AI-based nursing service tracking maintenance method, device and equipment

Abstract

The invention discloses an AI-based nursing service tracking maintenance method, device and equipment, relates to the technical field of AI nursing, and is used for solving the problems that the nursing effect is difficult to quantify and the nursing service tracking maintenance cannot be performed on a user in the prior art. Comprising the following steps: acquiring skin data of a target user from a long-term care database, and analyzing the skin data of the target user by adopting a constructed AI framework to obtain a skin state change trend of the target user; dynamically adjusting the nursing scheme of the target user according to the skin state change trend of the target user; and performing nursing service tracking maintenance on the target user according to the nursing scheme. According to the technical scheme provided by the invention, by combining time sequence analysis, multi-modal learning and enhanced decision making, the nursing scheme can be dynamically adjusted according to the skin state change trend of the user, the personalized nursing requirements of the user are met, the nursing effect is improved, and data-driven decision making is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of AI nursing, in particular to a nursing service tracking and maintenance method, device and equipment based on AI. BACKGROUND

[0002] With the increasing demand for skin care and the deep-rooted concept of personalized skin care, the skin care industry is facing unprecedented opportunities and challenges. The traditional skin care service mode mainly relies on manual experience and simple skin type classification, which is difficult to meet the increasingly diversified and refined skin care needs of customers. In the prior art, although some skin care brands have introduced basic skin testing equipment and questionnaires to understand the skin condition of customers, these methods can only obtain limited skin parameters, such as simple moisture content and oil level, and the accuracy and comprehensiveness of the data are insufficient, which cannot in-depth analyze the potential problems and dynamic change trend of the skin.

[0003] In addition, the existing skin care service lacks effective tracking and systematic analysis of customers' long-term skin data. The skin condition of customers is affected by various factors such as age, environment, season, and living habits, and the traditional service mode cannot monitor the impact of these factors on the skin in real time, which makes it difficult to adjust the treatment plan in time, resulting in unsatisfactory nursing effect. At the same time, for the periodic maintenance of skin care products, the existing service also has obvious shortcomings. Most brands can only provide general maintenance suggestions, lack of precise guidance for customers' individual skin characteristics and real-time conditions, making it difficult for customers to achieve the best effect in daily care.

[0004] Therefore, it is urgent to provide a more reliable AI-based nursing service tracking and maintenance solution. SUMMARY

[0005] The purpose of the present application is to provide an AI-based nursing service tracking and maintenance method, device and equipment, which solves the problem that the nursing effect in the prior art is difficult to quantify and cannot track and maintain the nursing service for users.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0007] In a first aspect, the present application provides an AI-based nursing service tracking and maintenance method, which comprises:

[0008] obtaining skin data of a target user from a long-term nursing database; the long-term nursing database at least includes a mapping relationship between skin data and user information;

[0009] adopting the constructed AI framework to analyze the skin data of the target user, to obtain the skin state change trend of the target user; the AI framework is a multi-task integrated learning framework; the AI framework at least includes a time series processing layer and a core prediction layer; the core prediction layer at least includes an LSTM neural network, an XGBoost integrated tree and a random forest classifier;

[0010] According to the skin state change trend of the target user, the nursing scheme of the target user is dynamically adjusted;

[0011] According to the nursing scheme, the target user is tracked and maintained for nursing service.

[0012] Optionally, adopting the constructed AI framework to analyze the skin data of the target user, to obtain the skin state change trend of the target user, includes:

[0013] Adopting the time series processing layer in the AI framework, based on the Prophet algorithm, the skin data of the target user is analyzed, and the time series features are extracted; the time series features at least include the seasonal characteristics and trend items of the skin indicators of the target user;

[0014] The core prediction layer in the AI framework utilizes multiple machine learning models to analyze the time series features, to obtain the skin state prediction result of the target user;

[0015] Adopting the output layer in the AI framework, the skin state prediction result is weighted and fused through the Attention mechanism, to generate the skin state prediction probability matrix of the target user; the skin state prediction probability matrix is at least used to represent the change probability and change trend of the skin state of the target user.

[0016] Optionally, the core prediction layer in the AI framework utilizes multiple machine learning models to analyze the time series features, to obtain the skin state prediction result of the target user, including:

[0017] Adopting the LSTM neural network, based on the time series features, the long-term evolution law of the skin indicators is captured;

[0018] Adopting the XGBoost integrated tree to select the first target feature related to the skin state change in the time series features; the first target feature at least includes the skin indicator feature, the environmental feature and the nursing record feature of the multi-modal data; the first target feature is data fused to obtain a target feature matrix, and a key turning point is predicted based on the target feature matrix;

[0019] The second target feature related to the skin problem combination pattern is selected by using a random forest classifier, and the second target feature at least includes a covariance matrix of skin indicators; feature dimension reduction is performed on the second target feature, and a key feature combination is determined; and a classification result of the skin problem combination pattern is determined based on the key feature combination.

[0020] Optionally, a time series processing layer in the AI framework is used to analyze the skin data of the target user based on a Prophet algorithm, and time series features are extracted, including:

[0021] The skin data of the target user is preprocessed to obtain preprocessed data; the preprocessing operation at least includes timestamp alignment, missing value filling, and normalization processing;

[0022] The Prophet model is initialized, trend item modeling, seasonal modeling, and environmental covariate expansion are performed, and the Prophet model is fitted;

[0023] The skin data of the target user is analyzed based on the Prophet model, and trend item features and seasonal features of skin indicators are extracted.

[0024] Optionally, the output layer in the AI framework is used to weight and fuse the skin state prediction results through an Attention mechanism to generate a skin state prediction probability matrix of the target user, including:

[0025] The output layer collects output results of LSTM neural networks, XGBoost integrated trees, and random forest classifiers;

[0026] The output weight of each output result is determined through an Attention mechanism;

[0027] The output results are weighted and fused according to the output weight to generate a skin state prediction probability matrix.

[0028] Optionally, the care plan of the target user is dynamically adjusted according to the skin state change trend of the target user, including:

[0029] A skin knowledge graph and a care strategy decision tree are constructed according to the skin state change trend of the target user; the skin knowledge graph at least contains multiple skin care rules, various skin problems, and corresponding care measures; and the care strategy decision tree at least includes skin type, environmental factors, care history, and care plan;

[0030] Based on the knowledge graph and the care strategy decision tree, a care effect reward function is constructed, and a DQN algorithm is used to simulate the strategy effect to determine a target reward strategy;

[0031] adjust the care program according to the trend slope of the skin state change trend of the target user;

[0032] monitor the care effect of the care program and feedback data in real time;

[0033] continuously dynamically adjust the care program of the target user based on the care effect and feedback data.

[0034] Optionally, the target user is tracked and maintained according to the care program, including:

[0035] automatically generating a periodic maintenance plan for the target user according to the care program;

[0036] visualizing the periodic maintenance plan and the skin data change of the target user;

[0037] tracking the target user for care services, collecting feedback information of the target user, and adjusting the care program based on the feedback information.

