Household heart rehabilitation suggestion system based on user portrait

Abstract

The invention discloses a household heart rehabilitation suggestion system based on a user portrait, and relates to the technical field of digital medical systems. Comprising a user portrait database, a rehabilitation knowledge base, a content module, an event response module, a portrait matching module and a feedback adjustment module, and the user portrait database stores real-time signs, behavior feature vectors and historical feedback weight data; the rehabilitation knowledge base stores rehabilitation content through a map structure and a triple mark; the content generation module calculates a dynamic association weight through the matched data; the event response module monitors signs and triggers signals when the signs are abnormal; the portrait matching module ranks the rehabilitation content based on the trigger signal and the weight; and the feedback adjustment module updates the data according to the user operation response to realize dynamic iteration of the database. According to the system, through multi-module cooperation, precision, individuation and dynamic optimization of family heart rehabilitation suggestions are achieved, and effectiveness and suitability of rehabilitation guidance are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital medical systems, and in particular to a home cardiac rehabilitation recommendation system based on user portraits. Background Art

[0002] Existing outpatient management systems for CHF patients usually use fixed rules to push content after user profiles are constructed, such as recommending standard exercise programs according to disease stages. The weight distribution of feature tags relies on static values preset by experts and does not change dynamically according to patient data. For example, when compliance decreases, the reminder weight fails to be automatically increased. Database fuzzy retrieval is used to match content, and the response delay is high. It cannot meet the timeliness requirements of real-time symptom warnings such as sudden dyspnea. The data collection cycle is fixed, such as weekly updates, and there is no embedded real-time feedback drive mechanism, which causes the adjustment of rehabilitation programs to lag behind changes in patient status. For example, the existing technology recommends rehabilitation videos by establishing a patient feature tag library, but the tag weights are manually set by medical staff, and the push algorithm is not linked to device alarm events, resulting in a misalignment between the pushed content and the patient's immediate needs. Therefore, there is an urgent need for a home cardiac rehabilitation recommendation system based on user profiles to meet the more needs of CHF patients for the optimization of digital medical rehabilitation system configuration during outpatient home rehabilitation. Summary of the Invention

[0003] In response to the shortcomings of the above-mentioned existing technologies, the present invention provides a home cardiac rehabilitation recommendation system based on user portraits. Through a dynamically updated patient portrait library, it achieves accurate matching and hierarchical delivery of management support content and the patient's real-time status, and ensures that the weight of the pushed information tag is dynamically optimized with the treatment stage, behavioral data, and environmental risks.

[0004] The specific technical solutions are as follows:

[0005] The user portrait-based home cardiac rehabilitation recommendation system includes:

[0006] User portrait database, used to store patients' real-time vital sign data, behavioral feature vectors, and historical feedback weight data;

[0007] The rehabilitation knowledge base uses a graph structure to classify and store rehabilitation content. The rehabilitation content is marked by a triplet field. The triplet includes: a physical sign matching field, a behavioral feature threshold condition field, and a historical feedback weight field.

[0008] a content generation module, connected to the user portrait database and the rehabilitation knowledge base, configured to match the physical sign matching field and the behavioral feature threshold condition field of the triple with the user's real-time physical sign data and behavioral feature vector, and dynamically calculate the dynamic association weight based on the historical feedback weight field and the real-time data;

[0009] an event response module configured to monitor real-time vital sign data and generate an event trigger signal if abnormal vital signs exceeding a dynamic threshold are detected;

[0010] The portrait matching module uses an improved Trie tree structure and is configured to prioritize the triples in the rehabilitation knowledge base based on event trigger signals and dynamic association weights, and output the highest priority rehabilitation content;

[0011] The feedback adjustment module is configured to collect the user's operational response data to the rehabilitation suggestions, generate updated data of the behavioral feature vector, and adjust the value of the historical feedback weight field based on the updated data to achieve dynamic iteration of the user portrait database.

