Intelligent recommendation system for health management

Abstract

The invention discloses an intelligent recommendation system for health management, which belongs to the technical field of intelligent medical treatment and comprises an acquisition module, a knowledge analysis module, a calculation processing module and an interaction module. The system collects original health data, manages a health knowledge graph and a safety recommendation rule base, constructs an initial health portrait by using the original health data, generates a first recommendation list and a second recommendation list based on the initial health portrait and a logic rule base, generates a recommendation card through the first recommendation list and the second recommendation list, and sends the recommendation card to a user. And collecting user behavior data for feedback. According to the method, the problems of high cold start difficulty, insufficient safety guarantee and limited personalized effect in the existing health recommendation service are solved, the initial availability and recommendation scientificity of the system are improved, the health behavior of the user can be effectively guided, and the method has positive significance in improving the health level of people.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart medical treatment, and in particular to an intelligent recommendation system for health management. BACKGROUND

[0002] The existing recommendation system relies heavily on user behavior data, which causes new users to face the cold start dilemma and cannot obtain effective personalized recommendations. At the same time, most recommendation models lack deep integration of medical safety rules, making it difficult to ensure the safety of recommended content for users with specific health conditions, and there is a potential risk. In addition, the system often cannot dynamically integrate scientific health knowledge and real-time user preferences, resulting in either stereotyped and rigid recommendations or blindly catering to the recommendations and deviating from the health goal. These defects limit the application effect of the system in large-scale health guidance, and an intelligent solution that can balance safety, personalization and dynamic evolution ability is urgently needed.

[0003] At present, a health information intelligent recommendation system and method based on big data are disclosed in Chinese Patent No. CN117877676B. The present application provides a health recommendation system, which includes a client and a management server. The client receives user health information retrieval requests and sends them to the server. The server has a health database and a search engine, which can accurately recommend dietary programs for different constitutions, chronic diseases and special populations by matching traditional Chinese medicine constitution test data, traditional Chinese medicine attribute food and recipe library, and intelligently display food and recipe taboos. Users can conveniently use mobile APP or WeChat mini program. However, the related technology does not solve the cold start problem by integrating a health knowledge graph and a safety recommendation rule library, and the recommendations lack safety and scientificity, cannot solve the connection problem between cold start and long-term personalization through dynamic self-adaptive double-rule smoothing, lack continuity and growth, and cannot continuously optimize health guidance strategies as user status changes, limiting the effect of large-scale application in population health management. SUMMARY

[0004] The technical problem solved by the present application is that related technology does not solve the cold start problem by integrating a health knowledge graph and a safety recommendation rule library, and the recommendations lack safety and scientificity, cannot solve the connection problem between cold start and long-term personalization through dynamic self-adaptive double-rule smoothing, lack continuity and growth, and cannot continuously optimize health guidance strategies as user status changes, limiting the effect of large-scale application in population health management.

[0005] To solve the above technical problems, the present application provides the following technical solutions: An intelligent recommendation system for health management includes a collection module, a knowledge analysis module, a calculation processing module and an interaction module: The collection module is used to collect the original health data of the user; The knowledge analysis module is used for storing and managing a health knowledge graph and a safety recommendation rule base; The computing processing module is used for constructing an initial health portrait based on user original health data, and generating a first recommendation list and a second recommendation list based on the initial health portrait and a logic rule base; The interaction module is used for generating a recommendation card and collecting user feedback data.

[0006] As a preferred scheme of the intelligent recommendation system for health management, the original health data includes user declaration data, user measurement data and third-party data: The user declaration data includes basic information, health goals, health condition constraints, lifestyle preferences and psychological preferences; The basic information includes gender, age, height, weight, blood type, occupation type, previous major illness history, drug allergy history and family genetic disease history; The user measurement data includes index data and behavior data; The index data includes heart rate, blood pressure, blood glucose, blood oxygen saturation and body temperature; The behavior data includes exercise duration, exercise intensity, sleep duration, sleep quality, steps and calorie consumption; The third-party data includes activity data and physiological data; The activity data includes exercise course completion records, diet check-in records and health activity participation records; The physiological data includes electronic health records, physical examination report data and medical diagnosis records.

[0007] As a preferred scheme of the intelligent recommendation system for health management, the knowledge analysis module includes a health knowledge graph unit and a rule auditing unit; The knowledge graph unit is used for storing and managing a health knowledge graph, the health knowledge graph includes entities and relationships, the entities include user portrait entities, health content entities, physiological medical entities and resource entities, and the relationships include hierarchical relationships, causal relationships, dependency relationships, contraindications and beneficial relationships; The management includes adding, deleting, modifying and querying operations on entities and relationships; The rule auditing unit is used for storing a safety recommendation rule base, auditing all new health content schemes input into the safety recommendation rule base, and generating health content tags for the health content schemes that pass the audit, the safety recommendation rule base includes first rules and second rules, and the health content schemes include exercise courses, diet recipes and health knowledge content.

