuracy of the recommendation algorithm and the user's satisfaction. The user can enter their opinions and suggestions on the recommendation results in the comment area. The system performs text analysis on the comment content, extracts keywords and sentiment tendencies, and uses them to optimize the recommendation algorithm and improve the user experience. The system records the click, browse, purchase, and other behavior data of the user on the recommendation results. These data are used to analyze the user's behavior habits and preferences and provide data support for the subsequent recommendation algorithm.

[0086] The system collects the user's satisfaction score, comment content, and operation behavior data in real time through the user feedback module. Use data analysis tools (such as the Pandas, NumPy, etc. libraries in Python) to conduct in-depth analysis on the collected feedback data. Through statistical analysis, text mining, and other methods, mine the user's real needs and opinions, and provide strong support for optimizing the recommendation algorithm and improving the user experience. According to the feedback analysis results, continuously optimize the recommendation algorithm and improve the recommendation effect. At the same time, according to the user's feedback opinions and suggestions, continuously improve the user interface and operation process to improve the user experience.

[0087] Through the above design, the user interface will be more friendly and intuitive, and the personalized notification function will improve the user's participation and satisfaction. At the same time, the user feedback module will help the system continuously optimize the recommendation algorithm and improve the user experience, realizing more accurate and personalized product service recommendations.

[0088] An AI-driven product service recommendation system, as shown in the appendix Figure 3 shown, includes;

[0089] The data collection layer is used to collect users' medical and health data and intelligent medical product service data from multiple data sources and transmit them to the cloud server. It obtains users' medical and health data, including medical records, diagnosis and treatment records, physical examination reports, lifestyle habits, etc., from different data sources (such as hospital databases, medical applications, intelligent devices, etc.) using APIs. By collaborating with medical devices, applications, online platforms, etc., it collects product service data, such as medical devices, health management services, drug recommendations, etc. The collected data is transmitted to the cloud server using encrypted channels (such as HTTPS, SSL / TLS) and stored in the database to ensure data security and privacy protection.

[0090] The data processing layer is used to process the collected user data and build a personalized profile. It identifies users' health status, preferences, and needs, performs unsupervised learning on users' health data based on clustering analysis (such as K-means, DBSCAN, etc.), extracts health status and potential disease risks, and combines users' health history, lifestyle, disease prevention needs, etc. Using machine learning algorithms (such as decision trees, random forests, etc.), it builds a personalized user profile, including information such as health needs, service preferences, and disease risks.

[0091] The database management platform is used to manage and optimize product service data and perform association analysis. It builds a data warehouse for intelligent medical products and services, integrates various product service information, such as product type, function, price, indications, etc., and uses association rule learning (such as Apriori, FP-growth, etc. algorithms) to mine potential connections between different product services, identify which products / services are often used or purchased together, thereby optimizing the recommendation logic. It classifies medical product services using clustering and classification algorithms (such as K-means, support vector machines, etc.) and assigns tags to each product.

[0092] The model training and recommendation layer is used to build a multi-dimensional AI recommendation model for personalized recommendations.

[0093] The full-process automatic learning framework: includes the following modules: The feature extraction module extracts key information, such as age, gender, health status, disease history, etc., from users' health data and product service data. The model selection module selects appropriate machine learning models (such as deep learning, collaborative filtering, matrix factorization, etc.) and evaluates the effects of different models based on the user profile and the product data warehouse. The model training module uses historical data to train the model and adopts methods such as cross-validation to optimize the model's parameters. The validation and tuning module tunes according to the model's performance metrics (such as accuracy, recall rate, F1-score, etc.) to avoid overfitting or underfitting.

[0094] The model evaluates the matching degree of different products and services based on the user's health status, historical data, and information in the product service library, and generates personalized recommendations for the user.

