A rehabilitation task generation method of personalized information

Abstract

The application relates to the technical field of wisdom-based old-age care, and discloses a rehabilitation task generation method for personalized information, which comprises the following steps: collecting rehabilitation data of old people in a nursing home and interactive data of visitors to form basic data; constructing a comprehensive portrait of the old people and the visitors based on the rehabilitation data and the interactive data; extracting comprehensive features of the comprehensive portrait, and performing initial task matching in a rehabilitation task library; performing multi-target weighted calculation on the initial task based on the old people and the visitors, obtaining a matching degree corresponding to the initial task, and selecting three tasks with the highest matching degrees as selected tasks; performing satisfaction score on the selected tasks by the old people and the visitors, and selecting a final rehabilitation task with the highest comprehensive satisfaction degree; and generating an optimal rehabilitation task through multi-target weighted calculation and a two-way scoring mechanism, so that dynamic and accurate matching and safety strength control are realized, and the rehabilitation task matching accuracy and safety are improved.

Description

Technical Field

[0001] This invention relates to the field of smart elderly care technology, specifically to a method for generating personalized rehabilitation tasks. Background Technology

[0002] As the core carrier of centralized elderly care services, nursing homes receive significant attention for their service quality and rehabilitation outcomes. Currently, nursing home rehabilitation services suffer from systemic deficiencies at three levels: Firstly, at the data collection level, existing solutions rely solely on manual collection of basic health information by medical staff, lacking systematic collection of multi-dimensional dynamic information such as real-time status data and emotional feedback data. Furthermore, a standardized collection mechanism for interactive data, including visitor abilities, strengths, and scheduling, has not been established. Secondly, at the task matching level, traditional methods use static assessment templates, failing to dynamically adjust task intensity based on real-time changes in the elderly's vital signs or accurately allocate tasks based on the visitor's professional skills, leading to a disconnect between rehabilitation tasks and participants' actual needs. Thirdly, at the optimization mechanism level, existing systems lack closed-loop optimization capabilities based on historical feedback. They cannot dynamically revise the profile model based on task performance or intelligently update the rehabilitation task database. This extensive management model results in significant fluctuations in rehabilitation outcomes, low visitor participation, and safety risks due to a mismatch between rehabilitation intensity and the elderly's tolerance. Especially when sudden changes in vital signs or temporary staff adjustments occur, existing systems struggle to generate timely alternative solutions, severely impacting the continuity and safety of rehabilitation services. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method for generating personalized rehabilitation tasks.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for generating personalized rehabilitation tasks includes: The basic data consists of rehabilitation data of elderly people in nursing homes and interaction data of visitors. The rehabilitation data includes basic health information, rehabilitation needs information, real-time status information and historical feedback information. The interaction data includes identity attribute information, ability and skill information, time adaptation information and interaction preference information. A comprehensive profile of the elderly and visitors is constructed based on the rehabilitation data and interaction data. Extract the comprehensive features of the comprehensive profile and match them with the initial task in the rehabilitation task library; The initial task was calculated using a multi-objective weighted calculation based on the elderly and visitors to obtain the matching degree of the initial task. Three tasks were selected as selection tasks based on the matching degree from high to low. The elderly and visitors rate the selected tasks, and the task with the highest overall satisfaction is selected as the final rehabilitation task.

[0005] In this invention, preferably, the rehabilitation task is also adjusted in real time: During rehabilitation activities, real-time data of the elderly is collected. Determine whether there is any data in the real-time data that exceeds a threshold. If it is determined that there is data that exceeds the threshold, match a suitable task type based on the data type. Determine whether the currently executed rehabilitation task matches the task type; if it does, continue the rehabilitation task. If it does not meet the requirements, check if there is a matching task type among the remaining selected tasks. If there is a matching task type, change it. If no matching task type is found, extract the current rehabilitation task and select a rehabilitation task of the same nature that matches the current task type to replace it.

[0006] In this invention, preferably, the collection of rehabilitation data from elderly people in nursing homes is carried out using a smart device non-intrusive collection strategy, a multi-scenario interactive collection strategy, and an auxiliary verification collection strategy. The smart device non-intrusive collection strategy drives the smart device to collect real-time heart rate, blood oxygen saturation, body temperature, limb activity data, and location data of the elderly person without being noticed throughout the process. The multi-scenario interactive collection strategy interactively collects data on the elderly person's physical condition, emotional state, activity preferences, and subjective preferences during the period when the elderly person begins rehabilitation tasks. The auxiliary verification collection strategy interactively collects data on the elderly person's mental state, social interaction, abnormal behavior, and daily life routines after the rehabilitation tasks are completed and in daily life.

[0007] The intelligent device's non-intrusive data collection strategy is specifically implemented by wearable devices, a global environmental sensor, and rehabilitation equipment. The wearable devices collect real-time heart rate, blood oxygen saturation, body temperature, limb activity data, and location data of the elderly. The global environmental sensor collects the elderly's respiratory rate and gait characteristics when in a fixed area. The rehabilitation equipment collects training data when the elderly use rehabilitation equipment.

[0008] The multi-scenario interactive data collection strategy includes voice interaction data collection in public scenarios, real-time interactive data collection in rehabilitation tasks, and scenario-based questionnaire data collection using wearable devices.

[0009] The auxiliary verification collection strategies include daily patrols by caregivers, professional assessments by medical staff, and interactive feedback from visitors.

[0010] In this invention, preferably, the multi-scene interactive acquisition strategy and the smart device non-intrusive acquisition strategy also perform data linkage acquisition. Based on the location data acquired by the smart device non-intrusive acquisition strategy, when the retention time of the location data exceeds a preset value, the multi-scene interactive acquisition strategy automatically locates the acquisition device in the area corresponding to the location data, and associates the data acquired by the acquisition device within the time period corresponding to the retention time with the location data. When the retention time of the location data is less than the preset value, the multi-scene interactive acquisition strategy only records the motion trajectory.

[0011] In this invention, preferably, when the corresponding area includes location data from multiple different elderly terminals, and the retention time of the location data exceeds a preset value, the multi-scenario interactive acquisition strategy activates the backup acquisition device in the corresponding area, so that each acquisition device corresponds to one elderly terminal, and the data collected by the acquisition device that is closer to the elderly terminal is associated with the location data.

[0012] In this invention, preferably, the rehabilitation task library contains multiple rehabilitation task templates, each of which is labeled with a task attribute vector, the task attribute vector including rehabilitation goal, intensity level, duration requirement, skill matching tag and task type tag.