[0008] As a preferred scheme of the intelligent recommendation system for health management, the user portrait entity includes user and user group paradigm. The health content entity includes exercise, exercise course, food and recipe. The physiological medical entity includes physical indicators, nutrients, symptoms and diseases. The resource entity includes equipment and venue.

[0009] As a preferred scheme of the intelligent recommendation system for health management, the rule form of the first rule and the second rule includes a condition part and an action part. The first rule has a first execution priority, wherein the condition part is a logical combination of the first feature label and the health content label, and the action part is an instruction for prohibiting the recommendation of specific health content solutions to the corresponding user. The second rule has a second execution priority, wherein the condition part is a logical judgment statement based on the user state and the health content label, and the action part is an instruction for de-weighting the first recommendation list or the second recommendation list.

[0010] As a preferred scheme of the intelligent recommendation system for health management, the computing processing module includes a user portrait unit and a recommendation unit. The user portrait unit is used to analyze the user declaration data to obtain corresponding health labels, and to standardize and analyze the physiological data, and to calculate the user's basic health indicator feature value based on a logic rule library. The health content label and the basic health indicator feature value are used as query conditions to query associated entities and relationships in a health knowledge graph, calculate semantic correlation degree to obtain an initial health portrait, and the initial health portrait is composed of a first feature label, which is used to represent the user's health status, ability level, resource constraint, user state and personal preference. The logic rule library includes a calculation formula for calculating body mass index (BMI) value according to user's height and weight, a judgment rule for determining user activity level grade according to the activity data, and a reasoning rule for inferring user recommendation contraindication and priority recommendation direction according to the health target and the health condition constraint. The recommendation unit includes a first recommendation subunit, a second recommendation subunit and a recommendation decision subunit. The first recommendation subunit is used to generate a first recommendation list by matching the initial health portrait with the health knowledge graph when the user behavior data amount threshold is not reached. The second recommendation subunit is configured to, when the user behavior data amount threshold is reached, generate a second recommendation list based on the user behavior data by calculating the basic health indicator characteristic value and the second recommendation score of the health content label; The recommendation decision subunit is configured to monitor the user behavior data amount, define a user behavior data amount threshold condition required for switching from the first recommendation subunit to the second recommendation subunit, and calculate the first recommendation score output by the first recommendation subunit and the second recommendation score output by the second recommendation subunit, according to the increase of the user behavior data amount, reduce the weight coefficient of the first recommendation score, and increase the weight coefficient of the second recommendation score, the weight coefficient being the proportion of the first recommendation score and the second recommendation score; The user behavior data includes explicit feedback data and implicit feedback data; The explicit feedback data includes a score, a like label, a dislike label, a completion status mark, and a text comment. The implicit feedback data includes content browsing duration, play progress, pause operation, continue operation, skip operation, and repeated access number.

[0011] As a preferred scheme of the intelligent recommendation system for health management, the first recommendation specifically includes: performing cosine similarity calculation on the initial health portrait of a new user and a user health paradigm, matching one or more user health paradigms for the user, obtaining a benchmark health scheme bound to the user health paradigm, and adding the benchmark health scheme to a recommendation candidate set. The first feature label is received, and traversal query is performed in the health knowledge graph to find the health content entity associated with the first feature label. The benchmark health scheme and the health content scheme corresponding to the health content entity are combined to form a first recommendation candidate set. The recommendation candidate set is subjected to first rule verification and second rule verification to filter the benchmark health scheme or the health content scheme that violates the first rule and the second rule, and a first recommendation list is formed. The user health paradigm is compared with the content scheme label to obtain a matching result, the matching result and a corresponding weight coefficient are calculated to obtain a first recommendation score, and the matching result includes complete matching, partial matching, and no matching. The second recommendation specifically includes: standardizing and encoding the basic health indicator characteristic value and the first feature label to generate a user feature vector, standardizing and encoding the health content label to generate a content feature vector, and constructing an interaction matrix including the user feature vector and the content feature vector. Based on the interaction matrix, the similarity between users or the cosine similarity between contents is calculated to obtain a calculation result, and a second recommendation candidate set of potential health content schemes of interest to the user is screened according to the calculation result; The basic health index characteristic value and the numerical characteristic corresponding to the health content label are taken as inputs, and a factor decomposition machine model is used to generate a second recommendation score of the user for the health content scheme; The contents in the second recommendation candidate set are sorted in descending order of the second recommendation score according to the user behavior data, and a second recommendation list is formed.