[0038] Compared with the prior art, the present application provides an AI-based care service tracking and maintenance method. By obtaining skin data of a target user from a long-term care database, the skin data of the target user is analyzed using a constructed AI framework to obtain a skin state change trend of the target user. The care program of the target user is dynamically adjusted according to the skin state change trend of the target user. The target user is tracked and maintained according to the care program. The technical solution provided by the present application combines time series analysis, multi-modal learning and reinforcement decision-making to realize real personalized dynamic adjustment. The AI framework is a multi-task integrated learning framework, which at least includes a time series processing layer and a core prediction layer. The core prediction layer at least includes an LSTM neural network, an XGBoost integrated tree and a random forest classifier. The AI framework can quantify the care effect and dynamically adjust the care program according to the skin state change trend of the user, meet the personalized care needs of the user, improve the care effect, and realize data-driven decision-making.

[0039] In a second aspect, the present application provides an AI-based care service tracking and maintenance device, which is applied to the AI-based care service tracking and maintenance method provided in the first aspect. The device comprises:

[0040] a target user skin data acquisition module for obtaining skin data of a target user from a long-term care database; the long-term care database at least includes a mapping relationship between skin data and user information;

[0041] The skin state change trend determination module is configured to analyze the skin data of the target user by using the constructed AI framework to obtain a skin state change trend of the target user; the AI framework is a multi-task integrated learning framework; the AI framework at least includes a time series processing layer and a core prediction layer; the core prediction layer at least includes an LSTM neural network, an XGBoost integrated tree, and a random forest classifier;

[0042] The nursing scheme determination module is configured to dynamically adjust a nursing scheme of the target user according to the skin state change trend of the target user.

[0043] The nursing service tracking maintenance module is configured to track and maintain the nursing service of the target user according to the nursing scheme.

[0044] In a third aspect, the present application provides an AI-based nursing service tracking maintenance device, which comprises:

[0045] a memory, a processor, and a communication interface coupled to the processor; the memory stores a computer program executable by the processor; and the processor executes the computer program to perform the AI-based nursing service tracking maintenance method described above.

[0046] In a fourth aspect, the present application provides a computer storage medium, which stores instructions, and when the instructions are executed, the AI-based nursing service tracking maintenance method described above is implemented.

[0047] The device scheme provided in the second aspect, the equipment scheme provided in the third aspect, and the computer storage medium scheme provided in the fourth aspect achieve the same technical effects as the method scheme provided in the first aspect, which will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 A flow chart of an AI-based nursing service tracking maintenance method provided by the present application;

[0050] Figure 2 A structural schematic diagram of an AI-based nursing service tracking maintenance device provided by the present application;

[0051] Figure 3 A structural schematic diagram of an AI-based nursing service tracking maintenance device provided by the present application. DETAILED DESCRIPTION

[0052] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the terms "first", "second", etc. are used to distinguish the same or similar items with basically the same function and role. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and do not limit the order. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the number and execution order, and the terms "first", "second", etc. also do not necessarily mean different.

[0053] It should be noted that in the present application, the words "exemplary" or "for example" are used to indicate an example, illustration or description. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of the words "exemplary" or "for example" is intended to present the relevant concept in a specific manner.

[0054] In the present application, "at least one" means one or more, and "multiple" means two or more. The association relationship of the associated objects is described, which means that there can be three relationships, for example, A and / or B, which can represent the following cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the front and rear associated objects. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b and c can be single or multiple.

[0055] In recent years, the rapid development of artificial intelligence (AI) technology in various fields has brought new opportunities to the skin care industry. AI technology has shown strong capabilities in image recognition, data analysis, machine learning, and other aspects, and can conduct in-depth mining and intelligent analysis on a large amount of complex skin data. For example, by using high-resolution cameras to obtain images of the skin surface, deep learning algorithms can accurately identify the details and distribution of skin problems such as wrinkles, spots, and acne; with the help of sensors to collect physiological data such as elasticity, texture, and pH value, AI systems can build a comprehensive skin health profile. However, although AI technology has been applied in the skin care industry to some extent, most of these applications are focused on single functions, such as skin problem detection or product recommendation, and there is a lack of a complete system that can integrate multi-source skin data and implement long-term care service tracking and periodic maintenance.

[0056] In summary, the existing skin care service has obvious limitations in personalization, data tracking analysis, and periodic maintenance. Therefore, developing an AI analysis-driven long-term skin care service tracking and periodic maintenance solution for skin care customers can effectively address the shortcomings of existing technology, providing more accurate, efficient, and personalized skin care services for customers, and has important practical significance and market value.

[0057] Next, the scheme provided by the embodiment of the present specification will be described in conjunction with the drawings:

[0058] As shown in the flowchart, the process can include the following steps: Figure 1

[0059] Step 110: Obtain the skin data of the target user from the long-term care database.

[0060] The long-term care database can be a systematic storage and management database for collecting, organizing and analyzing the skin data accumulated by the target user over a long period of time. It not only contains the basic information of the user, but also covers skin index data, care history records, environmental factors and other multi-dimensional information. The long-term care database at least includes the mapping relationship between the skin data and the user information. The user's basic information can include user ID, age, gender, skin type (such as oily, dry, mixed, etc.), health status (such as whether there is a history of allergies, chronic diseases, etc.), and living habits (such as whether to smoke, drink, exercise frequency, etc.), etc. The skin index data of each user can include the moisture value, oil value, elasticity value, erythema index, pigmentation, pore state, and texture of each user's skin, etc. The care history record can include skin care product usage record, care project record, and care effect feedback, etc. The environmental factors can include temperature and humidity, ultraviolet intensity, PM2.5, and seasonal changes, etc.

[0061] For example: The skin data of customers can be collected regularly by the skin tester in the store, including moisture, oil, elasticity, pigmentation, and pore index. The skin data and the basic information of the customers (such as age, gender, and care target) are integrated to form a long-term care database.

[0062] It should be noted that in order to ensure the data security of each user data in the long-term care database, data security and privacy protection can be performed on the long-term care database, such as: data encryption (sensitive data is encrypted and stored to ensure data security), access control (strict access permissions are set to ensure that only authorized personnel can access user data), privacy protection (protect user privacy to prevent data leakage).

[0063] ​Through the above constitution and function, the long-term care database provides clear and comprehensive skin data support for users, helping users to achieve scientific and effective skin care.

[0064] Step 120: using the constructed AI framework to analyze the skin data of the target user to obtain the skin state change trend of the target user.

[0065] More specifically, step 120 can specifically include:

[0066] Using the time series processing layer in the AI framework, the skin data of the target user is analyzed based on the Prophet algorithm to extract time series features; the time series features at least include seasonal characteristics and trend items of the skin indicators of the target user;

[0067] Using the core prediction layer in the AI framework, a variety of machine learning models are used to analyze the time series features to obtain the skin state prediction result of the target user;

[0068] Using the output layer in the AI framework, the skin state prediction result is weighted and fused through the Attention mechanism to generate the skin state prediction probability matrix of the target user; the skin state prediction probability matrix is at least used to represent the change probability and change trend of the skin state of the target user.