[0095] The result sorting and display layer is used to sort the recommendation results and display the recommendation results through the user interface to collect feedback. Different types of sorting are performed on the recommendation results, including: sorting based on matching degree, sorting according to the matching degree between the user and the product service, and giving priority to recommending products that best meet the user's needs. Sorting based on user feedback, adjusting the sorting strategy through the user's historical behavior, purchase records, evaluation feedback, etc.

[0096] Develop a user-friendly interface (such as a web application, mobile APP, etc.) to clearly display the recommended medical and health products and services, and allow users to perform operations (such as viewing details, purchasing, consulting, etc.). Collect user feedback on the recommendation results (such as ratings, click behaviors, purchase conversions, etc.) through the interface design, and use this feedback information to further optimize the recommendation model.

[0097] Continuously optimize the recommendation accuracy according to user feedback and system performance. Compare the effects of different recommendation strategies and interface designs through A / B testing to optimize the user experience and recommendation effects. Perform incremental learning of the model based on real-time data, and continuously update the user portrait and product service data. Combine new user health data and product updates to dynamically adjust the recommendation strategy to ensure that the recommendation results always match the user's needs.

[0098] Through the above content, a complete AI-driven medical and health product service recommendation system can be constructed. This system can not only provide personalized product and service recommendations based on the user's health data and preferences, but also improve the recommendation quality and user experience through continuous optimization and user feedback.

[0099] So far, the technical solution of the present invention has been described in combination with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An AI-driven product service recommendation method, characterized in that: The following steps are involved: Step 1: Use the application programming interface (API) and microservice architecture to obtain user medical health data and smart medical product service data, and transmit the obtained data to the cloud server for processing and storage; Step 2: Build a personalized user portrait based on an unsupervised learning model, where the unsupervised learning model identifies the user's health status, product and service preferences, and needs based on the acquired user medical and health data; Step 3: Use a database management platform to build a product service data warehouse, the database management platform uses an association analysis algorithm to mine the potential connections between different product services, and classifies the product services into health status, product service preferences and needs based on the association analysis results; Step 4: Use a full-process automatic learning framework to build a multi-dimensional AI recommendation model. The full-process automatic learning framework extracts feature information from user personalized portraits and product service data warehouses, and uses the extracted feature information for model training. The full-process automated machine learning framework includes a feature extraction module, a model selection module, a model training module, and a verification and tuning module. The feature extraction module uses a natural language processing library to filter out features that match health status, product service preferences and needs from user medical health data and smart medical product service data, and performs feature scaling, feature combination and feature conversion on the filtered features. The model selection module uses a recurrent neural network and cross-validation method to select and evaluate the recommendation algorithm; the model training module divides the filtered features into a training set, a validation set and a test set, and uses the training set features to train the selected algorithm to obtain an initial multi-dimensional AI recommendation model. The verification and tuning module tunes the hyperparameters of the initial multi-dimensional AI recommendation model through an automated search and evaluation mechanism, and uses the verification set features to verify the tuned multi-dimensional AI recommendation model; Step 5: Based on the constructed multi-dimensional AI recommendation model, the matching degree between the user and different types of products and services is evaluated from the user's health status, product and service preferences, and demand dimensions, and a personalized product and service recommendation result list is generated based on the matching degree evaluation results. In the process of generating the personalized product and service recommendation result list, a retrieval planning strategy is used to sort the personalized product and service recommendation results in different types and comprehensive sorting; In step 5, the multi-dimensional AI recommendation model evaluates the matching degree between users and different types of products and services from the dimensions of health status, product and service preferences, and demand. The data sets of different types of products and services are , Represents the product service of health status type, The product service indicating the product service preference type, The data set of product services representing product service demand types and user personalized portraits is , Represents the user's medical health data in the health status dimension, User medical and health data representing product and service preference dimensions, The user's medical health data represents the product service demand dimension. The matching degree between the user and different types of product services is calculated as follows: In formula (1), Indicates the matching degree between the user's medical health data in the health status dimension and the product services of the health status type. Indicates the matching degree between the user's medical and health data in the product service preference dimension and the product service of the product service preference type, Indicates the matching degree between the user medical health data in the product service demand dimension and the product service of the product service demand type, The relationship factor between the user's medical health data in the health status dimension and the product services of the health status type is used to represent the correlation between the user's medical health data in the product service preference dimension and the product services of the product service preference type. The comprehensive matching degree output function formula between the user and different types of product services is: In formula (2), P represents the comprehensive matching degree between users and different types of product services. represents the weight of the health status dimension, represents the weight of the product service preference dimension, Represents the weight of the product service demand dimension, Used to eliminate dimensional differences between dimensions. Used to adjust the sensitivity of comprehensive matching. Used to normalize the comprehensive matching results. represents the adjustment factor, which is used to balance the importance of each dimension; Step 6: Use a user interaction interface to display the generated personalized product and service recommendation result list to the user, and collect user feedback on the personalized product and service recommendation result list.

