Data processing method and device for integrated virtual employee care Saas platform

Abstract

The invention provides a data processing method and device for an integrated virtual employee care Saas platform. The data processing method comprises the steps of obtaining associated data of employees through the integrated virtual employee care Saas platform; the associated data comprises historical behavior data, working state data and emotion fluctuation data; adjusting and optimizing a preset psychological state recognition algorithm based on the associated data to obtain an adjusted and optimized psychological state recognition algorithm; evaluating the associated data based on the tuning psychological state recognition algorithm, generating a psychological state evaluation result of the employee, and automatically pushing care scheme information when the psychological state evaluation result of the employee triggers an early warning threshold value; and encrypting and storing the associated data by adopting a dynamic encryption algorithm. According to the invention, the accuracy of psychological state assessment is improved, and the data security is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method and device for an integrated virtual employee care SaaS platform. Background Art

[0002] With the widespread adoption of digital office work and the evolution of corporate management philosophies, employee care has become a core component of enhancing corporate cohesion and productivity. Demand for mental health support and career development guidance is particularly prominent. Currently, some companies are beginning to adopt AI-assisted tools or single-function platforms to provide employee care services, such as using chatbots to answer basic psychological questions and using independent systems to conduct career planning assessments.

[0003] However, the existing technology has the following significant limitations: Employees' behavioral data (such as platform usage records), work status data (such as task completion status), and emotional fluctuation data (such as emotional expressions in text / voice) are usually stored in different systems. There is a lack of cross-dimensional correlation analysis, resulting in psychological state assessment relying on a single data source and insufficient accuracy.

[0004] Most platforms use a universal mental state recognition algorithm that does not take into account individual differences among employees (such as personality traits, job stress thresholds) and dynamic changes (such as career stages, the impact of life events), resulting in large deviations between assessment results and actual needs.

[0005] Employee emotional fluctuation data, psychological assessment results, etc. are highly sensitive information. Existing storage methods have risks such as data leakage, especially in cross-platform data transmission and long-term storage, where privacy protection is insufficient. Summary of the Invention

[0006] The main purpose of the present invention is to provide a data processing method and device for an integrated virtual employee care SaaS platform, aiming to overcome the defects of current psychological state assessment that relies on single data, resulting in inaccurate assessment results and easy data leakage.

[0007] To achieve the above objectives, the present invention provides a data processing method for an integrated virtual employee care SaaS platform, comprising the following steps: Obtain employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; Tuning a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; Evaluate the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically push care plan information when the employee mental state evaluation result triggers a warning threshold; The associated data is encrypted and stored using a dynamic encryption algorithm.

[0008] Furthermore, the integrated virtual employee care SaaS platform integrates mental health support, career planning advice, social interaction services and life guidance functional modules into the same SaaS platform. The SaaS platform receives employee service requests through a unified interface and automatically matches the corresponding functional modules according to the request type.

[0009] Furthermore, the emotional fluctuation data includes the emotional tendency value of the text chat record and the intonation feature parameters of the voice interaction.

[0010] Furthermore, the preset mental state recognition algorithm is tuned based on the associated data to obtain a tuned mental state recognition algorithm, including: Extracting personalized strong correlation features from the correlation data; where, for the same employee in the same scenario, the Pearson correlation coefficient between the negative indicator in the emotional fluctuation data and the task delay rate in the work status data is ≥0.7, and it has occurred stably ≥4 times in the past six months, this is marked as a personalized strong correlation feature; Constructing a correction factor based on the personalized strong correlation feature; The correction factor is embedded in the feature extraction layer of the preset mental state recognition algorithm, and the feature weight of the data matched with personalized strong correlation features in the algorithm is automatically increased. At the same time, the weight of the general features of the group that are not related to the employees is frozen, thereby obtaining a tuned mental state recognition algorithm.

[0011] Furthermore, the calculation formula of the correction factor is: Correction factor = (individual characteristic value - group mean) / group standard deviation × 0.6 + characteristic occurrence frequency × 0.4.

[0012] Furthermore, the method further comprises: A points incentive mechanism is established based on employees' frequency of service usage and interactive participation on the platform. Points can be exchanged for professional psychological assessment services or career planning courses, and points can be shared and experiences can be shared among employees through the social interaction module.

[0013] Furthermore, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Classifying the sensitivity of the associated data; Determine a corresponding splitting rule based on the level corresponding to each associated data, split the corresponding associated data into 512-byte encrypted data blocks based on the splitting rule, and append a unique check code to each encrypted data block; The encrypted data blocks are stored in three physically isolated storage nodes, and the encrypted data blocks stored in any one storage node do not exceed 40% of the total; When calling the associated data, the calling conditions need to be met: The IP address of the calling node is in the preset whitelist; Obtain dynamic authorization tokens from three storage nodes, with token generation intervals within the preset time, and the tokens contain the version number of the data block reorganization algorithm; Through real-time facial feature verification of employees; After all three conditions are met, the reassembly algorithm is automatically triggered to splice the complete data according to the check code sequence.