[0069] In the specific implementation steps of the above 120, the AI framework is a multi-task integrated learning framework; specifically, the AI framework can realize time series feature extraction and multi-task integrated learning. The AI framework can at least include a time series processing layer, a core prediction layer and an output layer.

[0070] The time series processing layer is responsible for preprocessing the skin data and extracting features from the time series, including trend items and seasonal fluctuations. This provides clearer and more structured input data for the subsequent core prediction layer. Specifically, the time series processing layer can use the Prophet algorithm to decompose the time series data of the skin indicators, identify long-term trends, seasonal changes, and outliers. Then it outputs the processed time series features, such as trend items, seasonal items, or residual items. Trend items reflect the long-term trend of skin indicators over time, and can analyze the trend of growth or decline, the rate of change, etc. For example, by observing the trend chart, we can understand whether the skin indicators are gradually improving, deteriorating, or remaining stable. Seasonal items show the fluctuation pattern of skin indicators in different seasons or cycles, and can analyze the time and amplitude of the peak and trough of skin indicators in different seasons, so as to understand the impact of seasonal changes on the skin. The residual term represents random fluctuations or noise that the model cannot explain, and can analyze the distribution of the residual term to judge the fitting effect of the model. If there are obvious regularities in the residual term, the model parameters may need to be adjusted or other factors affecting the skin indicators may need to be considered.

[0071] The core prediction layer can be used to analyze the output of the time series processing layer using various machine learning models to predict the trend of skin state changes and future problems. It processes multi-modal data (such as skin indicators, environmental data, and care records) to capture complex patterns and relationships. The core prediction layer includes at least LSTM neural networks, XGBoost ensemble trees, and random forest classifiers; LSTM neural networks can handle time dependencies and capture long-term evolution patterns. XGBoost ensemble trees can handle multi-modal data fusion and predict key turning points. Random forest classifiers can identify problem combination patterns. The input of the core prediction layer is the output of the time series processing layer (trend items, seasonal items, etc.). The output of the core prediction layer is the prediction results of each model, including the probability and trend of skin state changes.

[0072] The output layer can be used to integrate the output of the core prediction layer to generate the final skin state prediction results. Specifically, the prediction results of each model can be weighted and fused through the Attention mechanism to generate a comprehensive prediction probability matrix. The output of the core prediction layer (prediction results of LSTM, XGBoost, and random forest) can be used as the input of the output layer, and the output layer outputs the final skin state prediction probability matrix, which is used to generate warning prompts and care suggestions.

[0073] The above AI framework involves feature engineering: constructing sliding window statistics (average / variance / slope of the last 3 detection indicators); extracting frequency domain features (Fourier transform main frequency of skin indicator changes); generating care intervention features (lag effect coefficients of each care project).

[0074] Corresponding training methods are: transfer learning: pre-training model on 100,000 skin medical image dataset; incremental learning: trigger model online update every 500 new customer data; adversarial training: introduce GAN to generate adversarial samples to enhance model robustness.

[0075] Early warning generation mechanism:

[0076] Build a dynamic threshold system: individual baseline (calculate the historical distribution percentile of each customer's indicators), group benchmark (establish a reference interval based on similar groups, such as people with the same skin quality + the same region).

[0077] For example, three-level early warning rules can be set:

[0078] Yellow warning: single indicator deviates from baseline by 2σ for 2 periods;

[0079] Orange warning: 3 associated indicators show a synergistic deterioration trend (Pearson correlation coefficient > 0.7);

[0080] Red warning: predicted model output problem probability > 85% and there is a clinical indication association.

[0081] Through this AI framework, complex skin data can be efficiently processed, future skin problems can be predicted, and personalized care recommendations can be provided to customers.

[0082] For example: use AI algorithms to analyze customer skin data and identify skin state trends. Predict future skin problems (such as dryness, sensitivity, aging, etc.) and generate early warning prompts.

[0083] Step 130: dynamically adjust the target user's care program according to the target user's skin state trend.

[0084] The customer's care program can be dynamically adjusted according to the trend of skin data changes. For example, for the trend of skin moisture decrease, increase the frequency or intensity of hydration care.

[0085] Step 140: track and maintain the target user's care service according to the care program.

[0086] Step 140 can include visual display and user feedback optimization.

[0087] Figure 1The method provided in the application comprises the following steps: obtaining skin data of a target user from a long-term care database; analyzing the skin data of the target user by using a constructed AI framework to obtain a skin state change trend of the target user; dynamically adjusting a care scheme of the target user according to the skin state change trend of the target user; and performing care service tracking and maintenance on the target user according to the care scheme. The technical scheme provided in the application combines time series analysis, multi-modal learning and reinforcement decision-making to realize real personalized dynamic adjustment. The AI framework is a multi-task integrated learning framework, and at least comprises a time series processing layer and a core prediction layer. The core prediction layer at least comprises an LSTM neural network, an XGBoost integrated tree and a random forest classifier. The AI framework can quantize the care effect, dynamically adjust the care scheme according to the skin state change trend of the user, meet the personalized care needs of the user, improve the care effect and realize data-driven decision-making.

[0088] The method provided in the application comprises the following steps: obtaining skin data of a target user from a long-term care database; analyzing the skin data of the target user by using a constructed AI framework to obtain a skin state change trend of the target user; dynamically adjusting a care scheme of the target user according to the skin state change trend of the target user; and performing care service tracking and maintenance on the target user according to the care scheme. The technical scheme provided in the application combines time series analysis, multi-modal learning and reinforcement decision-making to realize real personalized dynamic adjustment. The AI framework is a multi-task integrated learning framework, and at least comprises a time series processing layer and a core prediction layer. The core prediction layer at least comprises an LSTM neural network, an XGBoost integrated tree and a random forest classifier. The AI framework can quantize the care effect, dynamically adjust the care scheme according to the skin state change trend of the user, meet the personalized care needs of the user, improve the care effect and realize data-driven decision-making. Figure 1 The method provided in the application comprises the following steps: obtaining skin data of a target user from a long-term care database; analyzing the skin data of the target user by using a constructed AI framework to obtain a skin state change trend of the target user; dynamically adjusting a care scheme of the target user according to the skin state change trend of the target user; and performing care service tracking and maintenance on the target user according to the care scheme. The technical scheme provided in the application combines time series analysis, multi-modal learning and reinforcement decision-making to realize real personalized dynamic adjustment. The AI framework is a multi-task integrated learning framework, and at least comprises a time series processing layer and a core prediction layer. The core prediction layer at least comprises an LSTM neural network, an XGBoost integrated tree and a random forest classifier. The AI framework can quantize the care effect, dynamically adjust the care scheme according to the skin state change trend of the user, meet the personalized care needs of the user, improve the care effect and realize data-driven decision-making.

