A skin soothing efficacy data intelligent management method, system and medium

Abstract

This invention relates to the field of skin data management technology, and discloses a method, system, and medium for intelligent management of skin soothing efficacy data. The method includes constructing a time-series dataset by collecting multi-dimensional skin parameters and environmental variable data at multiple time points, performing quantitative evaluation based on a dynamic evaluation model, establishing an anomaly identification mechanism, generating differentiated intervention strategies, and achieving iterative optimization of the model through backtracking analysis and causal inference. By constructing a complete intelligent management system that integrates data-driven approaches, model prediction, precise intervention, in-depth analysis, and continuous optimization, the accuracy of skin soothing efficacy evaluation, the effectiveness of intervention, and the level of intelligent management are improved, providing a scientific, systematic, and sustainably optimized innovative solution for the field of skin management.

Description

Technical Field

[0001] This invention relates to the field of skin data management technology, and more specifically, to a method, system, and medium for intelligent management of skin soothing efficacy data. Background Technology

[0002] As people's living standards improve, the demand for skin health and soothing management is increasing. Currently, some skin analysis devices on the market, such as AI-FACE, can quantitatively analyze various skin parameters such as pores, pigmentation, wrinkles, redness, and acne based on machine learning and CNN neural network technology. While traditional skin management methods have made some progress in basic skin parameter collection and visualization, they have not yet systematically integrated multi-dimensional skin data to support dynamic tracking of the soothing process, quantitative evaluation of efficacy, and intelligent decision optimization. Therefore, they cannot fully meet users' needs for personalized, continuous skin soothing solutions and effect feedback at different stages.

[0003] Therefore, how to effectively integrate multi-source skin parameters, construct an iterative soothing efficacy evaluation model, realize personalized solution generation and long-term effect tracking, and have a highly adaptable intelligent management method for skin soothing data has become an urgent technical problem to be solved. Summary of the Invention

[0004] This invention provides a method, system, and medium for intelligent management of skin soothing efficacy data, solving the technical problem that existing technologies cannot effectively predict the very early risk of autoimmune diseases.

[0005] This invention provides a method, system, and medium for intelligent management of skin soothing efficacy data, including: Firstly, a method for intelligent management of skin-soothing efficacy data includes: Collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; A dynamic evaluation model for skin soothing effects is constructed, and quantitative evaluation results of skin condition are output based on time-series datasets; Based on the quantitative evaluation results, quantitative indicators for the soothing effect are set, and skin condition evaluation standards are established. Based on the skin condition evaluation criteria, and according to the user's skin soothing response characteristics, the user's skin condition is divided into multiple sub-state clusters; An anomaly identification mechanism was established based on skin condition evaluation standards. Anomalies were identified in multiple sub-category condition clusters to obtain anomaly identification results. Based on the anomaly identification results, generate and execute differentiated relief intervention allocation strategies; The soothing response characteristics of each subclass state cluster after the implementation of the differentiated soothing intervention allocation strategy are obtained, and backtracking analysis and causal inference are performed to obtain comprehensive analysis results; Based on the comprehensive analysis results, spurious correlation factors are eliminated, and real and effective soothing factors are extracted. The soothing factors are then used to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

[0006] Furthermore, the dynamic evaluation model adopts a neural network architecture that integrates gated recurrent units and attention mechanisms. The input is a vector of historical skin state and environmental variables with a sliding window length of 7 days, and the output is the predicted value of skin state at each time step in the next 3 days. The loss function of the dynamic evaluation model is the weighted mean square error, and the weights are set according to the clinical importance of skin parameters. The quantitative indicators of the soothing effect include: The changes in skin parameters are extracted from the quantitative assessment results, and the improvement of single skin parameters, the synergistic improvement of multiple skin parameters, the recovery speed of skin condition, and the stability coefficient of skin condition are calculated. Each quantitative indicator was assigned a different weight according to its clinical importance and normalized to a score of 0-100. A grading evaluation standard was set to obtain five scoring levels, including excellent (85-100 points), good (70-84 points), average (55-69 points), poor (40-54 points), and ineffective (0-39 points).

[0007] Furthermore, based on skin condition evaluation criteria and according to the user's skin soothing response characteristics, the user's skin condition is divided into multiple sub-state clusters, including: The score data of the user's skin parameter improvement range, recovery speed, stability coefficient and synergistic improvement were extracted from quantitative indicators; Extract soothing response characteristic parameters, including repair rate, inflammation resolution slope, barrier reconstruction stability, and pigment metabolism efficiency; Using the soothing response feature parameters as four-dimensional coordinates, the user's skin condition is located in the four-dimensional response space; The DBSCAN clustering algorithm was used to identify skin state point sets with similar soothing dynamics characteristics, and users were divided into four sub-state clusters according to their soothing response patterns: rising response, hysteresis response, fluctuating response, and steady response. Every 7 days, the user's subclass status cluster is recalculated based on updated quantitative indicators for dynamic adjustment, and cross-cluster migration is supported.

[0008] Furthermore, an anomaly detection mechanism is established based on skin condition evaluation standards to identify anomalies in multiple sub-category condition clusters, including: Based on quantitative indicators, multi-level skin condition deviation threshold ranges are set, and differentiated anomaly judgment standards are set for different sub-category state clusters; A deviation score is generated by calculating the Euclidean distance between the current skin parameters and the predicted trajectory; The deviation score is compared with the deviation threshold range of the corresponding subclass status cluster to determine the abnormal level of the current user's skin status in the subclass status cluster. A three-layer progressive anomaly identification architecture is constructed to distinguish between physiological fluctuations, ineffective product reactions, and external interference events; Based on a three-layer progressive anomaly recognition architecture, the anomaly recognition results of each subclass state cluster are output.

[0009] Furthermore, based on the anomaly identification results, differentiated palliative intervention allocation strategies are generated and implemented, including: Based on the anomaly identification results, the characteristic parameters and anomaly levels of each sub-category state cluster are analyzed to identify different types of relief responses and intervention needs. Based on the identified types of soothing responses and intervention needs, corresponding soothing intervention plans are generated for four state clusters: rising response, sluggish response, fluctuating response, and steady response. Based on the aforementioned soothing intervention plan, a differentiated soothing intervention allocation strategy is implemented, and corresponding soothing intervention plans are applied to users in each sub-category of status clusters. The timestamp of the intervention operation, product ingredient code, dosage, site of action, and associated abnormality level are recorded. During the implementation of the intervention program, skin parameter snapshots were continuously collected before and after the intervention to obtain the soothing response characteristics of each sub-category of state clusters after implementing differentiated soothing intervention allocation strategies.