[0012] Furthermore, in the above solution, the calculation formula of the dynamic association weight is: W_new=W_base×(1+α×S_urgency)×e^(-λ×t);

[0013] Where: W_base is the normalized value of the historical feedback weight field, α is the urgency coefficient, ranging from 0.1 to 0.5, S_urgency is a comprehensive score based on the deviation of real-time physical sign abnormalities and behavioral characteristics, λ is the time attenuation coefficient, 0.01≤λ≤0.1, and t is the time interval of the user portrait database update cycle.

[0014] Furthermore, in the above scheme, the S_urgency calculation method includes: weighted summing of the abnormal confidence of real-time physical signs, the decline in the completion of behavioral characteristics, and the trend change value of historical feedback weight, and selecting the largest component as the final score.

[0015] Furthermore, the above scheme collects real-time vital sign data through wearable devices, including heart rate, blood pressure, and blood oxygen, with a sampling frequency greater than 10Hz. After wavelet denoising processing, it is linearly interpolated and resampled to 100Hz through a multi-source synchronization unit, and then output to the user portrait database after sliding window time domain alignment.

[0016] Furthermore, the output data packet of the multi-source synchronization unit includes a timestamp, a sensor ID, an original value, a threshold mean μ_t, and a standard deviation σ_t, and the dynamic threshold adjustment formula is: θ_t=μ_t+kσ_t;

[0017] Wherein, k is a preset parameter, k=2~39.

[0018] Furthermore, in the above solution, the behavior feature vector generation method includes:

[0019] The three-axis acceleration sensor and GPS positioning data are input into the motion pattern recognition module to identify motion patterns including walking, running and standing still, and generate behavioral feature threshold conditions;

[0020] The motion feature threshold conditions are bound and matched with the "sports content" in the rehabilitation knowledge base, where the threshold conditions include running time, walking steps and stationary time.

[0021] Furthermore, in the above solution, the feedback adjustment module further includes:

[0022] The daily behavior response analysis unit calculates α and β based on the click delay time Δt, execution completion degree η, and feedback score σ of the day;

[0023] The calculation formula of α is: α = (Ση_i) / n×(1-Δt_avg / 300);

[0024] Where n is the number of valid records on the day, n≥5, and Δt_avg is the average click delay time;

[0025] The calculation formula of β is: β = 0.3σ + 0.7(Σclick_count) / total_push;

[0026] V = [α, β, Δt_max, η_std, σ_trend, last_7d_avg], where: Δt_max is the maximum click delay time of the day; η_std is the standard deviation of the execution completion rate of the day; σ_trend is the linear slope of the rating trend in the past 7 days.

[0027] Furthermore, in the above solution, the triggering conditions for updating the user portrait database include:

[0028] Vital sign data exceeds the dynamic threshold for three consecutive times;

[0029] The cosine similarity between the behavior feature vector and the historical behavior pattern is less than 0.7;

[0030] The environmental sensor detects that the temperature exceeds 32°C or the PM2.5 concentration exceeds 100μg / m 3 .

[0031] Furthermore, in the above solution, the improved Trie tree structure includes a hierarchical weight index unit, and the weight value of the hierarchical weight index unit is driven in real time by the dynamic association weight output by the content generation module.

[0032] Furthermore, in the above scheme, the hierarchical weight indexing unit is configured to automatically trigger the reordering of the weight path and update the matching priority of the triples and the user portrait database when the fluctuation of the dynamic association weight exceeds a preset threshold.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention collects real-time vital sign data at high frequency through wearable devices, processes and stores it in depth, and builds personalized archives by combining behavioral feature vectors with historical feedback weight data; adopts graph structure and triple tagging to realize structured storage of rehabilitation knowledge; uses set formulas to calculate dynamic association weights, and comprehensively evaluates content adaptability with multi-dimensional data; with technical points such as real-time vital sign abnormality monitoring, improved Trie tree structure priority sorting, and dynamic iteration based on user feedback, it realizes accurate output of rehabilitation suggestions that match the patient's real-time health status and behavioral habits, thereby improving the effectiveness of the suggestions; relies on real-time vital sign monitoring and dynamic threshold judgment to respond to user needs in a timely manner; continuously optimizes system parameters based on user feedback, enhances the personalization of rehabilitation suggestions, and improves user acceptance and rehabilitation effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a functional framework diagram of the system of the present invention;

[0036] Figure 2 This is the functional framework diagram of the system after optimization of the present invention. DETAILED DESCRIPTION

[0037] The following is a further detailed description of the embodiments of the invention in conjunction with the accompanying drawings to make the objectives, technical solutions and technical effects of the invention more clearly presented.