[0089] For the specific implementation scheme of step 120, the time series processing layer in the AI framework uses the Prophet algorithm to analyze the skin data of the target user and extracts time series features, which can include:

[0090] The skin data of the target user is preprocessed to obtain preprocessed data; the preprocessing operation at least includes timestamp alignment, missing value filling and normalization processing;

[0091] The Prophet model is initialized to perform trend item modeling, seasonal modeling and environmental covariant extension, and the Prophet model is fitted;

[0092] The skin data of the target user is analyzed based on the Prophet model to extract trend item features and seasonal features of the skin index.

[0093] Next, combined with an actual application scenario example, the specific implementation process of using the Prophet algorithm of the time series processing layer to decompose the seasonality and trend item (such as quarterly skin sensitive fluctuation) of the skin index is further described:

[0094] First step: data preprocessing and time alignment (Prophet algorithm input preparation), the Prophet algorithm requires that the input data is a regular time series, and the specific processing steps include:

[0095] 1) Time stamp alignment: Aggregate the sebum / moisture data collected by the skin detector by hour / day, and accurately align it with the environmental sensor (temperature / humidity, UV intensity) data through time stamp.

[0096] 2) Missing value filling: Use the linear interpolation method built into Prophet to fill in the data during the skin detection interval, while marking abnormal missing values caused by user-initiated device deactivation.

[0097] 3) Normalization: Min-Max normalization of multi-source data (e.g. sebum value 0-100, UV index 0-10) to eliminate differences.

[0098] Prophet directly supports missing value processing and non-uniform sampling time series, avoiding modeling bias caused by data discontinuity in traditional methods.

[0099] Step 2: Model construction and parameter definition (Prophet decomposition engine). Prophet uses an additive model framework of "trend + season + covariate" to customize for skin data:

[0100] 1) Trend modeling: Use piecewise linear functions to fit the long-term changes in skin indicators (e.g. slow decline in barrier function), automatically detect turning points: identify key turning points such as changes in care regimen, seasonal mutations, etc.

[0101] 2) Seasonal modeling: Build periodic basis functions through Fourier series: weekly cycle (difference in skincare habits between weekdays and weekends), monthly cycle (related to skin changes during women's menstrual period), custom annual cycle (e.g. spring and autumn season peak of sensitive skin).

[0102] 3) Environmental covariate extension: Incorporate UV intensity, PM2.5, etc. as additional regression terms to quantify the specific contribution of the environment to seasonal fluctuations. For example: increase the weight of the UV parameter in summer to explain the phenomenon of accelerated sebum oxidation.

[0103] Prophet decouples complex skin fluctuations into components with clear physical meaning through interpretable basis functions.

[0104] Step 3: Decomposition output and application (skin dynamic baseline generation), model output structured decomposition results:

[0105] 1) Trend: Reflects changes in the nature of the skin (e.g. a 3-month increase in moisture value corresponds to barrier repair), typical scenario examples:

[0106] Barrier repair long-term monitoring, scenario: user continuously uses ceramide-based repair skincare for 3 months. Trend item performance: moisture value trend line increases from initial 45 to 62 (+38%), slope significantly increases after 28 days (lag period for repair ingredient to take effect). After Prophet decomposition, seasonal fluctuations (e.g. daily fluctuations ±5 due to air-conditioned rooms) are excluded, confirming the intrinsic barrier function improvement. Application trigger: system automatically generates "barrier repair in progress" status label, recommending gradual replacement of overly occlusive products.

[0107] Chronic photo-damage accumulation, scenario: outdoor worker does not consistently apply sunscreen. Trend item performance: UV damage index trend item shows exponential growth (18% annual growth rate), forming a "three steps forward, one step back" pattern with short-term seasonal repair (decrease in winter). Covariate analysis shows that environmental UV intensity only explains 41% of the variation, with the remaining 59% attributed to cumulative effects. Application trigger: when the second derivative of the trend item is >0, initiate an intensive intervention program (e.g. recommend laser repair treatment).

[0108] 2) Seasonal item: fixed-cycle daily post-noon sebum secretion peak (circadian rhythm), environmental-driven cycle: 15% weekly increase in sensitivity during smog season (covariate correlation), scenario example: dry sensitive skin during winter. Data performance: seasonal item decomposition: from November to the following February, skin moisture value decreases periodically by 25±3% (strong correlation with humidity covariate); environmental correlation: Prophet model quantifies that for every 10% decrease in humidity, the rate of moisture loss increases by 1.8 times; system response is to push a "winter protection package" 2 weeks in advance: containing a high occlusive cream, dynamically adjust moisturizing program, automatically increase mask usage frequency recommendations, etc.

[0109] 3) Residual item: sudden skin events (e.g. single allergic reaction) or measurement noise; for example: sudden allergic reaction event detection, data performance is a sudden increase in residual error, traceability analysis. The sudden increase in residual error is specifically manifested as a sudden 15% increase in skin redness area on a certain day (outside the model prediction interval ±3σ); traceability analysis specifically shows that the residual peak matches the time of the user's new purchase of alcohol-containing toner usage record. System response is to trigger "emergency repair mode" immediately: recommend cold compress + stop using suspicious product list, correct user skin quality label: add "alcohol intolerance" risk label to skin profile.

[0110] As an optional embodiment, the core prediction layer includes three models: LSTM neural network, XGBoost ensemble tree, and random forest classifier. The LSTM neural network can handle the time dependence of skin indicators, capturing long-term evolution rules (such as the 6-month continuous decay rate of elasticity indicators); the XGBoost ensemble tree can handle multi-modal data fusion (skin indicators + environmental data + care records), predicting key turning points; the random forest classifier can identify problem combination patterns based on the skin indicator covariance matrix (such as the abnormal linkage of oil and pore indicators). More specifically, the core prediction layer in the AI framework uses multiple machine learning models to analyze the time series features to obtain the skin state prediction results of the target user, which can specifically include:

[0111] Using the LSTM neural network, the long-term evolution of the skin indicators is captured based on the time series features; the LSTM model usually includes one or more LSTM layers, each containing a certain number of hidden units to capture long-term dependencies in time series. In addition, a Dropout layer can be added to prevent overfitting, and a fully connected layer can be added after the LSTM layer to convert the output to the desired prediction value. The hyperparameters of the model are set, such as learning rate, batch size, number of training rounds, etc. Select an appropriate loss function (such as mean square error) and optimizer (such as Adam) to train the model.

[0112] Using the XGBoost ensemble tree to select the first target feature related to skin state changes in the time series features; the first target feature includes at least skin indicator features, environmental features, and care record features of multi-modal data; data fusion is performed on the first target feature to obtain a target feature matrix, and a key turning point is predicted based on the target feature matrix; an XGBoost library is used to create a model, with basic parameters such as learning rate, tree depth, and sub-sample ratio set. Skin indicators, environmental data, and care records are used as input features for the model, and XGBoost automatically processes these multi-modal data to learn the relationships and interactions between modalities. According to the prediction task, define the target variable, such as the score of the skin state or the identification of the key turning point.