[0012] As a preferred scheme of the intelligent recommendation system for health management, the interaction module comprises a display unit and a feedback unit. The display unit is configured to receive the first recommendation list and the second recommendation list, generate and display a recommendation card, the recommendation card comprising a content title, a picture and an intensity level, and pop up a corresponding health content entity and a contraindication relationship when a user selects one of the recommendation cards. The feedback unit is configured to receive user behavior data and send the user behavior data to the user portrait unit and the recommendation unit.

[0013] As a preferred scheme of the intelligent recommendation system for health management, the user portrait unit is further configured to receive the explicit feedback data and the implicit feedback data, and adjust the characteristic values in the initial health portrait and the weight values of the characteristic values based on an update strategy. The update strategy comprises explicit feedback updating and implicit feedback updating. The explicit feedback updating is configured to directly strengthen or weaken a related first characteristic label in the initial health portrait according to a user's active score or completion evaluation of the first recommendation list or the second recommendation list. The implicit feedback updating is configured to analyze user behavior data of the user for the first recommendation list or the second recommendation list, infer changes in user preferences and health status to obtain an inference result, and adjust the characteristic weight of a corresponding characteristic label in the initial health portrait based on the inference result.

[0014] As a preferred scheme of the intelligent recommendation system for health management, the initial health portrait adopts a vector representation method, wherein each dimension represents a first characteristic label and a corresponding characteristic weight.

[0015] The present application has the beneficial effects that: the present application ensures the safety and scientificity of the recommended content in medicine by constructing a health knowledge graph and utilizing semantic calculation and association mining, thereby providing reliable guidance for improving user health. By calculating the similarity between the user portrait and the standardized norm, a safe and relevant initial recommendation scheme is quickly generated, greatly improving the initial experience and trust of new users. With the accumulation of user data, the system realizes a smooth transition from rule-driven to data-driven, making the recommendation results more and more accurate and personalized. A closed-loop intelligent health guidance system is formed, significantly improving the user's health behavior compliance, and providing long-term, dynamic and scientific health improvement support for people in different health states. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A basic flowchart of an intelligent recommendation system for health management is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0018] Embodiments, with reference to Figure 1 For an embodiment of the present application, an intelligent recommendation system for health management is provided, including a collection module, a knowledge analysis module, a calculation processing module and an interaction module: The collection module is used to collect the original health data of the user; The knowledge analysis module is used to store and manage the health knowledge graph and the safe recommendation rule library; The calculation processing module is used to construct an initial health portrait based on the original health data of the user, and generate a first recommendation list and a second recommendation list based on the initial health portrait and the logical rule library; The interaction module is used to generate a recommendation card and collect user feedback data.

[0019] The present application constructs a health knowledge graph, combines semantic calculation and association mining technology, and ensures the safety and scientificity of the recommended content from the medical dimension, thereby providing reliable guidance for user health management. The system can quickly generate a safe and matching initial recommendation scheme by calculating the similarity between the user portrait and the standardized health norm, effectively improving the first-time use experience and trust of new users. With the continuous accumulation of user behavior data, the system can realize a smooth transition from rule-driven to data-driven, making the recommendation results gradually evolve towards precision and individualization. Finally, a closed-loop intelligent health guidance system is formed, significantly enhancing the user's compliance with health behaviors, and providing long-term, dynamic and scientific health improvement support for people in different health states.

[0020] The original health data includes user declared data, user measured data and third-party data: The user declared data includes basic information, health goals, health condition constraints, lifestyle preferences and psychological preferences; The basic information includes gender, age, height, weight, blood type, occupation type, previous major illness history, drug allergy history and family genetic disease history; The user measured data includes index data and behavior data; The index data includes heart rate, blood pressure, blood sugar, blood oxygen saturation and body temperature; The behavior data includes exercise duration, exercise intensity, sleep duration, sleep quality, step count and calorie consumption; The third-party data includes activity data and physiological data; The activity data includes exercise course completion records, diet check-in records and health activity participation records; The physiological data includes electronic health records, physical examination report data and medical diagnosis records.

[0021] In specific embodiments, the collection module provides a data basis for subsequent user portrait construction and precise recommendation. These multi-source data actively provided by users, classified and structured together constitute the perception basis of the system, so that the subsequent modules can accurately understand the user's state, constraints and needs, and thus calculate and generate safe and personalized exercise course or diet recipe recommendation schemes.

[0022] The knowledge analysis module includes a health knowledge graph unit and a rule review unit; The knowledge graph unit is used to store and manage the health knowledge graph, and the health knowledge graph includes entities and relationships, the entities include user portrait entities, health content entities, physiological medical entities and resource entities, and the relationships include hierarchical relationships, causal relationships, dependency relationships, contraindications and beneficial relationships; The user portrait entities include user and user group paradigms; The health content entities include exercise, exercise course, food and recipe; The physiological medical entities include body indicators, nutrients, symptoms and diseases; The resource entities include equipment and venues.