[0038] like Figure 1-2 As shown in the figure, the user-profile-based home cardiac rehabilitation recommendation system of the present invention achieves precise health management for cardiac rehabilitation patients through the collaborative work of multiple modules. The system mainly includes a user profile database, a rehabilitation knowledge base, a content module, an event response module, a profile matching module, and a feedback adjustment module. Specifically, it includes:

[0039] The user portrait database is used to store patients' real-time vital sign data, behavioral feature vectors, and historical feedback weight data;

[0040] The rehabilitation knowledge base uses a graph structure to classify and store rehabilitation content. The rehabilitation content is marked by a triplet field. The triplet includes: a physical sign matching field, a behavioral feature threshold condition field, and a historical feedback weight field.

[0041] The content generation module is connected to the user portrait database and the rehabilitation knowledge base, and is configured to match the physical sign matching field and the behavioral feature threshold condition field of the triple with the user's real-time physical sign data and behavioral feature vector, and dynamically calculate the dynamic association weight based on the historical feedback weight field and the real-time data;

[0042] An event response module is configured to monitor real-time vital sign data and generate an event trigger signal if abnormal vital signs exceeding a dynamic threshold are detected;

[0043] The portrait matching module adopts an improved Trie tree structure and is configured to prioritize the triples in the rehabilitation knowledge base based on event trigger signals and dynamic association weights, and output the highest priority rehabilitation content;

[0044] The feedback adjustment module is configured to collect the user's operational response data to the rehabilitation suggestions, generate updated data of the behavioral feature vector, and adjust the value of the historical feedback weight field based on the updated data to achieve dynamic iteration of the user portrait database.

[0045] The user portrait database is used to store the patient's real-time vital sign data, behavioral feature vectors, and historical feedback weight data, building a dynamic data archive of the patient for the system, completely recording the patient's health status and behavioral patterns, providing an accurate data foundation for subsequent modules, and supporting the system to make rehabilitation recommendations based on the patient's actual situation.

[0046] Real-time vital sign data, including heart rate, blood pressure, and blood oxygen, is collected via wearable devices at a sampling frequency greater than 10Hz. After wavelet denoising, it is resampled to 100Hz using linear interpolation using a multi-source synchronization unit. After sliding window time domain alignment, it is output to the user profile database. The output data packet from the multi-source synchronization unit includes a timestamp, sensor ID, raw value, threshold mean μ_t, and standard deviation σ_t. The dynamic threshold adjustment formula is: θ_t = μ_t + kσ_t; k is a preset parameter ranging from 2 to 39.

[0047] Behavioral feature vectors are generated by inputting triaxial accelerometer and GPS positioning data into a motion pattern recognition module. This module identifies motion patterns, including walking, running, and stationary, and generates behavioral feature threshold conditions. These motion feature threshold conditions are then matched to "sports content" in the rehabilitation knowledge base. These threshold conditions include running duration, walking steps, and stationary time.

[0048] The historical feedback weight reflects a user's past acceptance and implementation of various rehabilitation suggestions, and changes in its value reflect user preferences and feedback tendencies. This is based on the long-term accumulation of user responses to rehabilitation suggestions. The feedback adjustment module collects multi-dimensional data, such as click delay time, execution completion, and feedback score. This data is then processed in depth by the daily behavioral response analysis unit, calculating key parameters such as α and β. This data then generates updated behavioral feature vectors, which are used to adjust the value of the historical feedback weight field.