[0113] Using the random forest classifier to select the second target feature related to the skin problem combination pattern, which includes at least the covariance matrix of the skin indicators; performing feature dimensionality reduction on the second target feature and determining the key feature combination; determining the classification result of the skin problem combination pattern based on the key feature combination.

[0114] Next, the specific implementation steps of the above LSTM neural network, XGBoost ensemble tree, and random forest classifier are further described in detail with examples:

[0115] Implementation I, LSTM neural network processes the time dependence of skin indicators, captures long-term evolution rules, and the specific implementation process is as follows:

[0116] 1) Multimodal time series data preprocessing

[0117] Input data sources: ① Dermatometer time series data: daily / weekly collected skin indicators (moisture value / TEWL / redness index); ② Intervention records: skincare product usage time, type (such as the usage frequency of "No. 5 repair mask"); ③ Environmental covariates: temperature / humidity / ultraviolet intensity time series obtained by GPS positioning.

[0118] Key preprocessing steps: ① Time alignment: convert discrete skincare behavior events (such as mask use) into continuous influence curves (for example: set the mask effect duration to 48 hours, encode it as an exponential decay function); ② Missing value filling: predict the skin state during the period when the device is not detected; ③ Dynamic normalization: standardize individual user data while preserving personalized benchmarks.

[0119] 2) LSTM network architecture design

[0120] Use bidirectional time series modeling: consider the influence of historical state and future intervention plan propagation. Intervention effect coding: convert skincare product ingredients (such as ceramide concentration) into feature vectors, and affect cell state update through the gating mechanism. Personalized memory unit: maintain independent long-term memory weights for each user, realize "thousand people thousand models", and build a dedicated skin state evolution model for each user.

[0121] 3) Time-dependent pattern capture mechanism

[0122] Long-term evolution modeling: can extract cross-cycle features (such as setting a sliding window (such as 30 days) to extract the change rate of skin indicators, and track the barrier function decline / repair trend through the LSTM cell state) and key turning point detection (embed a change point detection module in the hidden layer, trigger an alarm when the second derivative of the skin indicator exceeds the threshold, for example: persistent sebum secretion rate > 2% / day for 7 days, marked as "oil-sensitive skin conversion risk period").

[0123] Intervention response modeling: ① Effect delay modeling: set the skincare product action function (such as mask effect = initial boost + exponential decay) ② Synergistic effect learning: capture ingredient-environment interactions through cross-features (such as "nicotinamide serum" x "ultraviolet intensity").

[0124] 4) Typical application scenarios

[0125] Scenario 1, take the repair product effect tracking as an example, input: facial moisture time series data during continuous use of "collagen cream" by the user; LSTM capture: cell state shows a penetration rate inflection point on the 9th day (corresponding to the dermal layer absorption saturation point of clinical testing); system output: automatically push "suggest switching to consolidation stage product" reminder on the 10th day.

[0126] Scenario 2, take the seasonal transition early warning as an example, pattern recognition: the same characteristic of transdermal water loss rate increase appears in the autumn for 3 consecutive years; proactive intervention: adjust the recommended solution 14 days in advance (such as reducing the cleansing frequency from 2 times / day to 1 time / day).

[0127] Implementation two, XGBoost integrated tree processing multi-modal data fusion, predicting key turning points, the specific implementation process is as follows:

[0128] 1) Data integration:

[0129] Time series data (such as skin moisture / oil value): extract sliding window features (past 7-day mean, variance, trend slope), difference features (adjacent time point change rate) and seasonal decomposition items (trend, period, residual extracted by Prophet).

[0130] Environmental data (temperature / humidity / ultraviolet): after aligning the timestamp, calculate the environmental stress index (such as the interaction term of ultraviolet x humidity), and bin processing (such as marking "dry warning" for humidity <30%).

[0131] User questionnaire data: text semantic analysis (such as "frequent overtime" → sleep quality score) and behavior coding (such as the number of skin care steps → one-hot encoding).

[0132] 2) Multi-modal data fusion strategy:

[0133] Feature concatenation: merge time series statistics, environmental index, questionnaire coding, etc. into a unified feature vector, dimension example: (moisture_3daymean, oil_7daytrend, ultraviolet x humidity, sleep quality score, skin care steps_cleaning frequency,...).

[0134] Interaction feature mining: explicit interaction: define key cross features (such as overtime behavior x barrier function trend). Implicit interaction: automatically capture feature combinations through the tree structure of XGBoost (such as humidity drop + winter → dry sensitivity risk).

[0135] Key turning point definition and label generation: first establish a classification task: label whether a state mutation (such as a sudden drop in moisture value >15%) occurs within T days in the future, and a regression task: predict the time point of the mutation (time series regression), and generate labels.

[0136] XGBoost model construction and optimization: remove low-contribution features (e.g., redundant options in the questionnaire) and retain core factors.

[0137] Key turning point prediction and interpretation: output probability value represents the likelihood of future mutation (e.g., probability > 0.7 triggers an early warning).

[0138] Time localization: if it is a regression task, predict the specific time window (e.g., 48-72 hours in the future) when the mutation occurs.

[0139] For example: take the target user's oil-sensitive skin deterioration warning as an example, the data input is the target user's skin data; the time series features are the past 7-day sebum fluctuation rate (+20%), transdermal water loss trend (slope > 0.5 / day); the environmental features are the continuous 3-day ultraviolet index > 8; the behavior features are recent frequent use of soap-based cleanser (questionnaire marked); XGBoost prediction: output probability = 0.89 → trigger "oil-sensitive skin deterioration" red warning.

[0140] Analysis: the main driving factors are "sebum fluctuation rate x ultraviolet" (contribution 52%) and "soap-based usage frequency" (28%). System response care plan: push "anti-inflammatory soothing" first aid plan (including B5 essence + physical sunscreen recommendation) etc.

[0141] Implementation three, random forest classifier identifies problem combination patterns based on skin index covariance matrix, the specific implementation process is as follows:

[0142] 1) Covariance feature matrix construction

[0143] Input data: individual user's continuous 30-day skin index time series data (such as moisture value, sebum secretion rate, trans-epidermal water loss rate, erythema index, etc. 8 indicators); environmental covariates (ultraviolet intensity, air humidity);

[0144] Calculation steps: covariance calculation, feature flattening processing. Specifically, extract the upper triangular elements of the covariance matrix (excluding the diagonal line), and convert the 8x8 matrix into a 28-dimensional feature vector. Append key statistics: maximum covariance value, negative correlation logarithm, principal component variance ratio.

[0145] 2) Problem combination pattern definition and label generation

[0146] Combination pattern: the synergistic change of two or more skin indicators triggers a clinical problem (e.g., "sebum increase + moisture decrease → oily sensitive skin"); then label generation.

[0147] Data augmentation: SMOTE oversampling for minority class samples (e.g., "acne-sebum peroxidation" combination) to balance class distribution.