[0013] In this invention, preferably, constructing a comprehensive profile of the elderly and visitors based on the rehabilitation data and interaction data specifically includes: The rehabilitation data and interaction data are tagged to obtain tagged data; Construct a three-dimensional feature matrix, map the labeled data to the corresponding dimensions, and dynamically assign initial weights based on the data's influence; By using interactive influence factors and feature correlation algorithms, a two-way fusion of profiles of the elderly and visitors is achieved, generating a comprehensive and correlated profile. The associated profile is dynamically corrected based on a time-series decay function and a feedback loop strategy. The feedback loop strategy generates a profile accuracy report after each period and monthly, and iteratively adjusts the feature dimensions and weight allocation rules by comparing the match rate between predicted preferences and actual behavior.

[0014] In this invention, preferably, the dynamic allocation of initial weights includes: First, the information entropy of each sub-dimension is calculated using the entropy weighting method. The lower the entropy value, the higher the information value and the greater the weight. At the same time, the weights are adjusted based on the scenario coefficient. In the case of rehabilitation training, the weight of the elderly's functional training data is increased, and in the case of social interaction, the weight of the elderly's emotional feedback data is increased.

[0015] In this invention, preferably, the extraction of the comprehensive features of the comprehensive portrait includes: The core main feature dimensions are determined, and key sub-features in the comprehensive profile are associated with each dimension. The core main feature dimensions include rehabilitation needs features, ability matching features, interaction preference features, and constraint condition features. Quantify the associated key sub-features; Normalize and dynamically assign weights to the quantized sub-features; Calculate the mutual information value between each character feature and the task matching satisfaction. The sub-information corresponding to the mutual information value at a set threshold is retained as the final comprehensive feature.

[0016] In this invention, preferably, the matching of initial tasks in the rehabilitation task library includes: First, a preliminary screening is conducted based on the elderly's needs information, and a rehabilitation task template that matches the rehabilitation goals in the elderly's needs information is selected. Next, based on the rehabilitation contraindications according to the elderly's needs, rehabilitation task templates related to rehabilitation contraindications were deleted; Then, based on the tolerance level in the elderly's needs information, rehabilitation task templates with task intensity greater than tolerance level are deleted. Finally, a rehabilitation task template that meets the time range is selected as the initial task based on the time adaptation information.

[0017] In this invention, preferably, the overall matching degree of the initial task is calculated based on a weighted Euclidean distance matching algorithm.

[0018] In this invention, preferably, it also includes: After the task is completed, the elderly's post-task satisfaction, physical feelings and improvement suggestions are collected through the elderly's terminal, and the task suitability rating, operation difficulty feedback and optimization opinions are collected through the visitor's terminal. Based on the collected feedback data, the initial weights are adjusted as follows: If the percentage of feedback indicating post-task satisfaction scores below 3 points is ≥30% for 7 consecutive days, the weight of A is increased by 5%-10%; if the percentage of feedback indicating task suitability scores below 3 points is ≥30% for 7 consecutive days, the weight of B is increased by 5%-10%; if the percentage of negative feedback related to duration is ≥30% for 7 consecutive days, the weight of C is increased by 5%-10%; if the task satisfaction improvement rate after historical feedback correction is ≥20%, the weight of D is increased by 3%-8%. Based on the collected feedback data, the rehabilitation task library is updated. If the elderly satisfaction score of a rehabilitation task template is ≤3 points for 30 consecutive days and the usage rate is ≤5%, it is removed from the rehabilitation task library. If the feedback rate of a certain type of uncovered rehabilitation needs is ≥20% for 15 consecutive days, a new rehabilitation task template is designed in collaboration with medical staff, and the task attribute vector is labeled and added to the rehabilitation task library.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a method and storage medium for generating personalized rehabilitation tasks. It collects rehabilitation data of elderly people in nursing homes and interaction data of visitors to form basic data. Based on the rehabilitation data and interaction data, it constructs a comprehensive profile of the elderly and visitors, extracts comprehensive features for task matching, and generates the optimal rehabilitation task through multi-objective weighted calculation and two-way scoring mechanism. This achieves dynamic and accurate matching and safety intensity control, and has the advantages of improving the accuracy and safety of rehabilitation task matching. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a personalized information rehabilitation task generation method according to the present invention.

[0021] Figure 2 This is a schematic diagram illustrating the real-time adjustment of rehabilitation tasks in a personalized rehabilitation task generation method according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Please see Figure 1 and Figure 2 A preferred embodiment of the present invention provides a method for generating personalized rehabilitation tasks. This application proposes a technical solution that collects rehabilitation data of elderly people in nursing homes and interaction data of visitors to form basic data, constructs a comprehensive profile based on the two types of data, extracts comprehensive features to match initial tasks in the rehabilitation task library, obtains matching degree ranking through multi-objective weighted calculation, selects the top three tasks for two-way satisfaction scoring, and finally determines the optimal rehabilitation task.

[0025] The rehabilitation data includes basic health information, rehabilitation needs information, real-time status information, and historical feedback information. This can be achieved using electronic health record systems, wearable device monitoring modules, and nursing record databases to accurately reflect the baseline level and dynamic trends of the elderly's physical function. Interaction data includes identity attribute information, ability and skill information, time suitability information, and interaction preference information. This can be collected through visitor registration forms, skill certification systems, and schedule management systems to assess the feasibility of visitors performing rehabilitation tasks. Multi-objective weighted calculation sets weight coefficients for three dimensions: task intensity suitability, skill matching, and time overlap. Specifically, the analytic hierarchy process (AHP) can be used to determine the weight ratio of each dimension, and the comprehensive matching value can be calculated using the weighted Euclidean distance formula.

[0026] Compared with existing technologies, this solution breaks through the traditional single-dimensional matching model and innovatively establishes a two-way dynamic evaluation system. Existing technologies only consider the basic health data of the elderly, while this solution simultaneously integrates the elderly's real-time physiological indicators, historical data, and the professional qualifications of the visitors, so that the task intensity dynamically adapts to the elderly's current state; traditional methods rely on human experience to match visitors, while this solution uses a quantitative evaluation model to accurately calculate the skill matching degree, reducing operational risks; conventional systems use fixed time schedules, while this solution introduces a flexible time window matching algorithm to improve the feasibility of task execution.

[0027] Through the above technical solution, this application effectively addresses the disconnect between the intensity of rehabilitation tasks and the real-time condition of the elderly, avoids the risk of operational errors by visitors during professional rehabilitation training, and significantly improves mutual satisfaction in task execution. The rehabilitation tasks automatically generated by the system, while ensuring safety, can enhance the elderly's enthusiasm for training, optimize the utilization rate of visitor resources, and form a virtuous cycle of rehabilitation service models.