[0014] Furthermore, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Calculate the data activity index of the associated data using the following formula: (access frequency within a preset time period × 0.5) + (number of associated warning triggers × 0.3) + (data update frequency × 0.2); When the data activity index is greater than a first preset value, the data is encrypted and stored using the AES-256-GCM algorithm, with an additional quantum-resistant hash lock, and automatically re-encrypted every preset time. When the data activity index is less than a second preset value, the core data in the associated data is retained and the original detail data is deleted, and encrypted storage is performed using a lightweight encryption algorithm; the first preset value is greater than the second preset value; When the data activity index is not greater than a first preset value and not less than a second preset value, a preset encryption algorithm is used for encrypted storage.

[0015] Furthermore, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Acquire multi-dimensional features of the associated data and classify them to obtain first dimensional features, second dimensional features, and third dimensional features; Obtain an encryption algorithm array; wherein the encryption algorithm array is an array including a plurality of encryption algorithms; Based on the first dimensional feature, the encryption algorithm array is deformed to obtain a deformed encryption algorithm array; Screening the array of deformed encryption algorithms based on the second dimensional feature to obtain a target encryption algorithm; The target encryption algorithm is tuned based on the third dimension feature to obtain a tuned encryption algorithm; and the associated data is encrypted and stored based on the tuned encryption algorithm.

[0016] The present invention also provides a data processing device for an integrated virtual employee care SaaS platform, comprising: An acquisition unit is used to acquire employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; a tuning unit, configured to tune a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; An evaluation unit, configured to evaluate the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically push care plan information when the employee mental state evaluation result triggers a warning threshold; The storage unit is used to encrypt and store the associated data using a dynamic encryption algorithm.

[0017] The data processing method and device of the integrated virtual employee care SaaS platform provided by the present invention include: obtaining employee related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; tuning a preset psychological state recognition algorithm based on the related data to obtain a tuned psychological state recognition algorithm; evaluating the related data based on the tuned psychological state recognition algorithm to generate an employee psychological state evaluation result, and automatically pushing care plan information when the employee psychological state evaluation result triggers a warning threshold; and encrypting and storing the related data using a dynamic encryption algorithm. In the present invention, related data of multiple dimensions of employees are obtained through the integrated virtual employee care SaaS platform, and the psychological state recognition algorithm is tuned, thereby improving the accuracy of psychological state evaluation; and the related data are encrypted and stored using a dynamic encryption algorithm to ensure data security. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the steps of a data processing method of an integrated virtual employee care SaaS platform in one embodiment of the present invention; Figure 2 This is a structural block diagram of a data processing device of an integrated virtual employee care SaaS platform in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0019] The implementation, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0021] Reference Figure 1 In one embodiment of the present invention, a data processing method for an integrated virtual employee care SaaS platform is provided, comprising the following steps: Step S1: obtaining employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; Step S2, tuning a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; Step S3, evaluating the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically pushing care plan information when the employee mental state evaluation result triggers a warning threshold; Step S4: encrypt and store the associated data using a dynamic encryption algorithm.

[0022] In this embodiment, as described in step S1 above, the goal is to achieve integrated collection of multi-dimensional employee data, providing a foundation for subsequent psychological status assessment and algorithm tuning. Specifically, the SaaS (Software as a Service) platform simultaneously obtains three types of core related data through a unified data interface system: Historical behavioral data: This includes employee activity within the platform, such as mental health consultation records (including consultation duration and question types), visit frequency and duration of visits to the career planning module, content of speeches in the social interaction module, and feedback on others' posts (likes, comments, and sentiment). This data is tracked throughout the entire process through logging and behavioral tracking technology. Work status data: Connect to internal enterprise OA, ERP and other systems to obtain employees' task completion progress (such as delay rate, quality score), overtime hours and frequency, and communication response speed in cross-departmental collaboration. At the same time, it can be linked to physiological indicators (such as sleep duration and average daily step count) synchronized with wearable devices (such as smart bracelets) to form linked data on work and physiological status; Emotional fluctuation data: Natural language processing (NLP) technology is used to extract emotional keywords (such as "stress" and "anxiety") and semantic tendencies in employees' text chats. This is combined with intonation features during voice interaction (such as pitch fluctuation and speech speed changes) and micro-expression features during video interaction (such as frown frequency and eye contact duration) to build a multimodal emotional feature library.

[0023] After deduplication and cleaning, the above data is stored in association with the timestamp and the employee's unique identifier (such as the employee ID hash value) to ensure the integrity and relevance of the data.