[0148] 3) Random forest model construction

[0149] Key interaction term reinforcement: prior knowledge driven cross-feature;

[0150] Training strategy: stratified 5-fold cross-validation, ensuring consistent combination pattern proportion in each fold; OOB (Out-of-Bag) error estimation replaces partial validation sets.

[0151] 4) Combination pattern recognition and interpretability analysis

[0152] Decision rule extraction: feature importance ranking (covariance).

[0153] Example path: IF sebum-corneometer thickness covariance > 0.37; AND UV x moisture covariance <-0.12 THEN classify as "photoaging-barrier compromised" combination pattern.

[0154] SHAP interaction analysis: analyze the joint impact of sebum and moisture covariance on prediction. Typical application scenarios include:

[0155] Example 1: Identify acne and sebum peroxide synergy, covariance feature: sebum secretion rate-lipid peroxide concentration covariance > 0.45 (strong positive correlation); corneometer thickness-erythema index covariance <-0.18 (negative correlation). Decision rule: when the above two conditions are met, trigger "inflammatory acne flare" warning; intervention strategy: recommend a combination of products from series 3 and 5, and simultaneously reduce cleansing frequency.

[0156] Example 2: Winter sensitive skin combination pattern, feature combination: moisture-trans epidermal water loss rate covariance <-0.30 (reverse fluctuation); humidity-erythema index covariance > 0.25 (environmentally driven sensitivity). System response: start "low temperature protection mode", increase the proportion of occlusive ingredients in the care program to 60%. Linkage environmental equipment suggestion: use a humidifier, etc.

[0157] In the above examples, in the "time series feature extraction + multi-task integrated learning" framework of skin analysis prediction, the entire process is illustrated with the example of a user taking skin photos and measuring moisture / oil values every day for 30 consecutive days:

[0158] Input data: user's 30-day skin image sequence, sensor measured oil and moisture values;

[0159] Processing process: ① Use LSTM / Transformer model similar to "time magnifying glass" to identify periodic fluctuations in oil secretion (such as premenstrual oil surge); ② Capture the gradual trajectory of blackhead size (such as a linear trend of 0.2mm per week); ③ Find specific patterns of night corneometer water loss (such as a 15% drop from 3-5am);

[0160] Output result: feature vector containing time regularity, for example: (daily oil increase rate 0.8%, water diurnal fluctuation amplitude 22%, pore expansion acceleration 0.05mm / day). 2

[0161] Input reception: the above-mentioned time sequence feature vector; then parallel processing is carried out based on the received time sequence feature vector:

[0162] Pimple prediction branch: focus on analyzing oil trend + pore blockage historical data; case: when it is detected that the oil increase rate exceeds the threshold for 3 consecutive days and the pore shrinkage function decreases, trigger level 2 acne warning;

[0163] Sensitive skin prediction branch: joint analysis of stratum corneum water fluctuation + erythema recurrence frequency; case: when it is found that the night water loss curve is highly consistent with the allergic period last month, it is suggested to repair the barrier;

[0164] Pigmentation branch: correlation between melanin deposition rate and time matching degree of sunscreen habit.

[0165] Feature sharing mechanism: each branch shares the basic feature extractor, but each has its own focus (such as the acne group paying more attention to oil, and the sensitive skin group focusing on barrier indicators).

[0166] Finally, the above results are fused: input processing: branch prediction results (acne risk value 72%, sensitive probability 65%, and color spot index 58%); dynamic weighting: adjust the weight according to the user's recent behavior data, case: when the system detects that the user has recently frequently used acid products, increase the decision weight of sensitive skin prediction; output: ① immediate warning: "detected that your oil increase rate is abnormal this week (+120%), it is suggested to perform deep cleaning tomorrow"; ② trend prediction: "according to the current care method, the probability of acne outbreak before the next menstrual period will reach 85%"; ③ intervention suggestion: "your skin renewal cycle is 28 days, it is suggested to use a scrub mask every weekend".

[0167] As an optional embodiment, in the technical solution provided by the present application, after the constructed AI framework is used to analyze the skin data of the target user and obtain the skin state change trend of the target user, the care scheme of the target user also needs to be dynamically adjusted according to the skin state change trend of the target user. Further, the specific implementation scheme of this step can include:

[0168] ​According to the skin state change trend of the target user, a skin knowledge graph and a care strategy decision tree are constructed; the skin knowledge graph at least contains a plurality of skin care rules, various skin problems and corresponding care measures; the care strategy decision tree at least includes skin type, environmental factors, care history and care scheme; based on the knowledge graph and the care strategy decision tree, a care effect reward function is constructed, and a DQN algorithm is used to simulate the strategy effect to determine the target reward strategy; according to the trend slope of the skin state change trend of the target user, the care scheme is adjusted; the care effect and feedback data of the care scheme are monitored in real time; based on the care effect and feedback data, the care scheme of the target user is continuously and dynamically adjusted.

[0169] Furthermore, the target user is tracked and maintained according to the care scheme, which can include: automatically generating a periodic maintenance plan for the target user according to the care scheme; visualizing the periodic maintenance plan and the skin data change of the target user; tracking the user for care service, collecting feedback information of the target user, and adjusting the care scheme based on the feedback information.

[0170] Next, still in the form of a specific example, "dynamically adjusting the care scheme of the target user according to the skin state change trend of the target user" is described:

[0171] First, 300+ skin care knowledge graphs can be established (for example, when the water loss rate is >15% / month, trigger intensive water light care); then maintain the care strategy decision tree (considering 12 decision dimensions such as skin type, environmental factors, care history, etc.). Build a care effect reward function: R = a*(index improvement degree) + β*(customer satisfaction) + γ*(nursing cost control). Use DQN algorithm to simulate the long-term effect of different adjustment strategies in a virtual environment. According to the trend slope, automatically calculate the adjustment amplitude: water frequency = baseline value × (1 + water loss rate / baseline loss rate). Introduce momentum factor: when the adjustment direction is consistent for 3 times in a row, increase the adjustment strength by 20%. Extract effective combination patterns from successful cases (for example, cleaning + radio frequency project improves the improvement effect of oily skin by 38%); and use graph neural network to simulate the superposition effect of different projects. Map the efficacy requirements in the care scheme (such as antioxidant) to the product ingredient database (containing 3000+ cosmetic ingredient data); cross-compare the history of allergic reactions and the product ingredient table to dynamically exclude risky formulations.

[0172] Then service cycle reconstruction is performed: first, a nursing interval optimization model is constructed: I(t) = ∫(index decay curve - nursing effect curve)dt; the maximum interval period that makes I(t) ≤ threshold is solved. Then the golden time window is automatically scheduled based on the customer visit probability prediction (using a survival analysis model). Customer self-evaluation data (swelling / tightness, etc.) is collected through the mobile terminal 48 hours after each nursing.