[0028] Furthermore, the collection of rehabilitation data from elderly residents in nursing homes is achieved through a three-pronged approach: a seamless data collection strategy using intelligent devices, a multi-scenario interactive data collection strategy, and an auxiliary verification data collection strategy. The seamless data collection strategy using intelligent devices is specifically accomplished by wearable devices, a global environmental sensor, and rehabilitation equipment. The wearable devices are lightweight smart bracelets or smart badges provided to the elderly, integrating heart rate, blood oxygen, body temperature, motion sensors, and a positioning module. These primarily collect real-time data on the elderly's heart rate, blood oxygen saturation, body temperature, limb activity levels, and location. The global environmental sensor consists of radar sensors installed at different locations in public areas such as the dining room, activity room, and corridors, used to collect data on the elderly's respiratory rate and gait characteristics in fixed areas. The rehabilitation equipment collects training data when the elderly use rehabilitation devices, mainly by connecting the devices to the Internet of Things (IoT) to collect data such as grip strength, joint flexion angle, balance maintenance duration, and number of training sessions completed. The multi-scenario interactive data collection strategy involves interactively collecting data on the elderly's physical condition, emotional state, activity preferences, and subjective preferences during the period when they begin rehabilitation tasks. This strategy includes voice interaction data collection in public settings, real-time interactive data collection during rehabilitation tasks, and scenario-based questionnaire collection using wearable devices. Voice interaction data collection in public settings involves using voice interaction terminals installed in public areas to conduct timed, lightweight inquiries and receive feedback from the elderly. Real-time interactive data collection during rehabilitation tasks uses the voice module of the rehabilitation device or a caregiver's handheld terminal to inquire about the elderly's feelings in real time, providing three voice options: comfortable, normal, or uncomfortable. The elderly can quickly provide feedback through voice or gestures. Scenario-based questionnaire collection using wearable devices involves conducting questionnaires every morning upon waking and after the rehabilitation tasks, receiving feedback from the elderly through a combination of voice broadcasts and text / images. The auxiliary verification data collection strategy involves interactively collecting data on the elderly's mental state, social interactions, abnormal behaviors, and daily routines after the rehabilitation tasks have concluded and during routine daily activities. This strategy includes data collection during daily rounds by caregivers, professional assessments by medical staff, and interactive feedback from visitors. Daily rounds by caregivers utilize smart monitoring terminals equipped with caregivers to collect data on the elderly's mental state, eating preferences, self-care abilities, social interactions, and abnormal behaviors. Professional assessments by medical staff involve measuring blood pressure, blood sugar, muscle strength, and cognitive function at fixed frequencies using specialized equipment. This data, combined with rehabilitation training data collected by the equipment, generates a rehabilitation progress assessment report. The collected data undergoes data cleaning and fusion to form the elderly's rehabilitation data.

[0029] Furthermore, the multi-scenario interactive acquisition strategy and the smart device's non-intrusive acquisition strategy also perform data linkage acquisition. Based on the location data acquired by the smart device's non-intrusive acquisition strategy, when the retention time of the location data exceeds a preset value, the multi-scenario interactive acquisition strategy automatically locates the acquisition device within the area corresponding to the location data and associates the data acquired by the acquisition device within the time period corresponding to the retention time with the location data. When the retention time of the location data is less than the preset value, the multi-scenario interactive acquisition strategy only records the motion trajectory.

[0030] Specifically, the multi-scenario interactive data acquisition strategy establishes a data communication link with the smart device non-intrusive data acquisition strategy. Based on the location data of the elderly obtained by the positioning module through the smart device non-intrusive data acquisition strategy, when the location data is held in a certain area for a longer time than a preset value, the multi-scenario interactive data acquisition strategy automatically retrieves the interactive data acquisition devices deployed in that area and associates and stores the subjective feedback data of the elderly collected by the devices within the time period corresponding to the holding time with the location data. When the holding time of the location data is less than the preset value, the multi-scenario interactive data acquisition strategy only records the elderly's movement trajectory and does not trigger the data association operation.

[0031] Among them, data linkage acquisition refers to real-time data collected by smart devices through non-intrusive acquisition strategies and multi-scenario interactive acquisition strategies. Location data retention time refers to the length of time an elderly person stays in the same area. The preset value can be adjusted according to the characteristics of the scenario, for example, 10 minutes in the activity room and 3 minutes in the corridor, to distinguish between brief passage and effective activity scenarios. In-area acquisition device positioning refers to quickly locating the interactive terminal in the area where the elderly person stays by matching the device deployment map with the location coordinates. Specifically, a spatial indexing algorithm is used to ensure that the device matching accuracy is ≤1 meter.

[0032] Specifically, when the corresponding area includes location data from multiple different elderly users simultaneously, and the retention time of the location data exceeds a preset value, the multi-scenario interactive collection strategy activates the backup collection device for the corresponding area, so that each collection device corresponds to one elderly user, and the data collected by the collection device closer to the elderly user's location is associated with that location data. In a specific embodiment, the smart bracelet uploads the elderly's location data in real time via Bluetooth beacons. When the system detects that the elderly's stay in the activity room reaches 8 minutes (the preset value is 5 minutes), it automatically triggers data linkage: it locates the No. 2 voice interaction terminal in the activity room area, and associates the subjective feedback collected by the terminal within 8 minutes, such as "likes handicraft activities" and "wants to participate in the afternoon choir," with the location data of "staying in the activity room for 8 minutes," and marks it as "expressed preference for handicrafts and choir during the activity in the activity room." When the elderly's stay in the corridor is only 2 minutes (less than the preset value of 3 minutes), the system determines it as a brief passerby, and only records the "movement trajectory from the room to the restaurant from 14:00 to 14:02," without associating it with the interactive device data in the corridor area.

[0033] Furthermore, when the corresponding area includes location data from multiple different elderly terminals, and the location data retention time exceeds a preset value, the multi-scenario interactive collection strategy activates the backup collection device in the corresponding area, so that each collection device corresponds to one elderly terminal, and the data collected by the collection device that is closer to the elderly terminal is associated with the location data.

[0034] Among these, "multiple elderly location conflict" refers to scenarios where two or more elderly people are present in the same area, with the preset value remaining consistent with the single elderly person scenario. "Backup data collection equipment" refers to redundant equipment not normally deployed within the area, ensuring its availability at all times through a device status monitoring module. "Distance-priority association" calculates the relative distance between the elderly person and the equipment based on the RSSI algorithm, prioritizing data from the device with the strongest signal to avoid cross-interference.

[0035] Specifically, if three elderly people stay in the restaurant area for more than 6 minutes (the preset value is 5 minutes), the two normally deployed voice terminals will be occupied. The system will automatically activate one backup mobile terminal and push an assignment instruction: through Bluetooth positioning calculation, elderly person A is closest to terminal 1 and is associated with the feedback "the food is a little salty" collected by terminal 1; elderly person B is closest to terminal 2 and is associated with the feedback "I want to drink hot water" collected by terminal 2; elderly person C is closest to the backup terminal and is associated with the feedback "today's soup is delicious" collected by the backup terminal, thus achieving accurate differentiation and association of the data of the three people.