[0023] In specific embodiments, health content entities and physiological and medical entities are bound together through taboo or beneficial relationships. High-intensity exercise courses (health content entities) are taboo to hypertension (physiological and medical entities), while low-salt diets (health content entities) are beneficial to hypertension (physiological and medical entities). Physiological and medical entities are interconnected through causal relationships. High-salt diets (physiological and medical entities - nutrient-related) are causative to hypertension (physiological and medical entities - disease), and obesity (physiological and medical entities - body indicators-related) is inducing to diabetes (physiological and medical entities - disease).

[0024] Management includes adding, deleting, modifying, and querying entities and relationships; The rule review unit is used to store the safety recommendation rule base, review all new health content proposals entered into the safety recommendation rule base, and generate health content tags for approved health content proposals. The safety recommendation rule base includes first rules and second rules, and health content proposals include exercise courses, diet recipes, and health knowledge content.

[0025] In a specific embodiment, health content schemes are input into the rule review unit of the knowledge analysis module through a standardized interface. These schemes can be entered by the system administrator in the backend, submitted by third-party health service organizations through authorized interfaces, or created and shared by users through specific channels. The rule review unit calls the safety recommendation rule library to review each newly input health content scheme. For example, a diet plan for gout patients submitted by a third-party organization will be automatically verified for safety based on conditions related to high purines and gout contraindications in the safety recommendation rule library. After approval, a standardized health content tag will be generated for the scheme, and an association will be established with relevant entities in the knowledge graph, thereby ensuring that all recommended content complies with safety specifications and facilitates accurate matching in the future.

[0026] The knowledge analysis module integrates and manages multi-dimensional health knowledge through health knowledge graph units. For example, when the system needs to determine whether a dessert recipe is suitable for a diabetic user, it can quickly make a rejection decision by querying the contraindications between diabetes and high-sugar foods in the knowledge graph. This ensures the medical safety and personalized suitability of all recommended content, providing knowledge support and risk control guarantees for generating reliable recommendations in the future.

[0027] The first and second rules consist of a condition part and an action part. The first rule has a first execution priority, wherein the condition part is a logical combination of the first feature tag and the health content tag, and the action part is an instruction to prohibit the recommendation of specific health content programs to the corresponding users; The second rule has a second execution priority, wherein the condition part is a logical judgment statement based on the user's status and health content tags, and the action part is an instruction to demote the first recommendation list or the second recommendation list.

[0028] In a specific embodiment, the first and second rules construct a two-tiered security protection and personalized adjustment mechanism for the system through conditional and action-based rules. The first rule, as the highest-priority mandatory security barrier, directly matches user characteristics and content tags (e.g., user tags include "pregnancy" and content tags include "high intensity"). Once triggered, the action part completely prohibits the recommendation of that content, serving as a recommendation boundary and effectively preventing health risks. The second rule, as a secondary-priority flexible adjustment mechanism, makes logical judgments based on the user's real-time status and content tags. For example, if a user has recently experienced poor sleep quality and encounters a course with a "vigorous exercise" tag, the action part, once triggered, downgrades the corresponding content in the recommendation list. This allows the system to further optimize the personalized experience while ensuring security. The collaborative work of the two-tiered rules ensures that the recommendation results are both safe and aligned with the user's dynamic needs.

[0029] The computational processing module includes a user profiling unit and a recommendation unit; The user profiling unit is used to parse the user's declared data, obtain the corresponding health tags, standardize and analyze the physiological data, and calculate the user's basic health indicator feature values ​​based on the logical rule base. In a specific embodiment, trend analysis of user physiological data enables dynamic monitoring of health status and adaptive adjustment of recommendation strategies. Trend analysis, based on time series data processing technology, can identify long-term patterns of change in user physiological indicators.

[0030] Using health content tags and basic health indicator feature values ​​as query conditions, related entities and relationships are queried in the health knowledge graph, and the semantic relevance is calculated to obtain an initial health profile. The initial health profile is composed of a first feature tag, which is used to represent the user's health status, ability level, resource constraints, user status, and personal preferences. In a specific embodiment, the semantic relevance is calculated by mapping health content tags and basic health indicator feature values ​​to entities in a health knowledge graph, extracting the relationship paths between entities, assigning basic weights according to the relationships, and accumulating the weights by combining the relationship path length attenuation coefficient.