[0049] The update trigger conditions of the user profile database include: (1) the physical sign data exceeds the dynamic threshold for three consecutive times; (2) the cosine similarity between the behavior feature vector and the historical behavior pattern is less than 0.7; (3) the environmental sensor detects that the temperature exceeds 32°C or the PM2.5 concentration exceeds 100 μg / m 3 .

[0050] The rehabilitation knowledge base uses a graph structure, categorizing and storing rehabilitation content and labeling it with a triplet consisting of a physical sign matching field, a behavioral feature threshold condition field, and a historical feedback weight field. Specifically, this massive amount of rehabilitation knowledge is organized in a structured and standardized manner to facilitate rapid retrieval and matching. This provides the system with a rich and organized rehabilitation knowledge resource, enabling it to accurately retrieve appropriate rehabilitation content from the knowledge base based on the patient's different physical signs and behavioral characteristics, laying a solid knowledge foundation for generating scientific and reasonable rehabilitation recommendations.

[0051] Rehabilitation content can include diet, exercise, psychological adjustment, and daily guidance, all in either video or text format. For example, diet content can include diet videos, diet recommendations, and diet prescriptions; exercise content can include exercise explanation videos and exercise tips; or it can be diet and exercise content compiled and organized by doctors based on their experience. This content can be continuously expanded and improved based on usage, which is an inevitable trend in the development of the system.

[0052] The content generation module connects the user portrait database and the rehabilitation knowledge base, and compares and matches the physical sign matching field and behavioral feature threshold condition field of the triples in the rehabilitation knowledge base with the real-time physical sign data and behavioral feature vectors in the user portrait database. It uses the formula W_new = W_base×(1+α×S_urgency)×e^(-λ×t) to dynamically calculate the dynamic association weight based on the historical feedback weight field and real-time data; it is used to achieve a deep association between rehabilitation content and the individual condition of the patient, and dynamically evaluate the adaptability of different rehabilitation content based on real-time data to ensure that the output rehabilitation suggestions are in line with the patient's current actual condition, thereby improving the pertinence and effectiveness of the suggestions.

[0053] Here, the calculation formula of the dynamic association weight is: W_new = W_base×(1+α×S_urgency)×e^(-λ×t);

[0054] Where: W_base is the normalized value of the historical feedback weight field, α is the urgency coefficient, ranging from 0.1 to 0.5, S_urgency is a comprehensive score based on the deviation of real-time physical sign abnormalities and behavioral characteristics, λ is the time attenuation coefficient, 0.01≤λ≤0.1, and t is the time interval of the user portrait database update cycle.

[0055] More specifically, the S_urgency calculation method includes: weighted summation of the abnormal confidence of real-time physical signs, the decline in the completion of behavioral characteristics, and the trend change value of historical feedback weights, and selecting the largest component as the final score.

[0056] The event response module monitors real-time vital sign data from the user profile database and generates an event trigger signal when an abnormal sign exceeds the dynamic threshold calculated by θ_t = μ_t + kσ_t. Specifically, through real-time data monitoring and dynamic threshold judgment, it promptly captures abnormal changes in the patient's body. It is used to quickly identify sudden changes in the patient's health. Once an abnormality is detected, subsequent processes are immediately triggered, enabling the system to respond quickly and provide patients with timely rehabilitation advice or intervention measures to ensure their health and safety.

[0057] The portrait matching module uses an improved Trie tree structure containing hierarchical weighted indexing units to prioritize the triplet content in the rehabilitation knowledge base based on the event trigger signal generated by the event response module and the dynamic association weight output by the content generation module. When the dynamic association weight fluctuation exceeds the preset threshold, it automatically re-sorts and outputs the highest priority rehabilitation content. By utilizing efficient data structures and dynamic weight calculations, rapid screening and prioritization of rehabilitation content can be achieved. When a patient experiences an abnormal condition or requires rehabilitation advice, it can quickly and accurately match the rehabilitation content that best suits the patient's current situation from the rehabilitation knowledge base, ensuring that the patient can receive the most effective rehabilitation guidance in a timely manner.