[0024] As described in step S2 above, the preset mental state recognition algorithm is tuned based on the associated data to obtain a tuned mental state recognition algorithm. The algorithm's recognition accuracy of individual employee mental states is improved through a dynamic learning mechanism. The core of this is to continuously optimize the algorithm parameters using the associated data. In one embodiment, this specifically includes: Sample pool construction: Valid training samples are screened from the linked data. Samples are required to contain complete behavioral, work-emotional features and cover a time span of at least 90 days. Samples are labeled by emotion type and influencing factors (e.g., "work stress-induced anxiety" and "life event-induced depression"). Samples with a labeling consistency rate of ≥85% are included in the core training set. Feature weight iteration: Using a gradient descent algorithm, the weights of the algorithm's feature extraction layer are adjusted based on individual data in the sample set. For example, if an employee's "working overtime for more than four hours for three consecutive days" and "a sudden increase of 30% in the proportion of negative words in the text" are strongly correlated (Pearson correlation coefficient ≥ 0.7), the weight of this combined feature in the algorithm will be automatically increased. Personalized Calibration: A two-dimensional model, Individual Baseline and Group Reference System, is constructed to address individual employee differences. Using an employee's average psychological state three months after joining the company as the individual baseline, this data is compared to the average data of the same position group and the deviation is calculated. When the deviation consistently exceeds 2 times the group standard deviation, a lightweight sub-model is customized for the individual. This approach achieves personalized algorithm adaptation by freezing the weights of common features and emphasizing unique individual characteristics (such as mood fluctuations before project deadlines).

[0025] It is understandable that through parameter iteration every 30 days, the algorithm's recognition accuracy of individual psychological states has increased by ≥10% compared to the initial version.

[0026] As described in step S3 above, the associated data is evaluated based on the tuned psychological state recognition algorithm to generate employee psychological state assessment results. When the warning threshold is triggered, care plan information is automatically pushed to achieve a closed loop from data evaluation to intervention execution. The core is to output accurate assessment results through the tuned algorithm and link the intelligent intervention mechanism.

[0027] The specific process is: The optimized algorithm performs a multi-dimensional analysis of the relevant data collected in real time, outputting an assessment result that includes emotion type (such as anxiety, burnout), impact level (1-10 points), duration, and related factors (work / life ratio). During the assessment process, the algorithm prioritizes individual calibrated feature weights to ensure that the results are tailored to the individual employee's characteristics. Three levels of pre-set warning thresholds (based on industry psychological crisis standards and historical company data) are available. For example, a Level 1 warning (mild fluctuation) is when the emotional score deviates from the personal baseline by 10%-20%, a Level 2 warning (significant fluctuation) is when the emotional score deviates by 20%-40% and persists for 72 hours or longer, and a Level 3 warning (high-risk fluctuation) is when the emotional score deviates by 40% or more or when extreme emotional signals (such as keywords related to self-harm tendencies) are detected. Automatically match differentiated solutions based on warning levels: Level 1 warning: Push preventive content, such as meditation audio and time management tool recommendations based on historical preferences; Level 2 warning: triggers a real-time conversation with the AI ​​psychological counseling robot, and simultaneously pushes an anonymous confession portal and a professional psychological assessment questionnaire; Level 3 warning: Immediately connect with the contracted psychological counselor resource pool, push the contact information of professionals who can respond immediately within a 30-kilometer radius, and generate an encrypted emotional snapshot (key features after desensitization) to assist in rapid intervention.

[0028] At the same time, the platform records employees' response data to care plans (such as click-through rate and completion rate) to provide feedback for subsequent algorithm tuning.

[0029] As described in step S4 above, the associated data is encrypted and stored using a dynamic encryption algorithm, and the storage security of the associated data (especially sensitive information) is guaranteed through a multi-level security mechanism, taking into account both privacy protection and data availability.

[0030] In one embodiment, it may include: Data is divided into three levels according to sensitivity: Level 1 sensitive data (such as raw psychological assessment scores and speech emotion feature sequences) uses a chaotic encryption algorithm. The key is generated from employee biometrics (such as fingerprint hash values), platform timestamps (accurate to milliseconds), and random prime numbers. The key is automatically rotated every 10 encrypted data items, and the key fragments are distributed and stored on three physically isolated nodes. Secondary sensitive data (such as work performance scores and overtime records) uses a hybrid mode combining the AES-256 algorithm and homomorphic encryption. This supports statistical operations (such as monthly task completion rate calculations) without decryption. The key is bound to the employee's dynamic password (updated every 60 seconds). Level 3 sensitive data (such as service access frequency) uses conventional hash encryption, retaining only statistical features; Establish a role-based access control matrix. For example, psychologists accessing primary data require triple verification: biometrics, dynamic password, and approval authorization. Single access durations are 10 minutes or less, and operation logs are uploaded to the blockchain in real time for evidence. For primary data older than 180 days and without triggering an alert, automatic desensitization is performed (for example, retaining sentiment trend curves but removing specific interaction content), mitigating storage risks while preserving data analysis value. In this embodiment, dynamic encryption combined with permission management ensures data security throughout the entire storage, transmission, and access process.