[0031] The logic rule base includes the calculation formula for calculating the body mass index (BMI) based on the user's height and weight, the judgment rules for determining the user's activity level based on activity data, and the reasoning rules for inferring user recommendation taboos and priority recommendation directions based on health goals and health status constraints. In a specific embodiment, the formula for calculating the Body Mass Index (BMI) is as follows: BMI = weight (kg) / [height (m)]²; The system automatically calls this formula to calculate the user's BMI value based on the height and weight values ​​in the user's declared data. According to the threshold range preset by the World Health Organization or relevant national health standards, BMI < 18.5 is "underweight", 18.5 ≤ BMI < 24 is "normal", 24 ≤ BMI < 28 is "overweight", and BMI ≥ 28 is "obese". It automatically generates characteristic labels such as obesity, overweight or underweight for the user profile. These labels are the key quantitative basis for subsequent assessment of health risks and recommendation of fat loss or muscle gain programs.

[0032] The user activity level assessment rule is based on a pre-defined logic using objective user measurement data and behavioral data from third-party sources. For example, if a user's average daily steps over the past seven days are less than 5,000, their activity level is classified as sedentary; if it's between 5,000 and 10,000 steps and they engage in moderate-intensity exercise 1-2 times per week, they are classified as lightly active; and if they exceed 10,000 steps and engage in regular high-intensity training, they are classified as highly active. This system dynamically assigns an activity level label to each user based on these rules, and this label is directly used to calculate their recommended daily calorie intake and exercise intensity.

[0033] The recommendation taboo and priority direction reasoning rules are a set of reasoning logic based on knowledge graphs and security rules. This logic derives absolute recommendation taboos and priority directions based on the user's health goals and health status constraints. For example, a rule can be defined as: IF The user's health status constraints include "kidney disease" AND the health content tag includes "high protein" THEN Execute the "prohibit recommendation" action. Accordingly, the system will automatically exclude any high-protein diet recipes from the user's recommendation candidate set.

[0034] The recommendation unit includes a first recommendation subunit, a second recommendation subunit, and a recommendation decision subunit; The first recommendation subunit is used to generate a first recommendation list by matching the initial health profile with the health knowledge graph when the user behavior data volume threshold is not reached. The second recommendation subunit is used to generate a second recommendation list based on user behavior data when the user behavior data volume threshold is reached, by calculating the second recommendation score of the basic health indicator feature value and the health content tag. The recommendation decision subunit is used to monitor the amount of user behavior data, define the threshold condition of the amount of user behavior data required to switch from the first recommendation subunit to the second recommendation subunit, and calculate the first recommendation score output by the first recommendation subunit and the second recommendation score output by the second recommendation subunit. According to the increase of the amount of user behavior data, the weight coefficient of the first recommendation score is reduced and the weight coefficient of the second recommendation score is increased. The weight coefficient is the ratio of the first recommendation score to the second recommendation score. User behavior data includes explicit feedback data and implicit feedback data.

[0035] In a specific embodiment, the recommendation unit achieves a seamless transition and intelligent decision-making process from cold start for new users to personalized recommendations for mature users. For example, when a new user registers and lacks behavioral data, the first recommendation subunit generates a safe and reliable first recommendation list (recommending low-sodium recipes and brisk walking courses) by matching the user's initial health profile (including tags for high blood pressure and weight loss) with a health knowledge graph. As the user continues to use the system, it continuously collects explicit feedback (course ratings) and implicit feedback (recipe browsing time). When the amount of user behavioral data reaches a preset threshold (30 cumulative behavioral records), the recommendation decision subunit automatically activates the second recommendation subunit. The latter discovers that users similar to John like a certain yoga course and calculates a second recommendation score based on John's recent sleep data to predict the appropriate exercise intensity for him, generating a more personalized second recommendation list. At the same time, the recommendation decision subunit implements a smooth weight transfer strategy. For example, the initial weight of the first recommendation score is 100%, which is gradually reduced to 30% as the amount of data increases, while the weight of the second recommendation score increases from 0% to 70%. Finally, the two are merged to form the final recommendation, thereby ensuring that the recommendation results are both safe and reliable in the early stages and accurate and personalized in the later stages.

[0036] The first recommendation specifically includes: calculating the cosine similarity between the initial health profile of the new user and the user health paradigm, matching one or more user health paradigms for the user, obtaining a baseline health plan bound to the user health paradigm, and adding the baseline health plan to the recommendation candidate set; Receive the first feature tag and perform a traversal query in the health knowledge graph to find health content entities associated with the first feature tag; The benchmark health plan and the health content plan corresponding to the health content entity are merged to form the first recommendation candidate set; The candidate recommendation set is validated by the first rule and the second rule, and the baseline health schemes or health content schemes that violate the first rule and the second rule are filtered to form the first recommendation list; The user health paradigm is compared with the content scheme tags to obtain the matching results. The matching results and the corresponding weight coefficients are calculated to obtain the first recommendation score. The matching results include complete match, partial match, and no match. In a specific implementation, the user health paradigm is a predefined standardized template representing a typical population with shared health characteristics and needs. These paradigms are developed by health experts based on clinical guidelines and population data analysis, and each paradigm is tied to a set of safe baseline health protocols.