[0036] Compared to existing technologies, traditional methods often store location and interaction data independently, making it difficult to trace elderly people's feedback in specific scenarios. For example, if someone mentions knee pain in an activity room, it's impossible to link that to the specific time they stayed. Furthermore, data confusion can easily occur when multiple elderly people are in the same area. This solution, through spatiotemporal correlation and device scheduling, not only achieves a complete record of when, where, and what was said, but also maintains data accuracy in scenarios with three or more people in the same area, reducing the correlation error rate to below 0.5%.

[0037] Through the above technical solutions, this application effectively solves the problems of correlation and accuracy in data collection across multiple scenarios. On the one hand, by triggering data linkage through location dwell time, redundant associations of invalid passing scenarios are avoided, improving data effectiveness. On the other hand, by using backup device scheduling and distance-priority association, collection conflicts when multiple elderly people are in the same area are resolved, ensuring that each feedback is accurately linked to the corresponding elderly person. This linkage mechanism makes the scenario-based features in the subsequent comprehensive profile construction more accurate, providing a more granular basis for matching rehabilitation tasks.

[0038] Furthermore, the real-time adjustment method for rehabilitation tasks includes the following steps: collecting real-time data of the elderly during the execution of the rehabilitation task; determining whether there is data exceeding a preset threshold in the real-time data, and matching a suitable task type according to the data type if such data exists; determining whether the currently executed rehabilitation task matches the matched task type, and continuing execution if it does not match, and checking whether there is a matching task type among the remaining candidate tasks and replacing it; if there is no matching task type, selecting a task with the same nature and matching the current task type from the current rehabilitation task for replacement.

[0039] Real-time data refers to the current physiological indicators, behavioral states, and environmental parameters of the elderly, acquired through non-intrusive collection strategies using smart devices and multi-scenario interactive collection strategies. This data dynamically reflects the body's responses during the rehabilitation process. Thresholds refer to pre-set safety ranges or adaptive indicators, specifically determined through clinical medical standards or historical data statistical analysis, used to identify abnormal states or changes in needs. Task type matching refers to mapping data based on the association rules between abnormal data characteristics and rehabilitation task attributes. This can be achieved using knowledge graphs or classification models to quickly locate suitable task adjustment directions. Similarity in nature refers to tasks having similarity in rehabilitation goals, implementation methods, or interaction modes. This can be achieved through similarity calculations between task attribute vectors, ensuring the continuity of task switching.

[0040] Specifically, during the rehabilitation task execution phase, wearable devices continuously collect data on the elderly person's heart rate, blood oxygen saturation, and limb movement range. When the heart rate is detected to consistently exceed 140 beats per minute, the system determines that the data exceeds the cardiopulmonary function safety threshold and automatically triggers the task adjustment process. At this time, based on the data type of cardiopulmonary overload, a low-intensity aerobic training task type is matched. If the currently executed balance training task is of the high-intensity anaerobic type, the candidate task list is immediately checked, and the second-ranked breathing regulation training task is selected as the replacement. If there are no low-intensity aerobic tasks among the candidate tasks, a sub-task containing seated static balance training is selected from the current balance training task. This sub-task maintains the same balance ability training goal as the original task, but the intensity level meets the requirements of the current cardiopulmonary condition.

[0041] Compared to existing technologies, traditional methods only conduct a one-time assessment before the task begins, failing to address sudden changes in physical condition during execution. For example, if an elderly person experiences abnormal blood pressure during rehabilitation training, the caregiver needs to manually observe, interrupt the training, and then reassess and select a new task, resulting in a delayed response. This solution, however, utilizes real-time data monitoring and an automated decision-making mechanism to adjust the task within 20 seconds of detecting abnormal indicators, avoiding secondary damage caused by delayed processing.

[0042] Through the above technical solution, this application effectively solves the problem of adaptation failure caused by unforeseen circumstances during the execution of rehabilitation tasks. By establishing a closed-loop mechanism of real-time monitoring, threshold judgment, and dynamic adjustment, the physiological safety of the elderly during rehabilitation training is ensured, while maintaining the continuity of the rehabilitation process. Especially in the event of an emergency, the system can automatically switch to a task version that suits the current physical condition, avoiding delays in rehabilitation progress caused by the complete termination of training. At the same time, by prioritizing high-scoring tasks in the candidate task list and task variations of similar nature, the implementation effect after task adjustment is ensured to be consistent with the original plan.

[0043] This application further proposes a method for generating personalized rehabilitation tasks. The rehabilitation task library contains multiple rehabilitation task templates, and each rehabilitation task template is labeled with a task attribute vector. The task attribute vector includes rehabilitation goals, intensity levels, duration requirements, skill matching tags, and task type tags.

[0044] The rehabilitation goal refers to the expected physiological or psychological improvement indicators achieved through the rehabilitation task. Specifically, it can be achieved using core indicators from the International Classification of Functional Disabilities (ICD-1), such as improved joint mobility or enhanced emotional stability. The intensity level characterizes the physical or mental exertion of the task on the performer, and can be categorized using metabolic equivalent classification, for example, dividing intensity into low, medium, and high levels. The duration requirement is defined by the time interval needed to perform the task, such as 15-30 minutes or 45-60 minutes. The skill matching tag identifies the specific abilities required to perform the task, which may include nursing skills, psychological counseling abilities, or exercise instruction qualifications. The task type tag distinguishes the functional attributes of the task, such as categorizing it as physical training, cognitive training, or social interaction.

[0045] Specifically, the rehabilitation task library is pre-built as a structured database, with each task template digitally described using multi-dimensional attribute vectors. During task matching, the system analyzes the rehabilitation needs features in the elderly person's profile and the ability features in the visitor's profile, comparing them with the task attribute vectors in multiple dimensions. For example, when an elderly person needs to improve upper limb function and the visitor has the qualifications to provide exercise guidance, the system prioritizes matching task templates whose rehabilitation goals include upper limb training and whose skill matching tags include exercise guidance. The multi-dimensional annotation of the task attribute vectors allows the matching process to be filtered based on combinations of different dimensions, such as task templates that simultaneously meet the requirements of intensity level ≤ 2, duration requirement ≤ 30 minutes, and task type as cognitive training. This effectively solves the adaptation bias problem caused by a single attribute dimension in the rehabilitation task matching process. The structured definition of multi-dimensional attribute vectors ensures that task selection takes into account multiple factors such as rehabilitation needs, execution ability, and operational constraints, reducing the risk of task execution due to excessive intensity, skill mismatch, or time conflicts.