[0031] In one embodiment, the integrated virtual employee care SaaS platform integrates mental health support, career planning advice, social interaction services and life guidance functional modules into the same SaaS platform. The SaaS platform receives employee service requests through a unified interface and automatically matches corresponding functional modules according to the request type.

[0032] In this embodiment, the mental health support module provides services such as psychological assessment, emotional counseling, and emergency psychological intervention, including a connection channel with professional psychologists and an anonymous confession function. Career planning advice module: Integrates career assessment tools, industry development information, and skills improvement course recommendations to support employees in querying job promotion paths and analyzing their ability matching; Social interaction service module: Builds an interactive space among employees, including interest groups, experience sharing areas, and team collaboration topic areas, supporting real-time chat, dynamic posting, and likes and comments; Life guidance function module: provides life service recommendations (such as nearby convenience facilities, parent-child activity information), health management suggestions (such as diet matching, exercise plans) and other life content.

[0033] The SaaS platform receives employee service requests through a unified access interface (e.g., web portal, mobile app, enterprise WeChat plug-in, etc.). When an employee initiates a request (e.g., "I'm stressed and need help" or "I want to learn about promotion requirements"), the request intent is analyzed using natural language understanding technology, automatically matching the request to the corresponding functional module and accessing relevant service resources. For example, if an employee sends a text request stating "I've been feeling stressed at work lately," the platform will directly jump to the mental health support module, prioritizing the emotional self-assessment tool and the instant counseling robot entrance. Employees no longer need to manually switch between modules, achieving a one-stop experience with multiple services through a single login, significantly reducing operational processes and improving employee convenience.

[0034] In one embodiment, the emotion fluctuation data includes the emotion tendency value of the text chat record and the intonation feature parameters of the voice interaction.

[0035] In this embodiment, the sentiment value of a text chat record refers to the result of a quantitative sentiment analysis of employee text interactions within the platform (such as one-on-one chats, social media posts, and psychological counseling messages) using natural language processing (NLP) technology. Specifically, the text is segmented to identify explicit emotional words (such as "anxiety," "tiredness," and "happiness") and implicit emotional expressions (such as "I've been suffering from insomnia lately" versus "overly stressed"). Based on a preset sentiment dictionary (containing positive, negative, and neutral lexicons), each word is assigned a weight (for example, "collapse" is assigned a value of -0.8, and "relaxation" is assigned a value of +0.6). Ultimately, a weighted calculation is performed to determine the sentiment value of the entire text (typically ranging from -1 to +1, with negative values ​​indicating negative sentiment and positive values ​​indicating positive sentiment. Larger absolute values ​​indicate higher sentiment intensity).

[0036] Intonation feature parameters for voice interaction: refers to the quantitative indicators of acoustic feature extraction of employee voice signals in scenarios such as voice calls and voice messages. Key parameters related to intonation are extracted through speech recognition technology, including: Pitch fluctuation: The amplitude of the change in the fundamental frequency (base frequency) of the speech signal. For example, the amplitude of pitch fluctuation during emotional excitement usually exceeds 30%. Speech rate variability: The change in the number of syllables per unit time. For example, in an anxious state, speech rate may increase by more than 20% compared to normal levels. Pause duration: the time interval between natural pauses in speech. For example, when someone is depressed, the pause interval may be extended to more than 1.5 times the normal level. The above parameters are converted into computable feature vectors through algorithms such as Mel-Met Cepstral Coefficients (MFCC), which are used to reflect the emotional fluctuation state in the speech (such as tension, calmness, excitement, etc.).

[0037] In one embodiment, optimizing a preset mental state recognition algorithm based on the correlation data to obtain an optimized mental state recognition algorithm includes: Extracting personalized strong correlation features from the correlation data; where, for the same employee in the same scenario, the Pearson correlation coefficient between the negative indicator in the emotional fluctuation data and the task delay rate in the work status data is ≥0.7, and it has occurred stably ≥4 times in the past six months, this is marked as a personalized strong correlation feature; Constructing a correction factor based on the personalized strong correlation feature; The correction factor is embedded in the feature extraction layer of the preset mental state recognition algorithm, and the feature weight of the data matched with personalized strong correlation features in the algorithm is automatically increased. At the same time, the weight of the general features of the group that are not related to the employees is frozen, thereby obtaining a tuned mental state recognition algorithm.