[0037] In the health knowledge graph, the system starts with user feature tags and traverses along relational edges within the graph. For example, starting from the diabetes node, it finds high-sugar food entities along the contraindication relation, then locates specific recipes through the belonging relation. Through multi-hop queries, it mines health content entities that are directly or indirectly related to user features, expands the candidate set, and ensures the relevance and safety of recommended content in terms of medical logic.

[0038] The system compares the characteristics of a user's health paradigm with content solution tags, classifying the match as: perfect match, partial match, or no match. Each level is assigned a fixed weight coefficient of 1.0, 0.6, and 0.1, respectively, and the initial recommendation score is calculated using a weighted average. The recommendation list is then sorted by score, prioritizing the display of solutions that best match the user's paradigm to ensure the rationality and interpretability of the initial recommendations.

[0039] The second recommendation specifically includes: standardizing and encoding the basic health indicator feature values ​​and the first feature label to generate a user feature vector; standardizing and encoding the health content label to generate a content feature vector; and constructing an interaction matrix, wherein the interaction matrix includes the user feature vector and the content feature vector. Based on the interaction matrix, the similarity between users or the cosine similarity between content is calculated to obtain the calculation results. Based on the calculation results, a second set of recommended candidate schemes that users may be interested in is selected. Using the basic health indicator feature values ​​and the numerical features corresponding to health content tags as input, a factor decomposition machine model is used to generate a second recommendation score for the user's health content plan. Based on user behavior data, the content in the second recommendation candidate set is sorted from high to low according to the second recommendation score to form a second recommendation list.

[0040] In a specific embodiment, by calculating the cosine of the angle between user feature vectors, user groups with similar health needs and preferences are identified; or by calculating the similarity between content feature vectors, health solutions with similar attributes are found. This process helps the system filter out potentially interesting content, forming a recommendation candidate set.

[0041] The user's basic health indicator features and the numerical features of health content tags are standardized and encoded, and then concatenated into a feature vector x. This feature vector x is then input into a factorization machine model, and the prediction equation is as follows: ; Where y(x) is the output second recommendation score, and w0 is the global bias term. This is the first-order linear part, representing the independent influence of each feature on the result. This is the second-order interaction part, used to model the combined effects between pairs of features. i and v j These are the latent vectors of features i and j. This represents the dot product of vectors. Feature interactions are learned through the dot product of latent vectors, and the function y(x) calculates a real-valued score, i.e., the second recommendation score.

[0042] The interactive module includes a display unit and a feedback unit; The display unit is used to receive the first recommendation list and the second recommendation list, generate and display recommendation cards. The recommendation cards include content titles, images and intensity levels. When the user selects one of the recommendation cards, the corresponding healthy content entities and taboo relationships will pop up. The feedback unit is used to receive user behavior data and send it to the user profile unit and the recommendation unit. Explicit feedback data includes ratings, like tags, dislike tags, completion status markers, and text comments; Implicit feedback data includes content viewing time, playback progress, pause operations, resume operations, skip operations, and number of repeated visits.

[0043] In a specific embodiment, the interaction module realizes the visualization of recommendation results and the closed-loop collection of user feedback. After receiving the recommendation list, the display unit converts it into intuitive recommendation cards. When the user clicks on the card, the unit extracts and displays the entity details and contraindications related to the content from the health knowledge graph, helping the user understand the reasons for the recommendation and potential risks. The feedback unit captures user interaction behavior in real time. Explicit feedback includes users giving a 5-star rating to a course or marking it "disliked," while implicit feedback includes users skipping the course video after only watching the first 30 seconds or repeatedly browsing a recipe details page more than 3 times. All this data is immediately sent to the user profile unit for dynamically updating user feature tags and weights, and synchronized to the recommendation unit to optimize subsequent recommendation ranking.

[0044] The user profile unit is also used to receive explicit and implicit feedback data, and adjust the feature values ​​and feature weights in the initial health profile based on the update strategy. Update strategies include explicit feedback updates and implicit feedback updates; Explicit feedback updates are used to directly strengthen or weaken the relevant primary feature labels in the initial health profile based on the user's active rating or completion evaluation of the first or second recommendation list. Implicit feedback updates are used to analyze user behavior data on the first or second recommendation list, infer changes in user preferences and health status, obtain inference results, and adjust the feature weights of corresponding feature tags in the initial health profile based on the inference results.

[0045] The initial health profile uses a vector representation, where each dimension represents a first feature label and its corresponding feature weight.