2. The AI-driven product service recommendation method according to claim 1, characterized in that: In the step 1, the application programming interface By connecting with smart medical devices, electronic medical record systems and telemedicine platforms, the user's personal information, medical consultation behavior data, health monitoring data and medical history records are received and updated in real time. The microservice architecture splits the smart medical product service data into multiple microservices according to functional modules. The microservices are deployed on the cloud server through virtualized containers, and data is accessed and updated through service registration and discovery mechanisms.

3. The AI-driven product service recommendation method according to claim 1, characterized in that: In the step 2, after receiving the user's medical health data, the medical health data is cleaned and preprocessed, and an autoencoder is used to extract the features of the user's personal information, medical history, physical examination data, living habits and purchase history. The unsupervised learning model identifies the user's health status, product service preferences and needs through self-organizing mapping, and vectorizes the user according to the health status, product service preferences and needs dimensions through a clustering algorithm to form a personalized user portrait, which is updated in real time using incremental learning.

4. The AI-driven product service recommendation method according to claim 1, characterized in that: In the step three, the preprocessed and cleaned smart medical product service data is obtained from the database management platform, and the data is converted into a binary matrix form. The association analysis algorithm uses a frequent item set mining method to obtain frequent item sets that meet the minimum support threshold from the data, and generates association rules that meet the minimum confidence threshold based on the frequent item sets. The minimum support and minimum confidence thresholds are used to screen frequent item sets and association rules.

5. The AI-driven product service recommendation method according to claim 1, characterized in that: The retrieval planning strategy automatically expands and updates different types of product services according to the user's historical health status, product service preferences and needs, and sorts the recommended product services by different types and comprehensive sorting according to the matching technical results.

6. The AI-driven product service recommendation method according to claim 1, characterized in that: The user interaction interface includes the name, function, price and user evaluation of the recommended product, and the multi-dimensional AI recommendation model is verified and optimized based on the user evaluation.

7. An AI-driven product service recommendation system, applied to an AI-driven product service recommendation method according to any one of claims 1 to 6, characterized in that: include; Data collection layer, using application programming interface The microservice architecture obtains user medical health data and smart medical product service data from various data sources and transmits them to the cloud server for processing and storage; The data processing layer uses unsupervised learning models to build personalized user portraits and identify users’ health status, product and service preferences, and needs based on their medical and health data. The database management platform builds a product and service data warehouse based on smart medical product and service data, and uses association analysis algorithms to mine the potential connections between different product services and classify them; The model training and recommendation layer uses a full-process automatic learning framework to build a multi-dimensional AI recommendation model. The full-process automatic learning framework includes a feature extraction module, a model selection module, a model training module, and a verification and tuning module. The multi-dimensional AI recommendation model evaluates the matching degree between users and different types of products and services based on the user's personalized portrait and the information in the product and service data warehouse; The result sorting and display layer performs different types of sorting and comprehensive sorting on the personalized product and service recommendation results, and uses a user interaction interface to display the generated personalized product and service recommendation result list to the user, and then collects user feedback.

Images