[0010] Furthermore, differentiated palliative intervention allocation strategies include: For the rising response state cluster, which is characterized by a high repair rate and a steep slope of inflammation reduction, a soothing intervention plan is generated: a strategy of prioritizing the supply of highly active ingredients, including high concentrations of niacinamide, centella asiatica extract and other fast-repairing ingredients, to shorten the product iteration testing cycle and increase the weight of new product trials. For the delayed response state cluster, characterized by low repair rate and slow barrier reconstruction, a soothing intervention plan is generated: activate the auxiliary stimulation module, combine microcurrent or phototherapy equipment to improve product penetration efficiency, configure penetration enhancers and long-lasting moisturizing ingredients, and extend the single intervention observation window time. For the fluctuating response state cluster, which is characterized by low stability coefficient and large parameter fluctuation variance, a soothing intervention plan is generated: a sliding window mechanism is used to dynamically adjust the intervention frequency, set an elastic response threshold, and configure mild anti-inflammatory and barrier repair ingredients. For the stable response state cluster, which is characterized by stable repair rate and balanced pigment metabolism efficiency, a soothing intervention plan is generated: maintain the basic care plan, configure stabilizing ingredients and daily moisturizing ingredients, and only trigger fine-tuning of the plan when the season changes or the environment changes. In the soothing intervention program of the rising response state cluster, the component linkage matrix simultaneously activates the recommended weight of synergistic antioxidant components when the core repair ingredients perform well, forming a combined intervention program. In the soothing intervention program for hysteretic response state clusters, when the main active ingredient is not responding well, a derivative ingredient with a similar structure but higher transdermal efficiency is retrieved from the ingredient linkage matrix, the original program is replaced and pushed to the user terminal. Based on the component linkage matrix, a cross-state cluster component early warning mechanism is established. When a component sensitization signal appears in an intervention program for a certain soothing response type, the recommendation of components containing the same sensitizing group in other response type intervention programs is immediately frozen, and a cross-risk report is generated. The differentiated soothing intervention allocation strategy responds in real time to the dynamic changes of each subclass state cluster, and automatically adjusts the soothing intervention plan when the user migrates across clusters.

[0011] Furthermore, the soothing response characteristics of each subclass state cluster after implementing differentiated soothing intervention allocation strategies are obtained, and retrospective analysis and causal inference are performed, including: Extract the soothing response features of each subclass state cluster from the response data after implementing differentiated soothing intervention allocation strategies; The soothing response characteristics include skin parameter improvement rate, intervention response delay duration, rebound frequency, coupling degree of multiple skin parameters, environmental sensitivity coefficient, and product tolerance threshold. The soothing response characteristics are compared with the predicted values ​​of the dynamic evaluation model to identify sub-state clusters whose prediction deviations exceed a preset deviation value. Backtracking analysis is performed on the sub-state clusters whose prediction deviations exceed the preset deviation values. The historical relief intervention records of users in this state cluster are traced back to obtain user life logs, regional climate data, product batch information, and usage compliance records to form a multi-source dataset. Causal inference is performed based on the multi-source dataset. Graph neural networks are used to mine the transmission paths between influencing factors under different soothing response types and to identify soothing influencing factors and their networks for subclass state clusters. Obtain the deviation between the predicted and measured values ​​of skin parameters for each subclass state cluster, plot the deviation heatmap, and use the correlation coefficient to calculate the time-varying correlation between different skin parameters to construct a dynamic correlation network; Based on the aforementioned dynamic correlation network, causal tests are used to infer the driving direction of changes in skin parameters and distinguish the causal transmission patterns of different soothing response types. By integrating deviation heatmaps, dynamic correlation networks, causal transmission models, and influence factor networks, a comprehensive analysis result is generated that includes anomaly root cause localization, failure paths of mitigation interventions, and identification of external interference sources.

[0012] Furthermore, based on the comprehensive analysis results, spurious correlation factors were eliminated, and truly effective mitigation factors were extracted, including: Candidate factors related to the prediction bias of each subclass state cluster are obtained from the comprehensive analysis results, resulting in a candidate factor set; A three-dimensional weighted evaluation framework was constructed to quantitatively score the candidate factor set from three dimensions: factor contribution, time sensitivity, and cross-user consistency, and to identify pseudo-correlation factors. A pseudo-correlation factor screening pipeline was established to perform outlier filtering, redundancy detection, low information content removal, and non-stationarity testing to screen out pseudo-correlation factors in the candidate factor set and obtain a purified factor set. The purified factor set is re-input into the dynamic evaluation model for verification. The change in prediction error before and after removal is compared to obtain the verification results. If the error decreases, the removal of pseudo-correlation factors is confirmed to be effective. Based on the verification results, the verified and effective factors are extracted as mitigation factors for subsequent iterative optimization of the dynamic evaluation model.

[0013] Secondly, a skin-soothing efficacy data intelligent management system includes: Data acquisition module: used to collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; Evaluation module: Used to construct a dynamic evaluation model for skin soothing efficacy, output quantitative evaluation results of skin condition based on time-series dataset; based on the quantitative evaluation results, set quantitative indicators for soothing effect, and set skin condition evaluation standards; Anomaly Detection Module: Based on skin condition evaluation criteria and the user's skin soothing response characteristics, the module divides the user's skin condition into multiple sub-state clusters; it establishes an anomaly detection mechanism based on the skin condition evaluation criteria to identify anomalies in the multiple sub-state clusters and obtain the anomaly detection results. Strategy generation module: Used to generate and execute differentiated relief intervention allocation strategies based on anomaly identification results; Deviation analysis module: used to obtain the soothing response characteristics of each subclass state cluster after the execution of the differentiated soothing intervention allocation strategy, and to perform backtracking analysis and causal inference to obtain comprehensive analysis results.

[0014] Iterative optimization module: This module is used to eliminate spurious correlation factors based on the comprehensive analysis results, extract real and effective soothing factors, and use the soothing factors to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

[0015] Thirdly, a computer-readable storage medium is provided for storing computer-readable instructions that, when read by a computer, enable the execution of the aforementioned intelligent management method for skin-soothing efficacy data.

[0016] The beneficial effects of this invention are as follows: By collecting multi-dimensional skin parameters and environmental variable data at multiple time points to construct a time-series dataset, quantitative evaluation is performed based on a dynamic assessment model, an anomaly identification mechanism is established, differentiated intervention strategies are generated, and iterative optimization of the model is achieved through retrospective analysis and causal inference. This solves the problems of strong subjectivity in assessment, poor intervention targeting, difficulty in effect tracking, and lack of model adaptability in traditional skin soothing management. By constructing a complete intelligent management system that integrates data-driven approaches, model prediction, precise intervention, in-depth analysis, and continuous optimization, the accuracy of skin soothing efficacy assessment, the effectiveness of intervention, and the level of intelligent management are significantly improved. This provides a scientific, systematic, and sustainably optimized innovative solution for the field of skin management, possessing significant clinical application value and promising prospects for wider application. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a method for intelligent management of skin soothing efficacy data provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a skin soothing efficacy data intelligent management system module provided in an embodiment of the present invention. Detailed Implementation

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a method, system, and medium for intelligent management of skin soothing efficacy data, including: like Figure 1 As shown, a method for intelligent management of skin soothing efficacy data includes the following steps: Step 1: Collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; Step 2: Construct a dynamic evaluation model for skin soothing efficacy, and output quantitative evaluation results of skin condition based on time-series datasets; based on the quantitative evaluation results, set quantitative indicators for soothing effect and set skin condition evaluation standards; Step 3: Based on the skin condition evaluation criteria, the user's skin condition is divided into multiple sub-state clusters according to the user's skin soothing response characteristics; an anomaly identification mechanism is established according to the skin condition evaluation criteria to identify anomalies in the multiple sub-state clusters and obtain the anomaly identification results; Step 4: Based on the anomaly identification results, generate and execute differentiated palliative intervention allocation strategies; Step 5: Obtain the soothing response characteristics of each subclass state cluster after implementing the differentiated soothing intervention allocation strategy, and perform backtracking analysis and causal inference to obtain comprehensive analysis results.