[0173] A / B test framework: conduct a strategy control experiment on 5% of users, continuously optimize and adjust the rules. Interpretation module: generate adjustment basis report (such as: "due to 22% increase in melanin deposition rate, it is recommended to increase the frequency of skin care to once every 2 weeks").

[0174] The above embodiments provided by the present application can at least include the following technical effects:

[0175] 1) The present application combines time series analysis, multi-modal learning and reinforcement decision-making to achieve true individualized dynamic adjustment, which improves the persistence of nursing effect and increases customer repurchase rate.

[0176] 2) Quantify nursing effect: through long-term tracking of skin data, customers can intuitively understand the nursing effect and enhance trust.

[0177] 3) Dynamic adjustment of scheme: dynamically adjust the nursing scheme according to the change of skin state to improve the nursing effect.

[0178] 4) Improve customer stickiness: through periodic maintenance reminders and long-term tracking mechanism, enhance customer interaction with the brand and reduce churn rate.

[0179] 5) Scientific maintenance: periodic maintenance plan based on data analysis is more in line with the actual needs of customers, improving the persistence of nursing effect.

[0180] 6) Data-driven decision-making: through the accumulation and analysis of long-term nursing data, provide scientific basis for product research and development and service optimization.

[0181] Based on the same idea, the present application also provides an AI-based nursing service tracking and maintenance device, as shown in Figure 2 The device can include:

[0182] A skin data acquisition module 210 for target users is configured to acquire skin data of target users from a long-term nursing database; the long-term nursing database at least includes a mapping relationship between skin data and user information;

[0183] The skin condition change trend determination module 220 is used to analyze the skin data of the target user using the constructed AI framework to obtain the skin condition change trend of the target user; the AI ​​framework is a multi-task ensemble learning framework; the AI ​​framework includes at least a time series processing layer and a core prediction layer; the core prediction layer includes at least an LSTM neural network, an XGBoost ensemble tree, and a random forest classifier;

[0184] The nursing plan determination module 230 is used to dynamically adjust the nursing plan for the target user based on the skin condition change trend of the target user.

[0185] The nursing service tracking and maintenance module 240 is used to track and maintain nursing services for the target user according to the nursing plan.

[0186] based on Figure 2 The device may also include specific implementation units:

[0187] Optionally, the skin condition change trend determination module 220 may include:

[0188] The time series feature extraction unit is used to analyze the target user's skin data based on the Prophet algorithm using the time series processing layer in the AI ​​framework to extract time series features; the time series features include at least the seasonality and trend of the target user's skin indicators;

[0189] The skin condition prediction result determination unit is used in the core prediction layer of the AI ​​framework to analyze the time series features using multiple machine learning models to obtain the skin condition prediction result of the target user.

[0190] The skin condition prediction probability matrix determination unit is used to perform weighted fusion of the skin condition prediction results through the Attention mechanism in the output layer of the AI ​​framework to generate the skin condition prediction probability matrix of the target user; the skin condition prediction probability matrix is ​​used to at least represent the probability of change and the trend of change of the target user's skin condition.

[0191] Optionally, the skin condition prediction result determination unit can be specifically used for:

[0192] An LSTM neural network is used to capture the long-term evolution pattern of the skin indicators based on the time series features.

[0193] adopting XGBoost ensemble tree to select first target features related to skin state change in the time series features; the first target features at least include skin index features, environmental features and nursing record features of multi-modal data; data fusion is performed on the first target features to obtain a target feature matrix, and a key turning point is predicted based on the target feature matrix;

[0194] adopting a random forest classifier to select second target features related to skin problem combination patterns, the second target features at least including a covariance matrix of skin index; feature dimension reduction is performed on the second target features, and a key feature combination is determined; a classification result of the skin problem combination pattern is determined based on the key feature combination.

[0195] Optionally, the time series feature extraction unit can be specifically used for:

[0196] performing preprocessing operations on the skin data of the target user to obtain preprocessed data; the preprocessing operations at least include timestamp alignment, missing value filling and normalization processing;

[0197] initializing a Prophet model, performing trend item modeling, seasonal modeling and environmental covariate expansion, and fitting to obtain the Prophet model;

[0198] analyzing the skin data of the target user based on the Prophet model, and extracting trend item features and seasonal features of the skin index.

[0199] Optionally, the skin state prediction probability matrix determination unit can be used for:

[0200] the output layer collects output results of the LSTM neural network, the XGBoost ensemble tree and the random forest classifier;

[0201] output weights of each of the output results are determined through an Attention mechanism;

[0202] the output results are weighted and fused according to the output weights to generate a skin state prediction probability matrix.

[0203] Optionally, the nursing scheme determination module 230 can include:

[0204] a nursing strategy decision tree construction unit, configured to construct a skin knowledge graph and a nursing strategy decision tree according to a skin state change trend of the target user; the skin knowledge graph at least includes a plurality of skin care rules, various skin problems and corresponding nursing measures; the nursing strategy decision tree at least includes skin type, environmental factors, nursing history and nursing scheme;

[0205] The target reward policy determination unit is configured to construct a nursing effect reward function based on the knowledge graph and the nursing strategy decision tree, and determine a target reward policy by simulating a policy effect using a DQN algorithm.

[0206] The nursing scheme adjustment unit is configured to adjust the nursing scheme according to the trend slope of the skin state change trend of the target user.

[0207] The real-time monitoring unit is configured to monitor the nursing effect and feedback data of the nursing scheme in real time.

[0208] The dynamic adjustment unit is configured to continuously and dynamically adjust the nursing scheme of the target user based on the nursing effect and feedback data.

[0209] Optionally, the nursing service tracking and maintenance module 240 can include:

[0210] The periodic maintenance plan generation unit is configured to automatically generate a periodic maintenance plan for the target user according to the nursing scheme.

[0211] The visual display unit is configured to visually display the periodic maintenance plan and the skin data change of the target user.

[0212] The nursing service tracking unit is configured to track the nursing service of the user, collect feedback information of the target user, and adjust the nursing scheme based on the feedback information.

[0213] Based on the same idea, the present specification embodiment also provides an AI-based nursing service tracking and maintenance device. As shown in the Figure 3 , the device includes a memory, a processor, and a communication interface coupled with the processor; the memory stores a computer program executable by the processor; and the processor executes the computer program to perform the AI-based nursing service tracking and maintenance method described above.

[0214] In a specific implementation, as an example, as shown in the Figure 3 , the processor can include one or more CPUs, such as CPU0 and CPU1 in the Figure 3 . In a specific implementation, as an example, as shown in the Figure 3 , the terminal device can include multiple processors, such as the processors in the Figure 3 . Each of these processors can be a single-core processor or a multi-core processor.

[0215] Based on the same idea, the present specification embodiment also provides a computer storage medium corresponding to the above-mentioned embodiments, and the computer storage medium stores instructions, which, when executed, implement the method in the above-mentioned embodiments.