[0046] In a specific embodiment, the user profiling unit receives and processes feedback data from the interaction module to dynamically optimize the accuracy and real-time performance of the user profile. For example, when a user gives a low-fat recipe a 5-star rating, the explicit feedback update strategy will directly strengthen the weight of the low-fat healthy diet tag in their profile. At the same time, if the system detects that a user has repeatedly abandoned high-intensity courses and repeatedly performed medium-intensity courses, the implicit feedback update strategy will infer changes in the user's physical condition or preferences, thereby weakening the weight of the high-intensity exercise tolerance tag in the profile and correspondingly increasing the weight of the medium-intensity appropriate tag. The system can capture user preferences and status changes in real time, providing accurate and adaptive profile basis for subsequent recommendations.

[0047] This invention achieves accurate perception of users' health needs and personalized content recommendations through the coordinated operation of a data collection module, a knowledge analysis module, a computational processing module, and an interaction module. First, it integrates multi-source health data to construct a dynamically updated user health profile. Then, it uses a health knowledge graph and a safety rule base to conduct medical compliance review and semantic association mining of recommended content. Based on this, the system employs a dual-rule recommendation mechanism, combining paradigm matching in the cold start phase with collaborative filtering when data is sufficient, to generate safe and personalized exercise, diet, and knowledge plans. Finally, the system uses an interactive interface to visualize the recommendation results and collect user feedback in real time, forming a closed-loop optimization process. This invention ensures that the recommended content throughout the entire lifecycle, from the cold start phase to the mature recommendation phase, possesses medical safety and personalized adaptability, forming a closed-loop intelligent health guidance system that can sustainably improve users' health behavior adherence and health level.

[0048] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0049] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent recommendation system for health management, characterized in that, The system comprises a collection module, a knowledge analysis module, a calculation processing module, and an interaction module. The collection module is configured to collect original health data of a user. The knowledge analysis module is configured to store and manage a health knowledge graph and a safety recommendation rule base. The calculation processing module is configured to construct an initial health portrait based on the original health data of the user, and generate a first recommendation list and a second recommendation list based on the initial health portrait and the logic rule base. The interaction module is configured to generate a recommendation card and collect user behavior data.

2. The intelligent recommendation system for health management as claimed in claim 1 wherein: The original health data includes user declared data, user measured data, and third-party data. The user declared data includes basic information, health goals, health condition constraints, lifestyle preferences, and psychological preferences. The basic information includes gender, age, height, weight, blood type, occupation type, previous major medical history, drug allergy history, and family genetic disease history. The user measured data includes index data and behavior data. The index data includes heart rate, blood pressure, blood glucose, blood oxygen saturation, and body temperature. The behavior data includes exercise duration, exercise intensity, sleep duration, sleep quality, steps, and calorie consumption. The third-party data includes activity data and physiological data. The activity data includes exercise course completion records, diet check-in records, and health activity participation records. The physiological data includes electronic health records, physical examination report data, and medical diagnosis records.

3. The intelligent recommendation system for health management as claimed in claim 2 wherein: The knowledge analysis module includes a health knowledge graph unit and a rule review unit. The knowledge graph unit is configured to store and manage a health knowledge graph, which includes entities and relationships. The entities include user portrait entities, health content entities, physiological medical entities, and resource entities. The relationships include hierarchical relationships, causal relationships, dependency relationships, contraindication relationships, and beneficial relationships. The management includes addition, deletion, modification, and query operations on entities and relationships. The rule review unit is configured to store a safety recommendation rule base, review all new health content solutions input into the safety recommendation rule base, and generate health content tags for health content solutions that pass the review. The safety recommendation rule base includes first rules and second rules. The health content solutions include exercise courses, diet recipes, and health knowledge content.

4. The intelligent recommendation system for health management of claim 3, wherein: The user portrait entities include user and user group paradigms. The health content entities include exercise, exercise courses, food, and recipes. The physiological medical entities include physical indicators, nutrients, symptoms, and diseases. The resource entities include equipment and venues.

5. The intelligent recommendation system for health management of claim 4, wherein: The rule forms of the first rules and the second rules include condition parts and action parts. The first rules have a first execution priority, wherein the condition part is a logical combination of a first feature tag and a health content tag, and the action part is an instruction for prohibiting the recommendation of a specific health content solution to a corresponding user. The second rule has a second execution priority, wherein the condition part is a logical judgment statement based on a user state and a health content label, and the action part is an instruction to reduce the weight of the first recommendation list or the second recommendation list.