[0020] Step 6: Based on the comprehensive analysis results, eliminate spurious correlation factors, extract real and effective soothing factors, and use the soothing factors to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

[0021] In this embodiment, an integrated skin data acquisition terminal is first deployed. This terminal includes a high-precision skin detection module, an environmental sensing module, and a user interaction log recording module. The skin detection module uses multispectral imaging technology to simultaneously acquire at least seven skin parameters, including red area area, melanin index, transepidermal water loss rate, pore density, stratum corneum water content, sebum secretion, and elastic modulus. The environmental sensing module records external variables such as temperature, humidity, UV index, and PM2.5 concentration in real time. The user interaction log recording module collects sleep duration, dietary records, cosmetic usage lists, mood scores, and product adherence check-in information via a mobile application. All raw data is fully sampled every 24 hours and uploaded to the central data processing platform via an encrypted communication protocol.

[0022] After the central data processing platform receives the raw data, it executes a data preprocessing process. For missing values, a hybrid strategy combining time-series interpolation and imputation based on the mean of a similar user group is adopted: if a user has a missing parameter for two consecutive days, linear interpolation between the previous and next days is used first; if the missing value is for more than three days, imputation is performed by referring to the average value of the user group with the same historical cluster label under the same environmental conditions. Outlier removal uses the Tukey interquartile range (IQR) method, with upper and lower bounds set as the first quartile Q1 minus 1.5 times the IQR and the third quartile Q3 plus 1.5 times the IQR, respectively. Here, Q1 is the 25th quartile of the dataset, Q3 is the 75th quartile, and IQR is the difference between the third and first quartiles. Data points exceeding these upper and lower bounds are marked as outliers and removed. Then, all parameters are normalized by Min-Max, and each dimension is mapped to the [0,1] interval to form a structured skin state time series dataset. Each time step contains a skin parameter vector st and an environmental variable vector Et, which are concatenated into an input comprehensive vector Xt=[st;Et] of length 11.

[0023] After data preprocessing, a dynamic evaluation model was constructed. This model uses a gated recurrent unit (GRU) as the backbone network, with a hidden state dimension of 64. A multi-head attention mechanism with 4 heads is added after the output layer of the GRU to capture the importance weights between different time steps. The model input is the historical sequence {X{t-6},…,Xt} extracted by a 7-day sliding window, and the output is the predicted skin parameter values ​​S{t+1}, S{t+2}, S{t+3} for each time step in the next 3 days. The loss function is weighted mean squared error, with weights allocated according to clinical importance: pore density 0.3, red area 0.25, transepidermal water loss rate 0.25, melanin index 0.2, and other parameters 0. The Adam optimizer was used during training, with an initial learning rate of 0.001, a batch size of 32, 200 training epochs, an early stopping mechanism to monitor validation set loss, and a tolerance window of 15 epochs.

[0024] After model training, the skin condition quantitative assessment phase begins. Based on the predicted trajectory output by the dynamic assessment model, four core indicators are calculated: the improvement magnitude of a single parameter is defined as the absolute value of the difference between the current measured value and the baseline value divided by the baseline value; the synergistic improvement degree of multiple parameters is obtained by calculating the consistency ratio of the improvement direction of each parameter, i.e., the proportion of the number of improved parameters to the total number of parameters; the recovery speed is the reciprocal of the number of days required to achieve the first significant improvement (defined as an improvement magnitude exceeding 10%); and the stability coefficient is measured by the reciprocal of the standard deviation of the parameters within a rolling 7-day window. These four indicators are multiplied by preset weights (0.3, 0.25, 0.25, 0.2) and summed, then normalized to 0–100 points through a linear transformation to form a comprehensive score. Based on this score, a five-level evaluation standard is set: Excellent (85–100 points), Good (70–84 points), Average (55–69 points), Poor (40–54 points), and Ineffective (0–39 points). The system establishes a time-series-based dynamic scoring mechanism, automatically triggering a quantitative indicator update process every 3 days (72 hours) to recalculate the user's current four core indicators and comprehensive score, tracking the persistence and stability of the relief effect, and generating a dynamic curve of the relief effect. This dynamic curve uses time as the horizontal axis and the comprehensive score as the vertical axis, employing cubic spline interpolation for smoothing, and marking key event nodes (such as intervention plan adjustments, occurrence of abnormal fluctuations) on the curve, serving as the core data basis for subsequent abnormal fluctuation identification, evidence binding, partitioning strategies, and model optimization.

[0025] Subsequently, user response behavior clustering is performed. The system first extracts scores from the quantitative indicator evaluation results, including the user's skin parameter improvement magnitude, recovery speed, stability coefficient, and synergistic improvement degree. Then, it retrieves the user's historical intervention record chain from the tamper-proof evidence storage system, analyzing the product ingredients, dosage, changes in skin parameters before and after each soothing intervention, and the associated quantitative indicator scores. Based on the aforementioned quantitative scoring data and the intervention records in the evidence storage system, four soothing response characteristic parameters are extracted from the user's historical response data: repair rate (barrier function recovery slope per unit time), inflammation regression slope (red area reduction rate), barrier reconstruction stability (transdermal water loss rate fluctuation coefficient), and pigment metabolism efficiency (melanin index decay rate), forming a four-dimensional feature vector. All user vectors are input into the DBsCAN clustering algorithm, with a neighborhood radius of 0.35 and a minimum sample size of 8, to identify user groups with similar soothing dynamic behaviors. The clustering results are divided into four sub-categories: rising response type (characterized by high repair rate, high inflammation regression slope, and low volatility), hysteretic response type (all indicators are below the population mean and the response delay exceeds 5 days), fluctuating response type (barrier reconstruction stability coefficient is below 0.6 and parameters oscillate repeatedly), and stable response type (all indicators change slowly, with small but continuous improvement). At the end of each 7-day cycle, the user's current four-dimensional feature vector is recalculated, and the user's cluster is adjusted according to the updated DBsCAN clustering results, allowing users to migrate across clusters during different weeks.

[0026] For each sub-category of state clusters, differentiated skin state deviation threshold ranges are set. The threshold range is set to [0.10, 0.20] for rising response, [0.25, 0.35] for hysteresis response, [0.15, 0.25] for fluctuating response, and [0.08, 0.15] for stable response. In daily evaluation, the Euclidean distance between the current measured skin parameter vector Vt and the model predicted trajectory St is calculated as ||Vt - St||², and the normalized Euclidean distance is used as the deviation score. If the Euclidean distance falls below the lower limit of the threshold range for the corresponding cluster, it is judged as normal fluctuation; if it exceeds the upper limit, a three-layer progressive anomaly detection architecture is triggered. The first layer determines whether the Euclidean distance exceeds the physiological fluctuation tolerance zone of ±15% of the predicted value. If not, it is considered a physiological fluctuation. If so, the process proceeds to the second layer, comparing the mean response curve of similar user groups under the same intervention conditions. If the Euclidean distance exceeds the group mean ±2 standard deviations for more than 2 consecutive days, it is determined to be an ineffective product response. If the situation is still unclear, the process proceeds to the third layer, linking the event records in the user logs. If there are events such as a sudden increase in UV exposure intensity (UV exposure intensity ≥8), sleep duration <5 hours, or changing to new cosmetics within 24–72 hours before the deviation occurs, and the spatiotemporal correlation strength (calculated by the Jaccard similarity coefficient) exceeds 0.7, it is marked as an external interference event.