[0038] In this embodiment, first, individual-specific and stable association patterns between emotions and work status are screened out from the associated data to provide accurate personalized basis for algorithm tuning. Specifically, the associated data of employees are first divided according to the scenario dimension, and the scenarios include but are not limited to project breakthrough period, quarterly performance appraisal period, major meeting preparation period and other situations with clear time nodes or work characteristics. In each scenario, negative indicators in the emotional fluctuation data (such as the proportion of negative words in text chat records, the pitch fluctuation rate during voice interaction, and the duration of negative emotions in micro-expression features) and the task delay rate in the work status data (the proportion of the actual completion time of tasks in a certain period that exceeds the planned time) are extracted respectively, and the correlation strength between the two is calculated by the Pearson correlation coefficient in statistics. When the calculation result meets the following conditions, it is marked as a personalized strong correlation feature: A Pearson correlation coefficient of ≥0.7 indicates a strong positive correlation between the two types of data (i.e., when negative indicators increase, the task delay rate also increases significantly). This association pattern has stably appeared ≥4 times in the past six months (180 days), and the scenario type is consistent each time (for example, it occurs "3 days before the project goes online"), eliminating sporadic associations and ensuring the stability and repeatability of the features.

[0039] For example, in the three days before the launch of four projects in the past six months, an employee's "proportion of negative words in the text increased from the daily 15% to more than 50%" and "the task delay rate increased from an average of 5% to 30%". The Pearson correlation coefficient between the two reached 0.78. In this case, this combined feature was marked as the employee's personalized strong correlation feature.

[0040] Furthermore, a correction factor is constructed based on the personalized strong correlation features. Through quantitative calculation, the personalized strong correlation features are converted into correction parameters that can be directly applied to the algorithm to achieve an accurate measurement of the feature's importance. The correction factor is constructed using the core attributes of the personalized strong correlation features as input and adopts a weighted calculation model.

[0041] Then, the correction factor is embedded in the feature extraction layer of the preset mental state recognition algorithm to obtain a tuned mental state recognition algorithm. By integrating the correction factor with the algorithm structure, personalized features are strengthened and irrelevant features are suppressed, thereby improving the algorithm's recognition accuracy of individual mental states. Specifically, the constructed correction factor is embedded in the feature extraction layer of the preset mental state recognition algorithm (the module responsible for extracting key features from raw data and assigning initial weights), forming a dynamic adjustment mechanism: 1. When the algorithm matches a marked, personalized, and strongly correlated feature within the associated data it processes (for example, if an employee is detected to be in the "three days before project launch" scenario and there are corresponding negative indicators and changes in the task delay rate), a correction factor is automatically applied to multiply the initial weight of the feature in the algorithm by the correction factor (for example, if the initial weight is 0.3 and the correction factor is 1.4, the adjusted weight is 0.42), significantly increasing its influence in feature fusion and model inference. 2. At the same time, the algorithm automatically freezes the weights of common group features unrelated to the employee (such as typical stress characteristics of employees in other positions and emotional characteristics unrelated to the employee's work environment). In other words, the weights of these features remain fixed and are not included in this optimization, preventing group data from interfering with individual-specific characteristics.

[0042] Through this process, the preset mental state recognition algorithm is transformed from a general model to a personalized tuning model, which can more sensitively capture characteristic signals that are strongly related to individual employees and reduce the interference of irrelevant information. The final tuned mental state recognition algorithm has an assessment accuracy rate of the employee's mental state that is ≥10% higher than before tuning.

[0043] In one embodiment, the calculation formula of the correction factor is: Correction factor = (individual characteristic value - group mean) / group standard deviation × 0.6 + characteristic occurrence frequency × 0.4.

[0044] In this embodiment, the calculation formula of the correction factor is used to quantify the core attributes of the personalized strong correlation feature into adjustment parameters that can be applied to the algorithm. The meaning and calculation logic of each part are as follows: (Individual trait value - group mean) / group standard deviation × 0.6: This section uses statistical methods to measure the degree of deviation of individual characteristics from the group benchmark. The core is to standardize individual trait values ​​into comparable deviation coefficients. Individual trait values ​​refer to the quantitative value of an employee's personalized, strongly correlated characteristics (such as the Pearson correlation coefficient between negative emotional indicators and task delay rates in a specific scenario); The group mean is the average value of the characteristic across employees in the same position under the same circumstances. The group standard deviation reflects the dispersion of the characteristic values ​​across employees in the same position and is used to eliminate the impact of group differences on the deviation calculation. The standardized deviation is calculated by dividing (individual - mean) by standard deviation, and then multiplied by a weighting factor of 0.6 to emphasize the impact of individual and group differences on the correction factor. For example, if an employee's characteristic value is 0.8, the group mean is 0.3, and the group standard deviation is 0.2, then the result is (0.8 - 0.3) / 0.2 × 0.6 = 1.5.