[0046] This application further proposes a specific implementation method for constructing comprehensive profiles of the elderly and visitors based on rehabilitation data and interaction data, including: labeling rehabilitation data and interaction data to obtain labeled data; constructing a three-dimensional feature matrix and mapping the labeled data to the corresponding dimensions, and dynamically allocating initial weights based on data influence; achieving bidirectional fusion of elderly and visitor profiles to generate a comprehensive associated profile through interaction influence factors and feature correlation algorithms; dynamically correcting the associated profile based on time-series decay functions and feedback loop strategies, generating profile accuracy reports after each period and monthly, and iteratively adjusting feature dimensions and weight allocation rules by comparing the consistency rate between predicted preferences and actual behavior.

[0047] Tagging data refers to transforming basic health information in rehabilitation data into structured tags, such as labeling blood pressure values ​​as low-risk, medium-risk, or high-risk levels. This can be achieved using natural language processing technology combined with medical knowledge graphs for entity recognition and classification. A three-dimensional feature matrix is ​​a data container composed of physiological, behavioral, and psychological dimensions, implemented using a multi-dimensional array data structure, with sub-feature storage space under each dimension to hold the tagged data. Dynamically assigning initial weights involves assigning differentiated values ​​based on the data's influence on rehabilitation task matching. For example, entropy weighting is used to calculate the information entropy value of each sub-dimension, and the weight allocation ratio is adjusted based on scenario coefficients. The interaction influence factor is a parameter reflecting the degree of correlation between the elderly and visitors in two-way interaction, specifically achieved by calculating the similarity of behavioral patterns between the two parties using collaborative filtering algorithms. Feature correlation algorithms quantify the interaction relationships between different features; for example, Pearson correlation coefficients are used to analyze the correlation strength between emotional feedback data and task satisfaction. The time-series decay function is a calculation model that reduces the influence of historical data over time; for example, an exponential decay function is used to assign lower weights to interaction records from three months ago. The feedback loop strategy refers to a mechanism that periodically evaluates the accuracy of the profile and corrects the model parameters accordingly. Specifically, an automated report generation system can be established to compare the deviation rate between predicted preferences and actual behavior.

[0048] Specifically, in constructing a comprehensive profile, the collected rehabilitation and interaction data are first structured using a predefined tagging system. For example, heart rate data in real-time status information is labeled as "stable" or "abnormal." Next, a feature matrix is ​​constructed encompassing three dimensions: physiological, behavioral, and psychological. Tagged data is mapped to corresponding dimensions according to attributes. For instance, visitor skill certificate information is categorized as an ability / skill sub-feature and stored in the behavioral dimension. During the initial weight allocation process, the information entropy of each sub-feature is calculated using the entropy weighting method. For example, if the information entropy value of emotional feedback data is found to be lower than that of basic health data, it is assigned a higher weight. In social interaction scenarios, the system automatically increases the scenario coefficient of emotional feedback data, giving this feature a larger weight in the weight calculation. In the profile fusion stage, a matrix factorization algorithm is used to mine potential correlations between the elderly and visitor features. For example, a high degree of matching is identified between the visitor's music therapy skills and the elderly's emotional regulation needs. During dynamic correction, the monthly profile accuracy report calculates the deviation rate between predicted preferences and actual behavior. If the actual execution time of a limb rehabilitation task is 15% lower than the predicted value, the weight allocation rules are adjusted, reducing the initial weight of the time-requirement feature.

[0049] Compared to existing technologies, traditional nursing home profiling relies solely on static health records for one-way assessment, lacking a multi-dimensional dynamic feature matrix and failing to reflect real-time status and interaction relationships. This solution, however, achieves multi-source data fusion through a three-dimensional feature matrix, eliminates historical data interference using a time-series decay function, and continuously optimizes profiling accuracy using a feedback loop strategy, resulting in a gradual improvement in rehabilitation task matching accuracy with data accumulation. This addresses the problems of single-dimensional profiling, poor data timeliness, and lack of bidirectional correlation analysis in existing technologies. It achieves dynamic weight allocation and periodic model correction, enabling the comprehensive profiling to adapt to changes in the elderly's state and the evolution of interaction patterns, providing continuously improving feature basis for rehabilitation task matching accuracy.

[0050] This application further proposes a method for dynamically allocating initial weights, which includes firstly calculating the information entropy of each sub-dimension using the entropy weight method, where the lower the entropy value, the higher the information value and the greater the weight; and secondly adjusting the weights based on the scenario coefficient, increasing the weight of the elderly's functional training data in the rehabilitation training scenario and increasing the weight of the elderly's emotional feedback data in the social interaction scenario.

[0051] The entropy weighting method refers to determining weights by calculating the information entropy of data. Specifically, it uses the information entropy formula to calculate the dispersion of each sub-dimension. The lower the dispersion, the smaller the entropy value of the dimension, and the higher the information value of the data in that dimension, thus assigning it a higher weight. This feature is used to objectively quantify the contribution of different data dimensions to profile construction, avoiding subjective weighting bias. The scenario coefficient adjustment weight refers to a method of dynamically adjusting the weight allocation rules according to the rehabilitation task execution scenario. Specifically, it can identify the current task type through a pre-set scenario classification model. For example, when the scenario type is limb rehabilitation training, the weight ratio of muscle strength data and joint mobility data is automatically increased; when the scenario type is psychological counseling interaction, the weight ratio of emotion recognition data and social interaction feedback data is automatically increased. This feature makes the weight allocation scenario-adaptive, solving the problem that a single weight model cannot adapt to multiple task scenarios.

[0052] Specifically, in constructing a comprehensive profile, the multidimensional data is first divided into sub-dimensions. For example, basic health information is divided into sub-dimensions such as blood pressure, heart rate, and muscle strength. The information entropy of each sub-dimension is calculated using the entropy weighting method. For instance, muscle strength data has a standard deviation of 3.2 and a concentrated distribution, resulting in an entropy value of 0.15, while blood pressure data has a standard deviation of 12.8 and a discrete distribution, resulting in an entropy value of 0.82. Therefore, muscle strength data is assigned a higher weight. Simultaneously, the weights are dynamically adjusted based on the real-time scenario. For example, in a cognitive training scenario, the weight coefficient for attention duration data is increased from 0.2 to 0.35, while in a music therapy scenario, the weight coefficient for emotional fluctuation frequency data is increased from 0.1 to 0.25. The resulting dynamic weight matrix is ​​applied to feature fusion calculations, ensuring that core data dimensions receive differentiated attention in different scenarios.

[0053] Traditional weighting methods typically use fixed proportions or manual experience to set weights, such as uniformly setting the weight of physiological data at 60% and psychological data at 40%. This solution, however, quantifies the intrinsic value of data using entropy weighting and dynamically adjusts it based on scenario recognition. For example, in social scenarios, the weight of psychological data is increased to 55%, and in post-operative rehabilitation scenarios, the weight of physiological data is increased to 70%, achieving a dual adaptation between data value assessment and scenario requirements.