[0027] After anomaly identification, the system establishes an immutable intervention record storage system. Each time a soothing intervention occurs, the system automatically collects basic intervention information including a timestamp (accurate to the second, UTC format), product ingredient code (based on the International Cosmetic Ingredient Nomenclature (INCI) coding system), dosage (in milligrams per square centimeter), and application site (using facial zoning coding, dividing the face into 16 sub-regions: T-zone, U-zone, periorbital zone, and perilipal zone). Simultaneously, it extracts skin parameter snapshots within 24 hours before and after the intervention, including real-time measurements of seven multi-dimensional skin parameters: transepidermal water loss rate, stratum corneum moisture content, erythema index, melanin index, pore density, sebum secretion, and elastic modulus, forming a pre-intervention parameter vector (spre) and a post-intervention parameter vector (spost). The system obtains the comprehensive score and four core indicators (improvement magnitude, recovery speed, stability coefficient, and synergistic improvement degree) corresponding to the current intervention from the quantitative assessment module, forming a quantitative indicator vector for the soothing effect. From the anomaly identification module, it obtains the skin condition anomaly level (normal / mild / moderate / severe), deviation score, and abnormal fluctuation status identifier (physiological fluctuation / product ineffectiveness / external interference) at the current intervention time, forming an anomaly identification result vector, Aresult. The system integrates these four types of data {intervention basic information, skin parameter snapshots (spre, spost), quantitative indicators, and anomaly identification results} into a single intervention record dataset, establishing a precise correspondence between soothing intervention behaviors and changes in skin parameters.

[0028] For each intervention record dataset, the system uses the sHA-256 hash algorithm for encryption to generate a unique 256-bit intervention credential. This credential embeds quantitative indicators and anomaly identification results into a digital fingerprint, ensuring that intervention behavior and effect evaluation are inseparable. Specifically, the system serializes the intervention record dataset into a JSON format string, concatenates the user's unique identifier and the device's hardware fingerprint, and inputs it into the sHA-256 function to calculate the hash value. To achieve chained binding, the system uses a Merkle tree data structure, generating a new leaf node each time an intervention record is added. The hash value of this new leaf node is calculated by combining the current intervention credential, the hash of the preceding node, and the associated quantitative indicator score and anomaly level code. The calculation process is as follows: the current intervention credential, the hash of the preceding node, the quantitative indicator score, and the anomaly level code are concatenated sequentially and then input into the sHA-256 function to calculate the hash value of the new leaf node, forming a time-ordered chained association between each soothing intervention and changes in skin parameters. After each new node is generated, the system immediately performs bottom-up hash propagation of the Merkle tree, updating all parent nodes up to the root node, and calculating the Merkle tree root hash value. The root hash is automatically written to the consortium blockchain nodes every 24 hours or when the cumulative number of newly added intervention records exceeds 50. The consortium blockchain is jointly maintained by medical institution nodes, product manufacturer nodes, and user-authorized nodes, employing a practical Byzantine fault-tolerant consensus mechanism. It requires confirmation from at least two-thirds of the nodes before data can be uploaded to the blockchain, ensuring record consistency and tamper-proof security. The Merkle root hash, once uploaded to the blockchain, serves as the global credential for that batch of intervention records. Any modification to a single record will cause a change in the root hash, thus being detected as an illegal operation by other nodes.

[0029] The system incorporates a smart contract automatic traceability mechanism. When the anomaly detection mechanism detects that a user's skin condition anomaly level reaches the severe threshold (deviation score > 0.40) or remains moderately abnormal for three consecutive days, the smart contract is automatically triggered. The contract traces back along the Merkle tree chain structure from the latest leaf node, analyzing the historical quantitative indicator changes, abnormal fluctuation records, and intervention operation logs in the intervention certificate chain. The system extracts product component codes from all intervention records within the past 30 days, statistically analyzes the frequency of each component and the anomaly level change trend for the corresponding time period, and uses a chi-square test (significance level α = 0.05) to identify suspicious components strongly correlated with the anomaly. Simultaneously, the contract calls upon the product batch information stored in the consortium blockchain (including production date, quality inspection report, and raw material supplier traceability code) and matches it with the time window of the current anomaly event to pinpoint the specific mitigation intervention behavior that triggered the anomaly, the product components, and the responsible party (user's own operational error / product batch quality problem / sudden external environmental change). The traceability results are output in the form of a structured report, which includes the time point of the anomaly trigger, the batch number of the products involved, a list of suspected ingredients (arranged in descending order of relevance intensity), the determination of responsibility and suggested remedial measures. The report is automatically pushed to user terminals, product manufacturers' quality management departments and medical institutions' regulatory platforms to achieve a transparent accountability mechanism that involves multiple parties.

[0030] Based on the data support from anomaly identification results and evidence storage system, a differentiated soothing intervention allocation strategy is generated. For users with an upward response, a high-activity ingredient priority supply strategy is activated, increasing the concentration ratio of core ingredients such as niacinamide, ceramide NP, and asiaticoside, and shortening the usage interval to twice daily. For users with a delayed response, a microcurrent-assisted penetration module is activated, applying a pulse signal with a frequency of 50Hz and a current intensity of 0.3mA for 10 minutes per session, three times a week, combined with red light (630nm) phototherapy to promote cell metabolism. For users with a fluctuating response, a sliding window mechanism is used to dynamically adjust the intervention frequency. The window length is 5 days. If the stability coefficient is below 0.55 for 3 consecutive days, the treatment interval is automatically extended to once every other day, and the potent active ingredients are suspended. For users with a stable response, the basic moisturizing and repair regimen is maintained, with fine-tuning triggered only when environmental variables change abruptly (such as a sudden drop in humidity >30% or a jump in UV exposure intensity >4), temporarily increasing antioxidant ingredients such as ergothioneine 0.5%.

[0031] The system constructs a knowledge graph mapping ingredient efficacy based on historical response data of users with different soothing response types extracted from the evidence storage system. This knowledge graph is stored in the Neo4j graph database. Node types include ingredient nodes (containing attributes such as chemical structure formula, molecular weight, and water-oil balance), efficacy nodes (such as anti-inflammatory, moisturizing, and barrier repair), and user state cluster nodes (corresponding to four response types). Edge types include efficacy, synergistic effect, complementary function, and incompatible combination. The system analyzes tens of thousands of intervention records in the evidence storage system, statistically analyzing the improvement rate, response delay, and rebound frequency of each ingredient in different state clusters. It uses an association rule mining algorithm (Apriori algorithm, minimum support 0.05, minimum confidence 0.7) to identify synergistic effects and incompatible combinations between ingredients. For example, when niacinamide and ceramide (NP) are used simultaneously, if the barrier repair speed of rising response type users increases by more than 1.5 times when either ingredient is used alone (statistical significance < 0.01), a directed edge between niacinamide and ceramide is established in the knowledge graph, with the edge weight set to a synergistic effect coefficient of 1.5. Based on this knowledge graph, the system constructs an ingredient linkage matrix. Matrix elements represent the strength of synergistic relationships between two different ingredients: positive values ​​indicate synergistic effects, negative values ​​indicate antagonism, and values ​​close to zero indicate no significant interaction. The ingredient linkage matrix is ​​further dynamically updated by combining chemical structure similarity (similarity coefficient > 0.7) with a transdermal efficiency database (water-oil balance value, molecular weight, number of hydrogen bond donors / acceptors). When the main active ingredient, such as niacinamide, shows an improvement rate exceeding 80% in rising-response users within 7 days, the recommended weights of magnesium vitamin C phosphate and coenzyme Q10 are automatically increased by 0.2 each. When the improvement rate of hyaluronic acid in lagging-response users is less than 30%, acetylated hyaluronic acid (molecular weight 50kDa, degree of acetylation ≥ 85%) or hydroxypropyl tetrahydropyranotriol (3.2 times increase in transdermal efficiency) is retrieved from the matrix for replacement. If any user in any cluster exhibits an allergic reaction signal (such as an itch score ≥4 or a sudden increase in erythema index >50%), the system immediately freezes recommendations for ingredients in other clusters containing the same allergenic group (such as phenoxyethanol or methylisothiazolinone) and generates a cross-risk report listing all affected users and potentially risky ingredients.