[0058] In the above, the improved Trie tree structure may include a hierarchical weight index unit, a node feature mapping module, a weight lower limit guarantee unit and a data synchronization interface.

[0059] The hierarchical weight indexing unit is used to receive and process dynamic association weights and adjust the priority of triple paths.

[0060] The node feature mapping module is used to encode the physical sign matching parameters, behavioral feature threshold parameters, and historical feedback weight fields into triple node identifiers.

[0061] The weight lower limit protection unit consists of a hardware comparator and a preset threshold register to ensure that the node activation weight is not lower than the set value.

[0062] The data synchronization interface is connected to the user portrait database and is used to periodically update the basic weights of triplet nodes.

[0063] Here, the hierarchical weight indexing unit receives weight inputs from the content generation module in real time through a dynamic weight receiving interface and a weight activation calculation module, dynamically adjusting the retrieval priority of triple paths. When the dynamic association weight on a path fluctuates by more than a preset threshold ΔW, such as ΔW > 15%, path reordering is automatically triggered, and an activation signal is transmitted to the node feature mapping module.

[0064] Here, the node feature mapping module encodes the physical sign matching parameters, behavioral feature parameters and historical feedback weights into node identifiers; a greedy algorithm is used to select the node combination with the highest weight sensitivity of the subtree path, and dynamically generate a matching path with the current user profile.

[0065] Here, the weight lower limit guarantee unit can be composed of a hardware comparator array and a freeze instruction generator. The hardware comparator array compares the weights of different nodes and performs a lower limit judgment. If it is lower than the set value, the path is frozen through the logic gate circuit to prevent low-matching nodes from being misselected.

[0066] Here, the data synchronization interface supports periodically pulling the latest vital sign-behavior data from the user portrait database to generate a difference weight update package; a timestamp verification mechanism is used to synchronize data to prevent weight distortion due to pathological data delays.

[0067] The feedback adjustment module collects user response data to rehabilitation suggestions. The daily behavior response analysis unit calculates α and β based on the current day's click delay time Δt, execution completion rate η, and other factors. This generates updated data for the behavioral feature vector, and then adjusts the value of the historical feedback weight field to achieve dynamic iteration of the user portrait database. Specifically, the module analyzes user feedback data to quantify the user's acceptance and implementation of rehabilitation suggestions, thereby optimizing system parameters. This enables the system to continuously optimize rehabilitation suggestion strategies based on actual user feedback, continuously improve the accuracy of rehabilitation suggestions and user satisfaction, and make the system more tailored to the user's personalized needs.

[0068] The feedback adjustment module also includes:

[0069] The daily behavior response analysis unit calculates α and β based on the click delay time Δt, execution completion degree η, and feedback score σ of the day;

[0070] The calculation formula of α is: α = (Ση_i) / n×(1-Δt_avg / 300);

[0071] Where n is the number of valid records on the day, n≥5, and Δt_avg is the average click delay time;

[0072] The calculation formula of β is: β = 0.3σ + 0.7(Σclick_count) / total_push;

[0073] V = [α, β, Δt_max, η_std, σ_trend, last_7d_avg], where Δt_max is the maximum click delay time of the day;

[0074] η_std is the standard deviation of the execution completion rate on the day; σ_trend is the linear slope of the score trend in the past 7 days.

[0075] The specific operation process of this system is as follows:

[0076] (1) Data acquisition and preprocessing: Wearable devices are used to collect real-time vital signs data such as heart rate, blood pressure, and blood oxygen at a sampling frequency greater than 10 Hz. After wavelet denoising, the data are linearly interpolated and resampled to 100 Hz through a multi-source synchronization unit. After sliding window time domain alignment, the processed data are combined with information such as timestamp, sensor ID, original value, threshold mean μ_t, and standard deviation σ_t to form a data packet and output it to the user portrait database. At the same time, the three-axis acceleration sensor and GPS positioning data are input into the motion mode recognition module to identify motion modes such as walking, running, and standing still, generate behavioral feature threshold conditions, and store them in the user portrait database as part of the behavioral feature vector to complete the initial data construction.