6. The intelligent recommendation system for health management as claimed in claim 5 wherein: The computing processing module comprises a user portrait unit and a recommendation unit; The user portrait unit is configured to parse the user declaration data to obtain corresponding health labels, and to perform standardization processing and trend analysis on the physiological data, and to calculate a user's basic health index characteristic value based on a logic rule library; The health content label and the basic health index characteristic value are taken as query conditions to query associated entities and relationships in a health knowledge graph, calculate a semantic correlation degree to obtain an initial health portrait, and the initial health portrait is composed of a first characteristic label, and the first characteristic label is used to represent a user's health status, ability level, resource constraint, user state and personal preference; The logic rule library comprises a calculation formula for calculating a body mass index (BMI) value according to a user's height and weight, a judgment rule for determining a user activity level grade according to the activity data, and an inference rule for inferring a user's recommended taboo and priority recommendation direction according to the health goal and the health condition constraint; The recommendation unit comprises a first recommendation subunit, a second recommendation subunit and a recommendation decision subunit; The first recommendation subunit is configured to generate a first recommendation list by matching the initial health portrait with the health knowledge graph when a user behavior data amount threshold is not reached; The second recommendation subunit is configured to generate a second recommendation list by calculating a second recommendation score of the basic health index characteristic value and the health content label based on user behavior data when the user behavior data amount threshold is reached; The recommendation decision subunit is configured to monitor the user behavior data amount, define a user behavior data amount threshold condition required for switching from the first recommendation subunit to the second recommendation subunit, and calculate a first recommendation score output by the first recommendation subunit and a second recommendation score output by the second recommendation subunit, and according to an increase in the user behavior data amount, reduce a weight coefficient of the first recommendation score and increase a weight coefficient of the second recommendation score, wherein the weight coefficient is a proportion of the first recommendation score and the second recommendation score; The user behavior data comprises explicit feedback data and implicit feedback data; The explicit feedback data comprises a score, a like label, a dislike label, a completion status mark and a text comment; The implicit feedback data comprises content browsing duration, play progress, pause operation, continue operation, skip operation and repeated access times.

7. The intelligent recommendation system for health management according to claim 6, wherein: The first recommendation specifically comprises: calculating the similarity of the initial health portrait of a new user with a user health paradigm, and matching one or more user health paradigms for the user to obtain a benchmark health scheme bound with the user health paradigm, and adding the benchmark health scheme to a recommendation candidate set; The first characteristic label is received and iteratively queried in the health knowledge graph to obtain the health content entity associated with the first characteristic label; merge the benchmark health plan and the health content plan corresponding to the health content entity to form a first recommendation candidate set; perform first rule verification and second rule verification on the recommendation candidate set, filter benchmark health plans or health content plans that violate the first rule and the second rule to form a first recommendation list; compare the user health paradigm with the content plan label to obtain a matching result, the matching result including complete match, partial match, and no match, calculate the matching result and the corresponding weight coefficient to obtain a first recommendation score; the second recommendation specifically includes: standardizing and encoding the basic health indicator characteristic value and the first characteristic label to generate a user feature vector, standardizing and encoding the health content label to generate a content feature vector, and constructing an interaction matrix including the user feature vector and the content feature vector; based on the interaction matrix, calculate the similarity between users or the cosine similarity between contents to obtain a calculation result, and filter potential health content plans of interest for the user based on the calculation result to form a second recommendation candidate set; input the basic health indicator characteristic value and the numerical value corresponding to the health content label into a factorization machine model to generate a second recommendation score of the user for the health content plan; sort the contents in the second recommendation candidate set from high to low according to the second recommendation score based on user behavior data to form a second recommendation list.

8. The intelligent recommendation system for health management as claimed in claim 7 wherein: The interaction module includes a display unit and a feedback unit. The display unit is configured to receive the first recommendation list and the second recommendation list, generate and display a recommendation card, the recommendation card including a content title, a picture, and an intensity level, and pop up the corresponding health content entity and contraindication relationship when the user selects one of the recommendation cards. The feedback unit is configured to receive user behavior data and send the user behavior data to the user portrait unit and the recommendation unit.

9. The intelligent recommendation system for health management according to claim 8, wherein: the user portrait unit is further configured to receive the explicit feedback data and the implicit feedback data, and adjust the feature values in the initial health portrait and the weight values of the feature values based on an update strategy; the update strategy includes explicit feedback update and implicit feedback update; the explicit feedback update is configured to directly strengthen or weaken the relevant first characteristic label in the initial health portrait according to the user's active rating or completion evaluation of the first recommendation list or the second recommendation list; the implicit feedback update is configured to analyze the user behavior data of the user for the first recommendation list or the second recommendation list, infer the changes in user preference and health status to obtain an inference result, and adjust the feature weight of the corresponding characteristic label in the initial health portrait based on the inference result.

10. The intelligent recommendation system for health management as claimed in claim 9 wherein: the initial health portrait adopts a vector representation method, wherein each dimension represents a first characteristic label and a corresponding feature weight.

Images