[0032] During the implementation of differentiated soothing intervention allocation strategies, the system continuously collects response data for each sub-category of condition clusters after the corresponding intervention plan is implemented, guided by the strategy. Specifically, the system continuously collects skin parameter snapshots within 48 hours before and after the intervention, extracting six soothing response features from the response data: improvement rate ((baseline - current) / baseline), response delay duration (time to first significant improvement), rebound frequency (number of deteriorations after improvement / total number of cycles), parameter coupling degree (average absolute value of the Pearson correlation coefficient matrix), environmental sensitivity coefficient (absolute value of the regression coefficient between changes in environmental variables and changes in skin parameters), and product tolerance threshold (stimulation score corresponding to the maximum tolerated concentration). The system cross-compares these soothing response features with the scoring data of quantitative indicators (including the improvement magnitude, recovery speed, stability coefficient, and synergistic improvement degree of skin parameters for each sub-category of condition cluster at different time points), and performs residual analysis with the predicted values ​​of the dynamic evaluation model to identify overestimated or underestimated sub-category of condition clusters. If any feature deviation exceeds a preset threshold (improvement rate deviation > 15%, delay > 2 days, rebound frequency > 0.3, etc.), the sub-category of condition cluster is marked as a condition cluster with significant prediction deviation. For the marked state clusters with significant prediction deviations, the system immediately initiates a deep retrospective analysis process. This process traces the historical soothing intervention records of users in that state cluster under the guidance of differentiated soothing intervention allocation strategies. Intervention voucher chains are retrieved from the evidence storage system, and information such as product ingredients, dosage, and intervention time is analyzed. Simultaneously, user lifestyle logs (sleep, diet, mood), regional climate data (from the meteorological bureau API), product batch information (including raw material suppliers, production dates, and quality inspection reports), and usage compliance records are collected (a check-in rate <80% is considered low compliance). The system performs structured coding on the historical response data, quantitative indicator scores, soothing response characteristics, and multi-source influencing factors (lifestyle logs, climate data, batch information, and compliance records) obtained through retrospective analysis. This integrates the entire soothing intervention process data for each sub-category of state clusters (rising response, lag response, fluctuating response, and stable response) to form a survey dataset for skin soothing intervention. This survey dataset includes data dimensions such as user ID, state cluster type, intervention time series, product component vector, skin parameter time series changes, quantitative indicator score series, abnormal event markers, environmental variable matrix, life log characteristics, and usage compliance score, which serve as the data foundation for subsequent bias analysis, correlation analysis, and causal inference.

[0033] Based on the skin soothing intervention survey dataset, the system performs bias analysis, correlation analysis, and causal inference to obtain comprehensive analysis results. First, the actual skin parameter change trajectories of each sub-category state cluster (rising response, hysteretic response, fluctuating response, and stable response) are extracted from the survey dataset. This involves the time series of measured values ​​of seven skin parameters (red area, transepidermal water loss rate, stratum corneum moisture content, melanin index, pore density, sebum secretion, and elastic modulus) during the soothing intervention process. Simultaneously, the predicted skin parameter change trajectories corresponding to each sub-category state cluster are extracted from the dynamic evaluation model to obtain the predicted skin parameter value sequence based on historical data. For each sub-category of state clusters, the system calculates the L2 norm deviation between the predicted and measured values ​​of each skin parameter at each time point (t=1,2,...,30 days), and plots a deviation heatmap for each state cluster. The horizontal axis of the heatmap represents time (1-30 days), and the vertical axis represents the seven skin parameters. The color depth indicates the magnitude of the deviation (using the Viridis color scheme, dark colors represent high deviation, and light colors represent low deviation), which is used to locate the high deviation periods and skin parameter dimensions in each soothing response type.

[0034] Based on the deviation heatmaps of each subclass state cluster, the system performs correlation analysis, using Pearson and Spearman correlation coefficients to calculate the time-varying correlations between different skin parameters. Specifically, a rolling 7-day window is used to calculate the pairwise correlation coefficients between the seven skin parameters within each time window, generating a time-varying correlation coefficient matrix sequence. For different soothing response types, the system constructs a dynamic correlation network, with network nodes representing the seven skin parameters. The edge strength is determined by the time-varying correlation coefficient matrix sequence (a threshold is set: |time-varying correlation coefficient matrix sequence| > 0.6 to establish an edge). The edge strength evolves with the time window, used to identify differences in coupling patterns between parameters in different state clusters.

[0035] Furthermore, the system performs causal inference analysis. In the dynamic correlation network of each subclass state cluster, the system identifies the pivotal skin parameters (such as transdermal water loss rate and red area) with the highest centrality, and uses Granger causality tests to infer the driving direction of changes in these parameters. For each pair of skin parameters (i,j), a vector autoregressive model (lag coefficient of 3) is constructed to test whether the historical value of parameter i has a significant predictive ability for the current value of parameter j (F-test, significance level of 0.05). If the test rejects the null hypothesis (lag order < 0.055), a causal arrow i→j is established, indicating that the change in parameter i drives the change in parameter j. Based on the Granger causality test results, the system identifies causal transmission patterns for different soothing response types. For example, in the fluctuating response type, there may be a causal chain of transdermal water loss rate → red area → melanin index, while the causal relationship is weaker in the stable response type.

[0036] Based on the survey dataset, the system constructs a graph neural network model to explore the transmission paths of influencing factors. Nodes in the graph neural network model represent influencing factors (such as insufficient sleep, high UV exposure, batch #2023A, and low product compliance), and edge weights are determined by the factor co-occurrence frequency and time-lag correlation. Through the message passing mechanism of the graph neural network model (using a graph attention network (GAT) architecture with an attention parameter of 8), the system identifies key intermediary nodes and the network of influence of influencing factors. This network describes how different influencing factors ultimately affect changes in skin parameters through transmission paths.

[0037] Finally, the system integrates the deviation heatmaps, dynamic correlation networks, causal transmission patterns, and influencing factor networks of each sub-category of state clusters to generate a comprehensive analysis result that includes the localization of abnormal root causes (e.g., excessive preservatives in batch #2023A leading to a continuous increase in transdermal water loss rate in users with delayed response), the failure path of soothing interventions (e.g., from microcurrent inactivation to insufficient penetration to delayed barrier repair and then to a decrease in stability coefficient), and the identification of external interference sources (e.g., continuous UV exposure intensity >9 for 3 days triggering oxidative stress → melanin index rebound). This comprehensive analysis result is output in the form of a three-dimensional analysis report and aligned with the soothing influencing factors and their networks in the survey dataset to screen out candidate factor sets that are significantly correlated with the prediction deviations of each sub-category of state clusters.

[0038] Candidate factor sets were extracted from the comprehensive analysis results, and a three-dimensional weighted evaluation framework was constructed. Factor contribution was calculated using sHAP values, and time sensitivity was defined as the decay rate of factor influence on the 3rd and 7th days after intervention. Cross-user consistency was measured using the Kappa coefficient to assess the consistency of the factor's direction of action among different users. A pseudo-association factor screening pipeline was implemented on the candidate factor set: first, outliers were removed using Tukey's rule; second, variance inflation factor was calculated, and redundant variables with variance inflation factor > 10 were removed; third, the variance of each factor was calculated using a rolling 30-day window, and low-information factors with variances lower than 5% of the total variance were removed; finally, the ADF unit root test was performed, and first-order differencing was applied to non-stationary sequences (p > 0.05). The purified factor set was re-input into the dynamic evaluation model, and the root mean square error before and after screening was compared. If the root mean square error decreased by more than 5%, the screening was confirmed to be effective, and the factors in the purified factor set were used as the truly effective mitigating factors to update the model's input feature set and weight allocation.