[0054] This application further proposes extracting comprehensive features for the comprehensive profile, including: determining the core main feature dimensions, associating key sub-features in the comprehensive profile with each dimension, the core main feature dimensions including rehabilitation needs features, ability matching features, interaction preference features, and constraint condition features; quantifying the associated key sub-features; normalizing and dynamically assigning weights to the quantified sub-features; calculating the mutual information value between each sub-feature and the task matching satisfaction; and retaining the corresponding sub-information at a set threshold as the final comprehensive feature.

[0055] The core principal feature dimension refers to the set of features selected from multidimensional data that have a decisive impact on the matching of rehabilitation tasks. Specifically, the analytic hierarchy process (AHP) combined with expert experience can be used to determine the dimensional composition, focusing on key influencing factors and avoiding data redundancy. Quantification refers to transforming unstructured features into a calculable numerical form. Specifically, a Likert scale can be used to quantify preference features at five levels, or physiological indicators can be standardized using sensor data. Normalization refers to eliminating the influence of different units on feature weights. Specifically, the minimax method can be used to map each sub-feature to the 0-1 interval. Dynamic weight allocation refers to adjusting the importance of features according to scenario requirements. Specifically, the entropy weight method combined with real-time scenario coefficients can be used to generate a dynamic weight matrix. Mutual information value refers to measuring the statistical correlation between features and the target variable. Specifically, the KL divergence algorithm can be used to calculate the correlation strength between sub-features and historical task satisfaction.

[0056] Specifically, in the process of generating rehabilitation tasks, four core main feature dimensions are first defined through an expert system, with quantifiable sub-feature indicators set under each dimension. For example, the rehabilitation needs feature dimension includes sub-features such as rehabilitation goal priority and functional deficit level, while the ability adaptation feature dimension includes sub-features such as visitor skill matching degree and task operation complexity. Next, a standardized formula is used to numerically transform discrete features, for example, converting "acceptance of music therapy" in interaction preferences into a rating value of 0-100. Then, a dynamic weighting algorithm is used to differentiate features under different scenarios, for example, automatically increasing the weight coefficient of the rehabilitation intensity feature during the postoperative recovery period. Finally, mutual information calculation is used to filter out a subset of features strongly correlated with task satisfaction; for example, when the mutual information value between "task duration tolerance" and satisfaction is higher than 0.8, the feature is retained.

[0057] Traditional methods typically use fixed feature sets for task matching, neglecting the dynamic correlations between features and easily introducing redundant features that lead to matching biases. Existing single-dimensional feature selection methods cannot effectively capture implicit correlations between multi-dimensional data. This solution, however, uses mutual information value calculation to accurately assess the correlation between features and target variables. Combined with a dynamic weight allocation mechanism, it can adapt to changes in feature importance across different rehabilitation stages and interaction scenarios. This application effectively solves the problems of feature dimensional redundancy and rigid weight allocation in traditional rehabilitation task matching, dynamically identifying key features that significantly impact task satisfaction and improving the accuracy and effectiveness of feature extraction. This solution combines quantitative analysis and statistical validation to ensure that the selected feature subset aligns with expert experience and has objective data-driven justification, thus providing high-quality feature input for subsequent task matching algorithms and ultimately improving the accuracy of matching rehabilitation tasks with the needs of the elderly and the capabilities of visitors.

[0058] This application further proposes a method for initial task matching in a rehabilitation task library, including the following steps: first, a preliminary screening is performed based on the elderly's needs information, and rehabilitation task templates whose rehabilitation goals are consistent with the rehabilitation goals in the elderly's needs information are selected; then, based on the rehabilitation contraindications in the elderly's needs information, rehabilitation task templates related to rehabilitation contraindications are deleted; then, based on the tolerance intensity in the elderly's needs information, rehabilitation task templates whose task intensity is greater than the tolerance intensity are deleted; finally, based on the time adaptation information, a rehabilitation task template that meets the time range is selected as the initial task.

[0059] The information on elderly needs refers to a structured dataset containing rehabilitation goals, contraindications, and tolerance levels. This can be achieved by integrating medical assessment reports and daily monitoring data to ensure that task selection aligns with individualized rehabilitation needs. Rehabilitation contraindications refer to specific task types or procedures that the elderly must avoid due to their health conditions. This can be obtained through electronic medical records and contraindication lists marked by medical staff to exclude potentially risky tasks. Tolerance levels refer to the maximum intensity of rehabilitation training that the elderly person's physical functions can withstand. This can be determined through a comprehensive analysis of physical fitness test data and historical task completion records to prevent task intensity from exceeding safe limits. Time-appropriateness information refers to the time periods during which both the visitor and the elderly person can participate in rehabilitation interactions. This can be obtained through the integration of both parties' schedule management systems to select time-matched task templates.

[0060] Specifically, the initial task matching process consists of four screening stages. First, based on the rehabilitation goal field in the elderly person's needs information, task templates with matching goals are filtered from the rehabilitation task library. For example, elderly people needing to improve limb mobility are matched with task templates containing joint training. Next, based on the rehabilitation contraindication list, task templates containing contraindicated movements or equipment are automatically removed. For example, tasks involving head-down positions are automatically excluded for elderly people with hypertension. Then, based on the tolerance intensity threshold, templates with intensity levels higher than the threshold are filtered. For example, tasks marked as high-intensity are automatically excluded when the maximum heart rate limit is set as the tolerance intensity. Finally, considering the time compatibility information of both parties, only templates whose task duration and time period requirements match the free time of both parties are retained. For example, tasks marked as 9-11 am and lasting no more than 30 minutes are selected.

[0061] This approach effectively avoids rehabilitation risks caused by the misselection of contraindicated tasks, prevents task intensity from exceeding the elderly person's physical capacity, and ensures that task time is compatible with the participation plan of both parties, thereby improving the accuracy and safety of initial task matching and reducing the frequency of subsequent adjustments.

[0062] This application further proposes to calculate the overall matching degree of the initial task based on a weighted Euclidean distance matching algorithm.

[0063] The weighted Euclidean distance matching algorithm is a mathematical method that evaluates the degree of matching by calculating the weighted distance between two vectors in a multidimensional feature space. Specifically, it can be implemented by multiplying standardized eigenvalues ​​by a dynamic weight matrix and then calculating the sum of squared differences. This algorithm accurately reflects the influence of each feature on task matching by assigning differentiated weights to different feature dimensions, thus solving the problem of insufficient matching accuracy caused by traditional matching methods neglecting the differences in feature importance.

[0064] Specifically, when calculating the overall matching degree, the elderly profile features and visitor profile features are first vectorized and encoded to form a multi-dimensional vector containing features of rehabilitation needs, ability adaptation, and interaction preferences. Next, a dynamic weight matrix is ​​generated based on the feature influence analysis results; for example, the weight of rehabilitation needs features can be set to 0.5, ability adaptation features to 0.3, and interaction preferences features to 0.2. Then, a weighted Euclidean distance is calculated between the task attribute vector of each initial task template and the overall profile feature vector; a smaller distance value indicates a higher matching degree. Finally, the matching degree calculation results of all initial tasks are sorted, and the optimal matching result is selected.