[0045] Feature frequency × 0.4: This component uses the frequency of a feature's occurrence to strengthen its stability weight. The frequency refers to the number of times the personalized, strongly correlated feature has consistently appeared in the past six months (must meet the prerequisite of ≥ 4 times). Multiplied by the weight coefficient of 0.4, it reflects the contribution of the feature's repeated stability to the correction factor. For example, if a feature appears 5 times, the result of this component is 5 × 0.4 = 2.0.

[0046] The sum of the results of the above two parts is the value of the correction factor. Through this formula, the degree of deviation between individuals and groups and the stability of the occurrence of characteristics are integrated into a single quantitative indicator, so that the correction factor can reflect both the individual specificity of the characteristics and the reliability of their repeated occurrence. It is ultimately used to adjust the weights of the corresponding features in the algorithm and improve the adaptability of the mental state recognition algorithm to individual differences.

[0047] In one embodiment, the method further comprises: A points incentive mechanism is established based on employees' frequency of service usage and interactive participation on the platform. Points can be exchanged for professional psychological assessment services or career planning courses, and points can be shared and experiences can be shared among employees through the social interaction module.

[0048] In one embodiment, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Classifying the sensitivity of the associated data; Determine a corresponding splitting rule based on the level corresponding to each associated data, split the corresponding associated data into 512-byte encrypted data blocks based on the splitting rule, and append a unique check code to each encrypted data block; The encrypted data blocks are stored in three physically isolated storage nodes, and the encrypted data blocks stored in any one storage node do not exceed 40% of the total; When calling the associated data, the calling conditions need to be met: The IP address of the calling node is in the preset whitelist; Obtain dynamic authorization tokens from three storage nodes, with token generation intervals within the preset time, and the tokens contain the version number of the data block reorganization algorithm; Through real-time facial feature verification of employees; After all three conditions are met, the reassembly algorithm is automatically triggered to splice the complete data according to the check code sequence.

[0049] In this embodiment, the associated data is first graded according to its sensitivity. This is done to differentiate the associated data based on the risk level that may be caused by data leakage, providing a classification basis for subsequent encrypted storage. Specifically, based on the privacy attributes of the data content and the enterprise data security specifications, the associated data is divided into three levels: Level 1 sensitive data: This includes voice emotion feature sequences in emotional fluctuation data, raw psychological assessment scores, and emergency psychological intervention conversation records. Leakage of this type of data could seriously damage employee privacy. Secondary sensitive data: This includes work status data such as project performance scores, overtime hours, and task delay rates, as well as historical behavioral data such as frequent access records to the mental health module. Leakage of these data could negatively impact an employee's professional image. Level 3 sensitive data: includes service access frequency in historical behavioral data, likes records of public social interactions, etc., which have low privacy attributes and relatively controllable leakage risks.

[0050] The grading process is automatically completed through a preset sensitive vocabulary library and rule engine. For example, text data containing keywords such as "depression" and "anxiety" are automatically classified as first-level sensitive data. It also supports manual review and adjustment to ensure the accuracy of the grading.

[0051] Then, based on the hierarchical splitting rules, the data is split into encrypted data blocks and a checksum is added. This step improves the security and integrity of data storage through a structured splitting and verification mechanism. The specific process is as follows: Determine splitting rules: Develop differentiated splitting strategies based on data sensitivity levels. Level 1 sensitive data is split by both time slicing and feature dimensions (e.g., every 30 seconds of voice data is a time slice, then split by feature dimensions such as fundamental frequency and amplitude). Level 2 sensitive data is split by record item (e.g., each task performance record is split independently). Level 3 sensitive data is split by batch (e.g., daily access records are aggregated and split). Splitting and encryption: Based on the above rules, sensitive data at all levels are uniformly split into encrypted data blocks of fixed size (512 bytes). During the splitting process, encryption is performed using the corresponding encryption algorithm (e.g., the national secret SM9 algorithm is used for level 1 data, and the AES-256 algorithm is used for level 2 data). Additional unique check code: A unique check code is generated for each encrypted data block. The check code includes the original serial number of the data block (identifying the position in the complete data), the timestamp (recording the split time), the sensitivity level label and the data block hash value, ensuring that the data block is not tampered with or replaced during storage and transmission.

[0052] Then, the encrypted data blocks are dispersed and stored in three physically isolated storage nodes. The distributed storage strategy reduces the risk of single point leakage and ensures the security of data physical storage. Specifically: Select three storage nodes with physical isolation properties, including computer rooms in different geographical locations (such as one each in North China, East China, and South China), servers of different network operators (such as one each in China Telecom, China Unicom, and China Mobile), and no direct data interaction links between nodes to avoid complete data leakage due to the compromise of a single node; use a dynamic allocation algorithm to distribute the encrypted data blocks to the three nodes, and strictly limit the storage ratio of a single node during the allocation process: the storage capacity of each node for level 1 sensitive data does not exceed 35% of the total, and the storage capacity of level 2 and level 3 sensitive data does not exceed 40%, to ensure that when any node is illegally accessed, the obtained data blocks cannot constitute complete information.