[0039] Furthermore, the system incorporates a dynamic monitoring mechanism for interference factors. It continuously calculates the influence weight (based on the average weights from the attention mechanism) and sHAP contribution of each input factor. If a factor's weight is <0.02 and its sHAP contribution is <0.5% for 14 consecutive days, it is automatically removed from the input variable set, triggering L2 regularized redistribution of the remaining factor weights to ensure the total weight sum is 1. Simultaneously, a backtesting verification procedure is initiated, calculating the mean absolute error (MAE) of the model before and after removal on an independent test set, and performing a paired t-test. If the lag order is <0.051, the removal is confirmed as effective; otherwise, the system returns to the monitoring mechanism, relaxing the weight threshold to 0.015 and adjusting the sHAP threshold to 0.8%, re-evaluating, and forming a closed-loop optimization process. Through these steps, the entire system achieves a complete intelligent management closed loop from data collection, modeling and evaluation, clustering and typing, anomaly identification, strategy generation to model self-evolution, ensuring the objectivity of skin soothing efficacy evaluation, the accuracy of intervention, and the system's continuous adaptability.

[0040] To enable those skilled in the art to fully understand and implement this invention, the following explanation of the implementation principle of this invention is provided in conjunction with a specific application scenario.

[0041] On the 10th day of a user's continuous use of soothing skincare products, the system obtained a snapshot of the user's skin parameters through an integrated skin data acquisition terminal: the red area was 12.3%, the transepidermal water loss rate was 18.7 g / m² / h, the stratum corneum moisture content was 42.5%, the environmental sensor module recorded the humidity as 35%, the UV index (UV exposure intensity) as 9, and the user's interaction log showed that the user slept for 4.2 hours the previous night and used a new facial cleanser. After the above raw data was encrypted and uploaded to the central data processing platform, a missing value check was first performed, confirming no missing values. Then, the Tukey interquartile range (IQR) method was used to identify outliers. It was found that the transepidermal water loss rate was within the boundary of the third quartile (Q3) of the historical distribution plus 1.5 times the IQR, but not exceeding it, so it was retained. Finally, all 11-dimensional input vectors Xt were Min-Max normalized to form a standardized time-series input.

[0042] The user's current cluster is an ascending response type. Its dynamic evaluation model, based on the previous 7-day sequence {X{t-6},…,Xt}, predicts that the red area on day t+1 should decrease to 10.8% and the transdermal water loss rate should decrease to 16.5 g / m² / h. However, the actual measured red area on day t+1 is 13.1%, and the transdermal water loss rate rises to 20.4 g / m² / h. The calculated Euclidean distance is 0.28, and the normalized deviation score is 0.31, exceeding the upper limit of the preset threshold range [0.10, 0.20] for ascending response types. The system then triggers a three-layer progressive anomaly identification architecture: The first layer determines whether d{t+1} exceeds the predicted value ±15% physiological tolerance zone. The calculated range of ±15% for the predicted value of 10.8% in the red zone is [9.2%, 12.4%]. The actual measured value of 13.1% has exceeded this range, so the system proceeds to the second layer. The second layer retrieves the average response curve of 200 rising response type users under the same intervention program. It finds that the average value of the red zone for the group is 11.0% ±1.2%. This user's 13.1% has been higher than the average value +2σ (i.e., 13.4%) for two consecutive days, which does not meet the condition of exceeding ±2σ for more than two consecutive days. Therefore, the system does not determine that the product is invalid. The system then proceeds to the third layer. It correlates with the user logs and finds that there are three events in the 48 hours before the deviation occurred: UV exposure intensity ≥8 (actual UV exposure intensity = 9), sleep <5 hours (4.2 hours), and changing to new cosmetics. The Jaccard spatiotemporal correlation coefficient between the events and the deterioration of skin parameters is calculated to be 0.78, which exceeds the 0.7 threshold. Therefore, the event is marked as an external interference event. This abnormal event, along with the current intervention baseline information (timestamp 2024-03-15T08:23:47Z, product ingredient code INCI-NAM-05%, dosage 0.8 mg / cm², application site T zone), skin parameter snapshots (spre=[18.7,42.5,12.3,...],spost=[20.4,40.1,13.1,...]), quantitative indicator score (overall score 52 points, poor grade), and abnormality identification results, was analyzed. (Moderate anomaly level, deviation score 0.31, status identifier external interference) are integrated into intervention record dataset 10. The system immediately performs sHA-256 hash operation to generate a unique intervention certificate H10=sHA256 (intervention record dataset), and links it with the previous node hash H9 to generate a new leaf node of Merkle tree (where 52 is the quantitative score and 2 is the moderate anomaly level code). The Merkle tree root hash is updated and synchronized to the consortium chain node to ensure that the anomaly event is recorded immutably in the evidence storage system.

[0043] Based on this identification result, the system did not adjust the core ingredient ratio, but instead generated a temporary intervention suggestion: suspend the use of the new facial cleanser for 3 days, add an antioxidant serum containing 0.5% ergothioneine at night, and push a UV protection reminder. Simultaneously, this event was recorded as an external perturbation sample and included in the survey dataset. In the subsequent 15-day periodic evaluation, if similar deviations are commonly observed among similar users under conditions of high UV exposure intensity + sleep deprivation, the graph neural network (graph neural network model) will enhance the edge weights between nodes of high UV exposure and sleep deprivation, and verify through Granger causality tests whether changes in transdermal water loss rate precede erythema deterioration—if the lag coefficient < 0.055, a directed causal edge between transdermal water loss rate and the red area will be established in the dynamic correlation network to correct the driving logic of the inflammatory response in the next round of prediction models.

[0044] In another application scenario, a user with a delayed response reached a severe level of skin abnormality (deviation score of 0.43) on day 18, triggering an automatic traceability process via the system's smart contract. The contract traced back along the Merkle tree chain structure in the evidence storage system, from the latest leaf node Nleaf18 to Nleaf1, parsing the intervention credential chain of all 15 intervention records for the user within the past 18 days. The system extracted the product component coding sequences from the credential chain and statistically analyzed the frequency of each component and the corresponding abnormality level trends over the time periods. Chi-square test results (test parameter 12.73, df=4, p=0.013<0.05) showed a significant correlation between the component INCI-HA-01% (hyaluronic acid) and the increased abnormality level. The system further accessed the product batch information stored in the consortium blockchain and discovered that the hyaluronic acid product used by the user from day 12 to 17 had batch number Batch#2024B0312. The quality inspection report for this batch showed a molecular weight of 1500kDa (high molecular weight hyaluronic acid). However, this user had a delayed-response skin type with low barrier permeability, making it difficult for the high molecular weight hyaluronic acid to penetrate effectively, resulting in an improvement rate of only 18%. The smart contract generated a structured traceability report, pinpointing the anomaly trigger time as 22:35 on day 15, involving the product batch number, with the suspected ingredient being hyaluronic acid (too high molecular weight). The responsibility was determined to be a mismatch between the product formula and the user's condition cluster. The recommended remedial measure was to replace it with acetylated hyaluronic acid with a molecular weight of 50kDa and increase microcurrent-assisted penetration. This report was automatically pushed to the user's terminal, the product manufacturer's quality management department, and the medical institution's monitoring platform. Upon receiving the report, the product manufacturer initiated a batch review process within 24 hours and pushed ingredient optimization suggestions to all delayed-response users, achieving transparent accountability and rapid response.