[0077] (2) Knowledge matching and weight calculation: The content generation module connects the user portrait database and the rehabilitation knowledge base, and matches the physical sign matching field and behavioral feature threshold condition field of the triple in the rehabilitation knowledge base with the real-time physical sign data and behavioral feature vector in the user portrait database; the dynamic association weight is calculated according to the formula W_new = W_base×(1+α×S_urgency)×e^(-λ×t), where W_base is the normalized value of the historical feedback weight field, α is the urgency coefficient between 0.1 and 0.5, S_urgency is obtained by weighted summing the abnormal confidence of the real-time physical sign, the decline in the completion of the behavioral feature, and the trend change value of the historical feedback weight and selecting the maximum component, λ is the time attenuation coefficient of 0.01-0.1, and t is the time interval of the user portrait database update cycle.

[0078] (3) Abnormal vital sign monitoring and event triggering: The event response module monitors the real-time vital sign data in the user portrait database in real time, calculates the dynamic threshold according to the dynamic threshold adjustment formula θ_t=μ_t+kσ_t, and generates an event trigger signal if an abnormal vital sign is detected that exceeds the dynamic threshold.

[0079] (4) Prioritization and output of rehabilitation content: The portrait matching module adopts an improved Trie tree structure containing a hierarchical weight index unit, receives event trigger signals and the dynamic association weight output by the content generation module, and prioritizes the triple content in the rehabilitation knowledge base; when the dynamic association weight fluctuation exceeds the preset threshold, the hierarchical weight index unit automatically triggers the reordering of the weight path, and finally outputs the highest priority rehabilitation content to provide rehabilitation advice to patients.

[0080] (5) Feedback data collection and system optimization: The feedback adjustment module collects the user's operational response data to rehabilitation suggestions, including the click delay time Δt, execution completion degree η and feedback score σ on the same day; through the daily behavior response analysis unit, α and β are calculated according to the formulas α=(Ση_i) / n×(1-Δt_avg / 300) and β=0.3σ+0.7(Σclick_count) / total_push, and the updated data of the behavioral feature vector containing parameters such as α and β are generated; based on the updated data, the value of the historical feedback weight field is adjusted to realize the dynamic iteration of the user portrait database, so as to prompt the system to optimize the generation strategy of subsequent rehabilitation suggestions according to user feedback.

[0081] (6) System loop operation: Repeat steps 2-5, and continue to match, calculate, respond and optimize rehabilitation suggestions based on the user's real-time data and feedback, to ensure that the system always provides patients with home cardiac rehabilitation guidance that meets their actual needs.

[0082] The various modules of this system work together to ensure that abnormal vital signs can be quickly responded to meet the needs of timely response. The system can support concurrent access by thousands of people, with small data synchronization errors, meeting the hospital's needs for simultaneous monitoring and advice on patients, and significantly improving the effectiveness and efficiency of home cardiac rehabilitation.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of the patent application of the present invention. Any equivalent changes, equivalent replacements or modified changes within the technical spirit and principles suggested by the present invention should be included in the scope of patent protection covered by the present invention.

Claims

1. A user-profile-based home cardiac rehabilitation recommendation system, characterized by: include: User portrait database, used to store patients' real-time vital sign data, behavioral feature vectors, and historical feedback weight data; The rehabilitation knowledge base uses a graph structure to classify and store rehabilitation content. The rehabilitation content is marked by a triplet field. The triplet includes: a physical sign matching field, a behavioral feature threshold condition field, and a historical feedback weight field. a content generation module, connected to the user portrait database and the rehabilitation knowledge base, configured to match the physical sign matching field and the behavioral feature threshold condition field of the triple with the user's real-time physical sign data and behavioral feature vector, and dynamically calculate the dynamic association weight based on the historical feedback weight field and the real-time data; an event response module configured to monitor real-time vital sign data and generate an event trigger signal if abnormal vital signs exceeding a dynamic threshold are detected; The portrait matching module uses an improved Trie tree structure and is configured to prioritize the triples in the rehabilitation knowledge base based on event trigger signals and dynamic association weights, and output the highest priority rehabilitation content; The feedback adjustment module is configured to collect the user's operational response data to the rehabilitation suggestions, generate updated data of the behavioral feature vector, and adjust the value of the historical feedback weight field based on the updated data to achieve dynamic iteration of the user portrait database.

2. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The calculation formula of the dynamic association weight is: W_new=W_base×(1+α×S_urgency)×e^(-λ×t); Where: W_base is the normalized value of the historical feedback weight field, α is the urgency coefficient, ranging from 0.1 to 0.5, S_urgency is a comprehensive score based on the deviation of real-time physical sign abnormalities and behavioral characteristics, λ is the time attenuation coefficient, 0.01≤λ≤0.1, and t is the time interval of the user portrait database update cycle.

3. The user portrait-based home cardiac rehabilitation recommendation system according to claim 2, characterized in that: The S_urgency calculation method includes: performing weighted summation on the abnormal confidence of real-time physical signs, the decrease in the degree of completion of behavioral characteristics, and the trend change value of historical feedback weights, and selecting the maximum component as the final score.

4. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The real-time vital sign data is collected through wearable devices, including heart rate, blood pressure, and blood oxygen, with a sampling frequency greater than 10Hz. After wavelet denoising processing, it is linearly interpolated and resampled to 100Hz through a multi-source synchronization unit, and then output to the user portrait database after sliding window time domain alignment.

5. The user portrait-based home cardiac rehabilitation recommendation system according to claim 4, characterized in that: The output data packet of the multi-source synchronization unit includes a timestamp, a sensor ID, an original value, a threshold mean μ_t, and a standard deviation σ_t. The dynamic threshold adjustment formula is: θ_t=μ_t+kσ_t; Wherein, k is a preset parameter, k=2~39.

6. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The behavior feature vector generation method includes: The three-axis acceleration sensor and GPS positioning data are input into the motion pattern recognition module to identify motion patterns including walking, running and standing still, and generate behavioral feature threshold conditions; The motion feature threshold conditions are bound and matched with the "sports content" in the rehabilitation knowledge base, where the threshold conditions include running time, walking steps and stationary time.

7. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The feedback adjustment module further includes: The daily behavior response analysis unit calculates α and β based on the click delay time Δt, execution completion degree η, and feedback score σ of the day; The calculation formula of α is: α = (Ση_i) / n×(1-Δt_avg / 300); Where n is the number of valid records on the day, n≥5, and Δt_avg is the average click delay time; The calculation formula of β is: β = 0.3σ + 0.7(Σclick_count) / total_push; V = [α, β, Δt_max, η_std, σ_trend, last_7d_avg], where: Δt_max is the maximum click delay time of the day; η_std is the standard deviation of the execution completion rate of the day; σ_trend is the linear slope of the rating trend in the past 7 days.

8. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The update trigger conditions of the user portrait database include: Vital sign data exceeds the dynamic threshold for three consecutive times; The cosine similarity between the behavior feature vector and the historical behavior pattern is less than 0.7; The environmental sensor detects that the temperature exceeds 32°C or the PM2.5 concentration exceeds 100μg / m 3 .

9. The user portrait-based home cardiac rehabilitation recommendation system according to claim 1, characterized in that: The improved Trie tree structure includes a hierarchical weight index unit, and the weight value of the hierarchical weight index unit is driven in real time by the dynamic association weight output by the content generation module.

10. The user portrait-based home cardiac rehabilitation recommendation system according to claim 9, characterized in that: The hierarchical weight indexing unit is configured to automatically trigger the reordering of the weight paths and update the matching priority of the triples with the user portrait database when the dynamic association weight fluctuation exceeds a preset threshold.

Images