[0053] Finally, conditional verification and data reorganization when calling related data ensure the legitimacy of data calls through a multi-dimensional verification mechanism, while achieving secure reconstruction of complete data. The details are as follows: Verify that the calling node IP is in the preset whitelist: A preset list of legal node IPs that are allowed to call data (such as corporate intranet servers and authorized psychological counselor terminal IPs) is created. When a call request is initiated, the initiating node IP address is first checked to see if it is in the whitelist. If not, the request is directly rejected, blocking illegal access at the network layer. Verify the dynamic authorization tokens of the three storage nodes: After a legitimate call request is triggered, the three storage nodes each generate a dynamic authorization token. The token contains the version number of the data block reassembly algorithm (to ensure that the reassembly logic matches the splitting logic), the token generation timestamp, and the node signature. Verify that the token generation interval is within the preset time (e.g., ≤10 seconds to prevent the token from being intercepted and valid for a long time), and that the reassembly algorithm version numbers of the three tokens are consistent. Otherwise, the token is considered invalid. Verify employees' real-time facial features: When calling for personal sensitive data, the employee's real-time facial image is captured through a camera and compared with the pre-stored facial feature template (collected upon joining the company). A match degree of ≥95% is considered verified, ensuring that the data caller is the employee himself or someone authorized by him.

[0054] When all three of the above conditions are met, the reassembly algorithm corresponding to the version number in the token is automatically triggered. According to the original serial number and timestamp in the check code of the encrypted data block, the dispersedly stored data blocks are spliced ​​into the complete original data in chronological order. During the splicing process, the integrity of each data block is verified by the hash value in the check code, and finally the complete and untampered associated data is returned to the calling node.

[0055] In one embodiment, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Calculate the data activity index of the associated data using the following formula: (access frequency within a preset time period × 0.5) + (number of associated warning triggers × 0.3) + (data update frequency × 0.2); When the data activity index is greater than a first preset value, the data is encrypted and stored using the AES-256-GCM algorithm, with an additional quantum-resistant hash lock, and automatically re-encrypted every preset time. When the data activity index is less than a second preset value, the core data in the associated data is retained and the original detail data is deleted, and encrypted storage is performed using a lightweight encryption algorithm; When the data activity index is not greater than a first preset value and not less than a second preset value, a preset encryption algorithm is used for encrypted storage.

[0056] In this embodiment, the data activity index of the associated data is first calculated. This quantitative indicator is used to assess the dynamic usage frequency and importance of the associated data, providing a basis for subsequent encryption strategy adaptation. Specifically, the data activity index is calculated based on three core indicators of the associated data within a preset period (e.g., 30 days) using a weighted summation model. The formula is: Data activity index = (access frequency within the preset period × 0.5) + (number of associated warning triggers × 0.3) + (data update frequency × 0.2); The access frequency within the preset time period refers to the total number of times the associated data is accessed and viewed by authorized users (such as employees themselves or psychological counselors). The weight of 0.5 reflects the core impact of the frequency of data access on activity. The number of associated warning triggers refers to the number of times the data participates in the psychological status assessment and triggers the warning threshold. The weight of 0.3 reflects the close correlation between the data and the abnormal psychological status of the employee. Data update frequency refers to the number of times data is added or modified within a unit of time (such as every day) (such as new text emotion records in sentiment fluctuation data). The weight of 0.2 reflects the degree of dynamic change of the data.

[0057] The higher the index value calculated by this formula, the more real-time, important, and frequently used the data is.

[0058] Furthermore, when the data activity index is greater than a first preset value, encrypted storage is performed. When the data activity index is greater than the first preset value (e.g., the preset value is set to 8, representing high-activity data), a high-strength encryption strategy is adopted to address the security risks brought about by high-frequency access, specifically including: Encryption algorithm selection: AES-256-GCM algorithm is used for encrypted storage. This algorithm combines high encryption strength (256-bit key) with integrity verification function (generating authentication tags through GCM mode to prevent data tampering); Enhanced quantum computing resistance: A hash lock based on lattice cryptography is added to the encrypted data header. The cracking complexity of this hash lock increases exponentially with the improvement of quantum computing capabilities, thus preventing the risk of future quantum computing cracking traditional encryption algorithms. Automatic re-encryption mechanism: Set the re-encryption process to be automatically triggered every preset time (such as 12 hours). During re-encryption, a new random key (not related to the original key) is generated and the hash lock is updated simultaneously. This ensures that even if the key is leaked at a certain moment, it only affects the encrypted data during that period, significantly reducing long-term storage risks.

[0059] This type of high-activity data usually includes recent frequently accessed psychological counseling records, emotional fluctuation data that frequently triggers warnings, etc. High-intensity encryption strategies match their high security requirements.