[0045] When the system detected that another user with a delayed response had only achieved a 22% improvement in transdermal moisture loss rate after using a product containing hyaluronic acid for 7 consecutive days (below the 30% threshold), the component linkage matrix immediately searched the transdermal efficiency database and screened acetylated hyaluronic acid (molecular weight 50kDa, degree of acetylation 87%, water-oil balance value -1.8, number of hydrogen bond donors 3) as a replacement candidate due to its molecular structure similarity coefficient of 0.73 and in vitro transdermal tests showing a 3.1-fold increase in penetration rate. The system automatically replaced the original hyaluronic acid in the formula with 0.2% acetylated hyaluronic acid and issued the replacement instruction to the user's terminal device in the next intervention command. Skin snapshots collected within the following 48 hours showed that the transdermal moisture loss rate decreased to 15.2 g / m² / h, an improvement rate of 38%, verifying the effectiveness of the replacement.

[0046] During the model self-optimization phase, the system found that the "emotional rating" factor had a mean attention weight of 0.018 and a sHAP contribution of 0.42% over 21 consecutive days, which was below the removal threshold. The system removed it from the input variable set and performed L2 regularization redistribution on the remaining 10 factors to maintain a total weight sum of 1. Backtesting was conducted on an independent test set containing 500 users. The MAE before and after the removal was 0.087 and 0.086, respectively. The paired t-test showed p=0.12>0.01, which was not statistically significant. Therefore, the monitoring mechanism automatically relaxed the threshold to weight <0.015 and sHAP <0.8%, and re-evaluated. Two weeks later, the factor weight rebounded to 0.021, and the system terminated the removal process, maintaining the original feature set, thus avoiding the accidental deletion of potentially slow-acting influencing factors.

[0047] Through the above mechanism, this invention achieves dynamic tracking, anomaly attribution, and strategy optimization of the individual skin soothing process, and ensures that the intervention plan adapts to the evolution of the user's state through multi-source data fusion and closed-loop model iteration.

[0048] like Figure 2 As shown, a skin-soothing efficacy data intelligent management system includes: Data acquisition module: used to collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; Evaluation module: Used to construct a dynamic evaluation model for skin soothing efficacy, output quantitative evaluation results of skin condition based on time-series dataset; based on the quantitative evaluation results, set quantitative indicators for soothing effect, and set skin condition evaluation standards; Anomaly Detection Module: Based on skin condition evaluation criteria and the user's skin soothing response characteristics, the module divides the user's skin condition into multiple sub-state clusters; it establishes an anomaly detection mechanism based on the skin condition evaluation criteria to identify anomalies in the multiple sub-state clusters and obtain the anomaly detection results. Strategy generation module: Used to generate and execute differentiated relief intervention allocation strategies based on anomaly identification results; Deviation analysis module: used to obtain the soothing response characteristics of each subclass state cluster after the execution of the differentiated soothing intervention allocation strategy, and to perform backtracking analysis and causal inference to obtain comprehensive analysis results.

[0049] Iterative optimization module: This module is used to eliminate spurious correlation factors based on the comprehensive analysis results, extract real and effective soothing factors, and use the soothing factors to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

[0050] The present invention further proposes a computer-readable storage medium for storing computer-readable instructions, which, when read by a computer, can execute the aforementioned intelligent management method for skin soothing efficacy data.

[0051] When computer program instructions are loaded and executed on a computer, the corresponding functions are implemented according to the process provided in the embodiments of this invention. The computer program instructions involved may be assembly instructions, machine instructions, or code written in a programming language, etc.

[0052] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for intelligent management of skin soothing efficacy data, characterized in that, include: Collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; A dynamic evaluation model for skin soothing effects is constructed, and quantitative evaluation results of skin condition are output based on time-series datasets; Based on the quantitative evaluation results, quantitative indicators for the soothing effect are set, and skin condition evaluation standards are established. Based on the skin condition evaluation criteria, and according to the user's skin soothing response characteristics, the user's skin condition is divided into multiple sub-state clusters; An anomaly identification mechanism was established based on skin condition evaluation standards. Anomalies were identified in multiple sub-category condition clusters to obtain anomaly identification results. Based on the anomaly identification results, generate and execute differentiated relief intervention allocation strategies; The characteristics of the soothing response of each subclass state cluster after the implementation of differentiated soothing intervention allocation strategies are obtained, and backtracking analysis and causal inference are performed to obtain comprehensive analysis results; Based on the comprehensive analysis results, spurious correlation factors are eliminated, and real and effective soothing factors are extracted. The soothing factors are then used to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

2. The intelligent management method for skin soothing efficacy data according to claim 1, characterized in that, The dynamic evaluation model adopts a neural network architecture that integrates gated recurrent units and attention mechanisms. The input is a vector of historical skin state and environmental variables with a sliding window length of 7 days. The output is the predicted value of skin state at each time step in the future preset period. The loss function of the dynamic evaluation model is the weighted mean square error, and the weights are set according to the clinical importance of skin parameters. The quantitative indicators of the soothing effect include: The changes in skin parameters are extracted from the quantitative assessment results, and the improvement of single skin parameters, the synergistic improvement of multiple skin parameters, the recovery speed of skin condition, and the stability coefficient of skin condition are calculated. Each quantitative indicator was assigned a different weight according to its clinical importance and normalized to a score of 0-100. A grading evaluation standard was set to obtain five scoring levels, including excellent (85-100 points), good (70-84 points), average (55-69 points), poor (40-54 points), and ineffective (0-39 points).

3. The intelligent management method for skin soothing efficacy data according to claim 1, characterized in that, Based on skin condition evaluation criteria, and according to the user's skin soothing response characteristics, the user's skin condition is divided into multiple sub-category state clusters, including: The score data of the user's skin parameter improvement range, recovery speed, stability coefficient and synergistic improvement were extracted from quantitative indicators; Extract soothing response characteristic parameters, including repair rate, inflammation resolution slope, barrier reconstruction stability, and pigment metabolism efficiency; Using the soothing response feature parameters as four-dimensional coordinates, the user's skin condition is located in the four-dimensional response space; The DBSCAN clustering algorithm was used to identify skin state point sets with similar soothing dynamics characteristics, and users were divided into four sub-state clusters according to their soothing response patterns: rising response, hysteresis response, fluctuating response, and steady response. The system dynamically adjusts the user's subclass status cluster based on updated quantitative indicators every preset period, and supports cross-cluster migration.

4. The intelligent management method for skin soothing efficacy data according to claim 1, characterized in that, An anomaly detection mechanism is established based on skin condition evaluation standards to identify anomalies in multiple sub-category condition clusters, including: Based on quantitative indicators, multi-level skin condition deviation threshold ranges are set, and differentiated anomaly judgment standards are set for different sub-category state clusters; A deviation score is generated by calculating the Euclidean distance between the current skin parameters and the predicted trajectory; The deviation score is compared with the deviation threshold range of the corresponding subclass status cluster to determine the abnormal level of the current user's skin status in the subclass status cluster. A three-layer progressive anomaly identification architecture is constructed to distinguish between physiological fluctuations, ineffective product reactions, and external interference events; Based on a three-layer progressive anomaly recognition architecture, the anomaly recognition results of each subclass state cluster are output.