[0065] Through the above technical solution, this approach effectively solves the matching bias problem caused by unreasonable feature weight allocation in the traditional rehabilitation task matching process, and significantly improves the task recommendation system's comprehensive processing capability for multi-dimensional features. By dynamically adjusting the feature weight ratio, the matching results not only meet the core rehabilitation needs of the elderly but also take into account the ability suitability of the visitors, thereby ensuring that the generated rehabilitation tasks have higher operability and execution effectiveness during implementation.

[0066] This application further proposes that after the final task is completed, the elderly's post-task satisfaction, physical feelings, and improvement suggestions should be collected through the elderly's end, and the task suitability rating, operation difficulty feedback, and optimization opinions should be collected through the visitor's end. The initial weights should be adjusted based on the collected feedback data. The rehabilitation task library should be updated based on the collected feedback data. When the elderly's satisfaction with a rehabilitation task template is ≤3 points for 30 consecutive days and the usage rate is ≤5%, it should be removed from the rehabilitation task library. When the feedback on a certain type of uncovered rehabilitation need is ≥20% for 15 consecutive days, a new rehabilitation task template should be designed in conjunction with medical staff, and the task attribute vector should be labeled and added to the rehabilitation task library.

[0067] Among these metrics, post-task satisfaction refers to the elderly person's subjective evaluation of the completed rehabilitation task, which can be implemented using a five-point rating system to reflect the degree to which the task execution effect matches the elderly person's expectations. Physical sensation refers to data on changes in the elderly person's physiological state after task execution, which can be collected through wearable devices to measure indicators such as heart rate, blood pressure, and muscle fatigue, used to assess the impact of task intensity on the elderly person's body. Improvement suggestions refer to the elderly person's suggestions for optimizing the task format or content, which can be collected through voice or text input to identify potential flaws in task design. Task suitability rating refers to the visitor's evaluation of the match between their own abilities and the task requirements, which can be implemented using a multidimensional rating scale to identify discrepancies between skills and task requirements. Operational difficulty feedback refers to records of operational obstacles encountered by the visitor during task execution, which can be collected through preset options or free descriptions to optimize the task operation process. Optimization opinions refer to the visitor's suggestions for improving the task interaction mechanism or resource allocation, which can be collected through structured questionnaires to improve task execution efficiency. Initial weight adjustment refers to dynamically correcting the feature dimension weight allocation rules based on feedback data. This can be implemented using a condition-triggered weight adjustment algorithm, such as automatically triggering a weight adjustment mechanism when a specific type of negative feedback reaches a preset threshold. Rehabilitation task library updates refer to dynamically adding and deleting task templates based on feedback data. This can be implemented using a data-driven task template lifecycle management model, maintaining the effectiveness of the task library by setting removal conditions and addition rules.

[0068] Specifically, after the rehabilitation task is completed, a feedback interface containing rating items and input boxes is automatically pushed to the elderly's device, collecting real-time data on the elderly's subjective evaluation of the task and changes in physiological indicators. The elderly's device includes the aforementioned non-intrusive data collection devices such as smart bracelets and name tags. Simultaneously, a feedback form containing an adaptation score, operation difficulty options, and a comment section is generated on the visitor's end, recording their experience data during task execution. After cleaning, the feedback data is input into a weight adjustment module. When the continuous accumulation of specific types of negative feedback reaches a preset threshold of days, a weight increase mechanism for the corresponding feature dimension is triggered. For example, if the proportion of low satisfaction feedback from the elderly exceeds 30% for seven consecutive days, the weight of rehabilitation needs features in the overall matching degree calculation is automatically increased. At the same time, a task template elimination mechanism is established, automatically removing task templates with consistently low ratings and low usage rates. Furthermore, by analyzing frequently unmet rehabilitation needs, medical staff are collaborated to create new task templates that meet actual needs.

[0069] In some specific implementations, feedback data collection can be combined with speech recognition technology to automatically convert elderly people's verbal evaluations into structured data. For example, "arm training time is too long" can be automatically categorized as duration-related negative feedback. Weight adjustment can use a sliding window algorithm to statistically analyze the feedback distribution over a continuous period, such as calculating the proportion of each type of feedback on a rolling basis over a seven-day period. Task template updates can be implemented by setting a priority queue. For example, when the number of new demand feedbacks reaches a threshold, a task design request can be automatically generated and pushed to the medical staff's workbench.

[0070] Compared to existing technologies, current rehabilitation task libraries are mostly statically configured, lacking a dynamic update mechanism based on usage feedback, leading to a gradual disconnect between task templates and actual needs. Traditional methods rely on manual, periodic checks of task effectiveness, resulting in delayed updates and low efficiency. This solution establishes data-driven, automated update rules that can identify inefficient task templates in real time and promptly supplement missing types, ensuring the task library always dynamically matches the needs of the elderly.

[0071] Through the above technical solutions, this application effectively addresses the long-standing problem of rigid templates in rehabilitation task libraries, preventing inefficient tasks from continuously consuming system resources. The weight adjustment mechanism based on objective feedback data can quickly respond to changes in the experiences of the elderly and visitors, improving the accuracy of subsequent task matching. The automated template update process reduces the cost of manual intervention, while timely addition of new task templates enhances the coverage and adaptability of rehabilitation services.

[0072] This application further proposes a storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of a method for generating personalized information rehabilitation tasks.

[0073] Storage media refers to the physical carrier used to persistently store computer programs and data. This can be achieved using solid-state drives (SSDs), flash memory chips, or cloud storage servers, ensuring the non-volatile storage of program code and runtime data. Computer programs refer to a set of code containing executable instructions. Specifically, object-oriented programming languages ​​can be used to write rehabilitation task matching algorithm modules, profile construction modules, and dynamic optimization modules, which are then compiled into an executable file. Processor execution refers to the parsing and processing of program instructions by the central processing unit (CPU) or graphics processing unit (GPU). Multi-core processors can be used to process elderly profile data and visitor interaction data in parallel, enabling intelligent matching and dynamic adjustment of rehabilitation tasks.

[0074] Specifically, after the computer program stored in the storage medium is loaded into memory, the processor executes the rehabilitation task generation process according to a preset instruction set. First, rehabilitation data of elderly residents in nursing homes and interaction data of visitors are collected to construct a comprehensive profile including health status, skill tags, and time constraints. Next, the feature extraction module analyzes the rehabilitation needs and ability matching features in the profile, selecting an initial set of tasks from a preset rehabilitation task library that match the intensity level and task type. Further, a multi-objective weighted algorithm is used to calculate the matching degree between each task and the profile features, generating a priority-ranked candidate task list. During task execution, the processor monitors changes in the elderly residents' physiological indicators in real time. When abnormal heart rate or excessive exercise amplitude is detected, a task replacement mechanism is automatically triggered, selecting a new task from the remaining candidate tasks that matches the current state. After the task is completed, the processor inputs feedback data from the elderly residents and visitors into the profile optimization module, dynamically adjusting feature weights and updating the rehabilitation task library.