[0216] The above describes the scheme provided by the embodiments of the application mainly from the perspective of the interaction between the modules. It can be understood that each module comprises a hardware structure and / or a software unit for performing the corresponding functions. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of the examples described in the embodiments disclosed herein, the application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is performed by hardware or computer software driving hardware depends on the specific application and design constraints of the technical scheme. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0217] The embodiments of the application can divide the functional modules according to the above method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The integrated module can be realized in the form of hardware or software functional module. It should be noted that the division of the modules in the embodiments of the application is illustrative, and is only a logical functional division. There can be another division manner in actual implementation.

[0218] Although the application is described herein in conjunction with various embodiments, other variations of the disclosed embodiments can be understood and implemented by those skilled in the art with reference to the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. Some measures described in mutually different dependent claims can be combined and produce beneficial results.

[0219] Although the application is described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations can be made to it without departing from the spirit and scope of the application. Accordingly, the specification and drawings are to be regarded in an illustrative manner and are to be regarded as covering any and all modifications, variations, combinations or equivalents that fall within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the application without departing from the spirit and scope of the application. Thus, if these modifications and variations of the application fall within the scope of the claims and their equivalents, the application is intended to include them.

Claims

1. An AI-based method for tracking and maintaining nursing services, characterized in that, The methods include: Retrieve the target user's skin data from a long-term care database; the long-term care database includes at least a mapping relationship between skin data and user information; The constructed AI framework is used to analyze the target user's skin data to obtain the skin condition change trend of the target user; the AI ​​framework is a multi-task ensemble learning framework; the AI ​​framework includes at least a time series processing layer and a core prediction layer; the core prediction layer includes at least an LSTM neural network, an XGBoost ensemble tree, and a random forest classifier; The skin care plan for the target user is dynamically adjusted based on the changing trends of the target user's skin condition. Nursing services will be tracked and maintained for the target user in accordance with the nursing plan.

2. The AI-based nursing service tracking and maintenance method according to claim 1, characterized in that, The constructed AI framework is used to analyze the target user's skin data to obtain the target user's skin condition change trend, including: The time series processing layer in the AI ​​framework is used to analyze the target user's skin data based on the Prophet algorithm and extract time series features; the time series features include at least the seasonality and trend of the target user's skin indicators; The core prediction layer in the AI ​​framework uses multiple machine learning models to analyze the time series features and obtain the skin condition prediction results of the target user. The skin condition prediction results are weighted and fused using the Attention mechanism in the output layer of the AI ​​framework to generate a skin condition prediction probability matrix for the target user; the skin condition prediction probability matrix is ​​used to at least represent the probability of change and the trend of change of the target user's skin condition.

3. The AI-based nursing service tracking and maintenance method according to claim 2, characterized in that, The core prediction layer in the AI ​​framework uses multiple machine learning models to analyze the time-series features and obtain the skin condition prediction results for the target user, including: An LSTM neural network is used to capture the long-term evolution pattern of the skin indicators based on the time series features. An XGBoost ensemble tree is used to select a first target feature related to changes in skin condition from the time series features. The first target feature includes at least skin index features, environmental features, and nursing record features from multimodal data. The first target feature is fused to obtain a target feature matrix, and key inflection points are predicted based on the target feature matrix. A random forest classifier is used to select a second target feature related to the combination pattern of skin problems. The second target feature includes at least the covariance matrix of skin indicators. The second target feature is then subjected to feature dimensionality reduction, and key feature combinations are determined. The classification result of the combination pattern of skin problems is determined based on the key feature combinations.

4. The AI-based nursing service tracking and maintenance method according to claim 2, characterized in that, Using the time-series processing layer in the AI ​​framework, the skin data of the target user is analyzed based on the Prophet algorithm to extract time-series features, including: The skin data of the target user is preprocessed to obtain preprocessed data; the preprocessing operation includes at least timestamp alignment, missing value imputation, and normalization. Initialize the Prophet model, perform trend term modeling, seasonality modeling, and environmental covariate expansion, and fit the Prophet model to obtain the Prophet model; Based on the Prophet model, the skin data of the target user is analyzed to extract trend features and seasonal features of skin indicators.

5. The AI-based nursing service tracking and maintenance method according to claim 2, characterized in that, The output layer of the AI ​​framework uses an attention mechanism to weight and fuse the skin condition prediction results to generate a skin condition prediction probability matrix for the target user, including: The output layer collects the output results of the LSTM neural network, XGBoost ensemble tree, and random forest classifier; The output weight of each output result is determined through an attention mechanism; The output results are weighted and fused according to their respective weights to generate a skin condition prediction probability matrix.

6. The AI-based nursing service tracking and maintenance method according to claim 1, characterized in that, Dynamically adjust the target user's skin care plan based on the target user's skin condition change trend, including: A skin knowledge graph and a nursing strategy decision tree are constructed based on the skin condition change trends of the target users; the skin knowledge graph contains at least multiple skin care rules, various skin problems and corresponding care measures; the nursing strategy decision tree includes at least skin type, environmental factors, care history and care plan. Based on the knowledge graph and nursing strategy decision tree, a nursing effect reward function is constructed, and the DQN algorithm is used to simulate the strategy effect to determine the target reward strategy. Adjust the care plan according to the slope of the trend of skin condition changes in the target user; Real-time monitoring of the nursing effect and feedback data of the nursing plan; The nursing care plan for the target user is continuously and dynamically adjusted based on the nursing outcomes and feedback data.

7. The AI-based nursing service tracking and maintenance method according to claim 1, characterized in that, According to the nursing plan, nursing service follow-up and maintenance are carried out on the target user, including: Based on the described care plan, a periodic maintenance plan is automatically generated for the target user; The periodic maintenance plan and the changes in the target user's skin data are presented in a visual format; The nursing service is tracked for the user, feedback information from the target user is collected, and the nursing plan is adjusted based on the feedback information.

8. An AI-based nursing service tracking and maintenance device, characterized in that, The device is applied to the AI-based nursing service tracking and maintenance method according to any one of claims 1 to 7, and the device comprises: The target user's skin data acquisition module is used to acquire the target user's skin data from a long-term care database; the long-term care database includes at least a mapping relationship between skin data and user information. The skin condition change trend determination module is used to analyze the target user's skin data using a constructed AI framework to obtain the target user's skin condition change trend; the AI ​​framework is a multi-task ensemble learning framework; the AI ​​framework includes at least a time series processing layer and a core prediction layer; the core prediction layer includes at least an LSTM neural network, an XGBoost ensemble tree, and a random forest classifier; The nursing plan determination module is used to dynamically adjust the nursing plan for the target user based on the changing trend of the target user's skin condition. The nursing service tracking and maintenance module is used to track and maintain nursing services for the target user according to the nursing plan.

9. An AI-based nursing service tracking and maintenance device, characterized in that the device... include: Memory, processor, and communication interface coupled to the processor; The memory stores computer programs that can be executed by the processor; When the processor runs the computer program, it executes the AI-based nursing service tracking and maintenance method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed, implement the AI-based nursing service tracking and maintenance method according to any one of claims 1 to 7.

Images