5. The intelligent management method for skin soothing efficacy data according to claim 4, characterized in that, Based on the anomaly identification results, a differentiated palliative intervention allocation strategy is generated and implemented, including: Based on the anomaly identification results, the characteristic parameters and anomaly levels of each sub-category state cluster are analyzed to identify different types of relief responses and intervention needs. Based on the identified types of soothing responses and intervention needs, corresponding soothing intervention plans are generated for four state clusters: rising response, sluggish response, fluctuating response, and steady response. Based on the aforementioned soothing intervention plan, a differentiated soothing intervention allocation strategy is implemented, and corresponding soothing intervention plans are applied to users in each sub-category of status clusters. The timestamp of the intervention operation, product ingredient code, dosage, site of action, and associated abnormality level are recorded. During the implementation of the intervention program, skin parameter snapshots were continuously collected before and after the intervention to obtain the soothing response characteristics of each sub-category of state clusters after implementing differentiated soothing intervention allocation strategies.

6. The intelligent management method for skin soothing efficacy data according to claim 5, characterized in that, Differentiated palliative intervention allocation strategies include: For the rising response state cluster, which is characterized by a high repair rate and a steep slope of inflammation reduction, a soothing intervention plan is generated: a strategy of prioritizing the supply of highly active ingredients, including high concentrations of niacinamide, centella asiatica extract and other fast-repairing ingredients, to shorten the product iteration testing cycle and increase the weight of new product trials. For the delayed response state cluster, characterized by low repair rate and slow barrier reconstruction, a soothing intervention plan is generated: a penetration enhancer and long-lasting moisturizing ingredients are configured, and the observation window time of a single intervention is extended. For the fluctuating response state cluster, which is characterized by low stability coefficient and large parameter fluctuation variance, a soothing intervention plan is generated: a sliding window mechanism is used to dynamically adjust the intervention frequency, set an elastic response threshold, and configure mild anti-inflammatory and barrier repair ingredients. For the stable response state cluster, which is characterized by stable repair rate and balanced pigment metabolism efficiency, a soothing intervention plan is generated: maintain the basic care plan, configure stabilizing ingredients and daily moisturizing ingredients, and only trigger fine-tuning of the plan when the season changes or the environment changes. In the soothing intervention program of the rising response state cluster, by using the ingredient linkage matrix, when the core repair ingredients perform well, the recommended weight of the synergistic antioxidant ingredients is simultaneously activated to form a combined intervention program. In the soothing intervention program for hysteretic response state clusters, when the main active ingredient is not responding well, a derivative ingredient with a similar structure but higher transdermal efficiency is retrieved from the ingredient linkage matrix, the original program is replaced and pushed to the user terminal. Based on the component linkage matrix, a cross-state cluster component early warning mechanism is established. When a component sensitization signal appears in an intervention program for a certain soothing response type, the recommendation of components containing the same sensitizing group in other response type intervention programs is immediately frozen, and a cross-risk report is generated. The differentiated soothing intervention allocation strategy responds in real time to the dynamic changes of each subclass state cluster, and automatically adjusts the soothing intervention plan when the user migrates across clusters.

7. The intelligent management method for skin soothing efficacy data according to claim 1, characterized in that, Obtain the soothing response characteristics of each subclass state cluster after implementing differentiated soothing intervention allocation strategies, and perform retrospective analysis and causal inference, including: Extract the soothing response features of each subclass state cluster from the response data after implementing differentiated soothing intervention allocation strategies; The soothing response characteristics include skin parameter improvement rate, intervention response delay duration, rebound frequency, coupling degree of multiple skin parameters, environmental sensitivity coefficient, and product tolerance threshold. The soothing response characteristics are compared with the predicted values ​​of the dynamic evaluation model to identify sub-state clusters whose prediction deviations exceed a preset deviation value. Backtracking analysis is performed on the sub-state clusters whose prediction deviations exceed the preset deviation values. The historical relief intervention records of users in this state cluster are traced back to obtain user life logs, regional climate data, product batch information, and usage compliance records to form a multi-source dataset. Causal inference is performed based on the multi-source dataset. Graph neural networks are used to mine the transmission paths between influencing factors under different soothing response types and to identify soothing influencing factors and their networks for subclass state clusters. Obtain the deviation between the predicted and measured values ​​of skin parameters for each subclass state cluster, plot the deviation heatmap, and use the correlation coefficient to calculate the time-varying correlation between different skin parameters to construct a dynamic correlation network; Based on the aforementioned dynamic correlation network, causal tests are used to infer the driving direction of changes in skin parameters and distinguish the causal transmission patterns of different soothing response types. By integrating deviation heatmaps, dynamic correlation networks, causal transmission models, and influence factor networks, a comprehensive analysis result is generated that includes anomaly root cause localization, failure paths of mitigation interventions, and identification of external interference sources.

8. The intelligent management method for skin soothing efficacy data according to claim 7, characterized in that, Based on the comprehensive analysis results, spurious correlation factors were eliminated, and the truly effective mitigating factors were extracted, including: Candidate factors related to the prediction bias of each subclass state cluster are obtained from the comprehensive analysis results, resulting in a candidate factor set; A three-dimensional weighted evaluation framework was constructed to quantitatively score the candidate factor set from three dimensions: factor contribution, time sensitivity, and cross-user consistency, and to identify pseudo-correlation factors. A pseudo-correlation factor screening pipeline was established to perform outlier filtering, redundancy detection, low information content removal, and non-stationarity testing to screen out pseudo-correlation factors in the candidate factor set and obtain a purified factor set. The purified factor set is re-input into the dynamic evaluation model for verification. The change in prediction error before and after removal is compared to obtain the verification results. If the error decreases, the removal of pseudo-correlation factors is confirmed to be effective. Based on the verification results, the verified and effective factors are extracted as mitigation factors for subsequent iterative optimization of the dynamic evaluation model.

9. A skin soothing efficacy data intelligent management system, used to execute the skin soothing efficacy data intelligent management method as described in any one of claims 1-8, characterized in that, include: Data acquisition module: used to collect multi-dimensional skin parameters and environmental variable data at multiple time points, and form a time-series dataset of skin condition after preprocessing; Evaluation module: Used to build a dynamic evaluation model for skin soothing efficacy, and outputs quantitative evaluation results of skin condition based on time-series datasets; Based on the quantitative evaluation results, quantitative indicators for the soothing effect are set, and skin condition evaluation standards are established. Anomaly Detection Module: Based on skin condition evaluation criteria and the user's skin soothing response characteristics, the module divides the user's skin condition into multiple sub-state clusters; it establishes an anomaly detection mechanism based on the skin condition evaluation criteria to identify anomalies in the multiple sub-state clusters and obtain the anomaly detection results. Strategy generation module: Used to generate and execute differentiated relief intervention allocation strategies based on anomaly identification results; Deviation analysis module: used to obtain the soothing response characteristics of each subclass state cluster after the implementation of the differentiated soothing intervention allocation strategy, and to perform backtracking analysis and causal inference to obtain comprehensive analysis results; Iterative optimization module: This module is used to eliminate spurious correlation factors based on the comprehensive analysis results, extract real and effective soothing factors, and use the soothing factors to iteratively optimize the dynamic evaluation model for the next round of iterative evaluation of skin condition.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable instructions, which, when read by a computer, enable the execution of a skin-soothing efficacy data intelligent management method as described in any one of claims 1-8.

Images