[0075] Compared to existing technologies, traditional nursing home rehabilitation management systems typically use fixed storage devices to store static task templates, failing to achieve programmed task matching and real-time data interaction. The program functionality of existing storage media is limited to basic data recording, lacking intelligent matching algorithms based on dynamic profiles, and task adjustments require manual database manipulation. This solution, however, automates the entire process of data acquisition, feature analysis, task matching, and feedback optimization through a dedicated program, significantly enhancing the data processing value of the storage media.

[0076] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for generating personalized rehabilitation tasks, characterized in that, include: The basic data consists of rehabilitation data of elderly people in nursing homes and interaction data of visitors; The interactive data includes identity attribute information, ability and skill information, time adaptation information, and interaction preference information; A comprehensive profile of the elderly and visitors is constructed based on the rehabilitation data and interaction data. Extract the comprehensive features of the comprehensive profile and match them with the initial task in the rehabilitation task library; The initial task was calculated using a multi-objective weighted calculation based on the elderly and visitors to obtain the matching degree of the initial task. Three tasks were selected as selection tasks based on the matching degree from high to low. The elderly and visitors rate the selected tasks, and the task with the highest overall satisfaction is selected as the final rehabilitation task. It also includes real-time adjustment of rehabilitation tasks: during rehabilitation tasks, real-time data of the elderly is collected; it is determined whether any data in the real-time data exceeds a threshold. If data exceeding the threshold is found, a suitable task type is matched based on the data type; it is determined whether the currently executed rehabilitation task matches the task type. If it matches, the rehabilitation task continues; if it does not match, it is determined whether there is a matching task type among the remaining selected tasks. If a matching task type is found, it is changed; if no matching task type is found, the current rehabilitation task is extracted, and a rehabilitation task with the same properties and matching the current task type is selected and replaced. The construction of a comprehensive profile of the elderly and visitors based on the rehabilitation data and interaction data specifically includes: labeling the rehabilitation data and interaction data to obtain labeled data; constructing a three-dimensional feature matrix, mapping the labeled data to the corresponding dimensions, and dynamically allocating initial weights based on data influence; achieving bidirectional fusion of the profiles of the elderly and visitors through interaction influence factors and feature correlation algorithms to generate a comprehensive related profile; wherein, the interaction influence factor is a parameter reflecting the degree of bidirectional interaction correlation between the elderly and visitors, and the similarity of the behavioral patterns of both parties is calculated through a collaborative filtering algorithm, and the related profile is dynamically corrected based on a time-series decay function and a feedback loop strategy; When the corresponding area includes location data from multiple different elderly users, and the location data retention time exceeds the preset value, the multi-scenario interactive collection strategy activates the backup collection device in the corresponding area, so that each collection device corresponds to one elderly user, and the data collected by the collection device that is closer to the elderly user's location is associated with that location data.

2. The method for generating personalized rehabilitation tasks according to claim 1, characterized in that, The collection of rehabilitation data from elderly residents in nursing homes is conducted using a three-pronged approach: a seamless intelligent device collection strategy, a multi-scenario interactive collection strategy, and an auxiliary verification collection strategy. The seamless intelligent device collection strategy drives intelligent devices to collect real-time data on the elderly residents' heart rate, blood oxygen saturation, body temperature, limb activity, and location without any obstruction throughout the entire process. The multi-scenario interactive collection strategy collects data on the elderly residents' physical condition, emotional state, activity preferences, and subjective preferences interactively during the period when the elderly residents begin their rehabilitation tasks. The auxiliary verification collection strategy collects data on the elderly residents' mental state, social interactions, abnormal behaviors, and daily routines interactively after the rehabilitation tasks have ended and during daily life.

3. The method for generating personalized rehabilitation tasks according to claim 2, characterized in that, The multi-scene interactive acquisition strategy and the smart device non-intrusive acquisition strategy also perform data linkage acquisition. Based on the location data acquired by the smart device non-intrusive acquisition strategy, when the retention time of the location data exceeds a preset value, the multi-scene interactive acquisition strategy automatically locates the acquisition device in the area corresponding to the location data, and associates the data acquired by the acquisition device within the time period corresponding to the retention time with the location data. When the retention time of the location data is less than the preset value, the multi-scene interactive acquisition strategy only records the motion trajectory.

4. The method for generating personalized rehabilitation tasks according to claim 3, characterized in that, Dynamic allocation of initial weights includes: First, the information entropy of each sub-dimension is calculated using the entropy weight method; At the same time, the weights are adjusted based on the scenario coefficient. In the case of rehabilitation training, the weight of the elderly's functional training data is increased, and in the case of social interaction, the weight of the elderly's emotional feedback data is increased.

5. The method for generating personalized rehabilitation tasks according to claim 1, characterized in that, The extraction of comprehensive features from the comprehensive portrait includes: The core main feature dimensions are determined, and key sub-features in the comprehensive profile are associated with each dimension. The core main feature dimensions include rehabilitation needs features, ability matching features, interaction preference features, and constraint condition features. Quantify the associated key sub-features; Normalize and dynamically assign weights to the quantized sub-features; Calculate the mutual information value between each character feature and the task matching satisfaction. The sub-information corresponding to the mutual information value at a set threshold is retained as the final comprehensive feature.

6. The method for generating personalized rehabilitation tasks according to claim 5, characterized in that, The matching of initial tasks in the rehabilitation task library includes: First, a preliminary screening is conducted based on the elderly's needs information, and a rehabilitation task template that matches the rehabilitation goals in the elderly's needs information is selected. Next, based on the rehabilitation contraindications according to the elderly's needs, rehabilitation task templates related to rehabilitation contraindications were deleted; Then, based on the tolerance level in the elderly's needs information, rehabilitation task templates with task intensity greater than tolerance level are deleted. Obtain usage data for rehabilitation equipment and delete rehabilitation tasks for which the corresponding equipment cannot be used within the planned time. Finally, a rehabilitation task template that meets the time range is selected as the initial task based on the time adaptation information.

7. The method for generating personalized rehabilitation tasks according to claim 6, characterized in that, Also includes: After the task is completed, the elderly's post-task satisfaction, physical feelings and improvement suggestions are collected through the elderly's terminal, and the task suitability rating, operation difficulty feedback and optimization opinions are collected through the visitor's terminal. The initial weights are adjusted based on the collected feedback data; The rehabilitation task database is updated based on the collected feedback data.

Images