[0060] Encrypted storage when the data activity index is less than a second preset value. When the data activity index is less than a second preset value (for example, the preset value is set to 3, indicating low-activity data), a strategy combining lightweight encryption and data reduction is adopted to optimize storage efficiency while ensuring basic security. Specifically, the strategy includes: Data streamlining: retain the core data in the associated data (such as the emotional trend curve in the emotional fluctuation data and the monthly task completion rate average in the work status data), delete the original detailed data (such as specific chat texts and hourly task progress records), and only retain the key features that support subsequent statistical analysis to reduce storage redundancy; Lightweight encryption algorithm: Use an encryption algorithm with low computational complexity (such as the national encryption SM3 hash algorithm combined with Base64 encoding) to reduce the system resource usage of the encryption and decryption process while meeting basic encryption requirements (preventing unauthorized access).

[0061] This type of low-activity data typically includes historical behavior records that have not been accessed for more than six months, early emotional data that has not triggered any warnings, etc. The streamlined and lightweight encryption strategy balances its security needs and storage costs.

[0062] Encrypted storage when the data activity index is between the first and second preset values. When the data activity index is not greater than the first preset value and not less than the second preset value (i.e., between 3 and 8, representing moderately active data), a preset conventional encryption algorithm is used for storage. This algorithm must meet the following requirements: Moderate encryption strength (such as AES-128 algorithm), balancing security and computational efficiency; Supports regular key rotation (e.g. weekly key updates), eliminating the need for frequent re-encryption. The complete structure of the data is retained (no streamlining is performed) to ensure that the data can be used for routine psychological status assessment and historical tracing.

[0063] This type of moderately active data typically includes employees' recent career planning consultation records, occasionally accessed work status data, etc. Conventional encryption strategies are adapted to their security needs and frequency of use to achieve a balance between security and efficiency.

[0064] In one embodiment, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Acquire multi-dimensional features of the associated data and classify them to obtain first dimensional features, second dimensional features, and third dimensional features; Obtain an encryption algorithm array; wherein the encryption algorithm array is an array including a plurality of encryption algorithms; Based on the first dimensional feature, the encryption algorithm array is deformed to obtain a deformed encryption algorithm array; Screening the array of deformed encryption algorithms based on the second dimensional feature to obtain a target encryption algorithm; The target encryption algorithm is tuned based on the third dimension feature to obtain a tuned encryption algorithm; and the associated data is encrypted and stored based on the tuned encryption algorithm.

[0065] In this embodiment, first, the multi-dimensional features of the associated data are obtained and classified. Through multi-dimensional feature extraction and classification, a refined basis is provided for the adaptation of the encryption algorithm. Specifically, three types of core dimensional features are extracted from the associated data: The first dimension reflects the sensitivity of the data, including the privacy level of the data content (such as whether it contains psychological diagnosis results) and the potential impact of data leakage (such as the potential risk to employees' professional reputation). It uses a quantitative score of 1-10 (10 is the most sensitive). The second dimension reflects the structural characteristics of the data, including data format (text, voice, image), data volume (such as the number of text characters, voice duration), and data update frequency (static data vs. dynamic real-time update data). For example, the structural characteristics of a "5-minute voice emotion recording" can be described as "audio format + 300KB + daily updates"; The third dimension features: characterizes the usage scenarios of the data, including the access subject (employee / administrator / psychological counselor), access frequency (such as daily access / monthly access), and access purpose (daily review / early warning analysis / emergency intervention). For example, "historical emotional data used by psychological counselors to intervene in psychological crises" belongs to a high-urgency usage scenario.

[0066] The classification process is completed through automatic matching of feature labels. For example, text data containing negative-leaning keywords has a first-dimensional feature score of 10 points, a second-dimensional feature of "text + 500 words + real-time update", and a third-dimensional feature of "psychologist + emergency visit".

[0067] Then, we obtain the encryption algorithm array. This step builds a candidate pool of diverse encryption algorithms, providing a basis for subsequent dynamic adaptation. The encryption algorithm array is a collection of multiple encryption algorithms, covering algorithm types with different encryption strengths, applicable scenarios, and computational complexities, for example: High-security algorithms: National Secret SM4, AES-256, RSA-2048, etc., suitable for highly sensitive data; Lightweight algorithms: DES, RC4, SM3 hash algorithm, etc., suitable for low-sensitivity or large-volume data; Special scenario algorithms: voiceprint encryption algorithm for voice data, homomorphic encryption algorithm for text data, etc.

[0068] The algorithm array is sorted in a two-dimensional matrix to facilitate the subsequent rapid location of candidate algorithms based on data characteristics.

[0069] Then, based on the first dimension feature deformation encryption algorithm array, this step adjusts the priority of the algorithm array according to the sensitive attributes of the data to achieve a preliminary match between encryption strength and sensitivity level. Specifically: For highly sensitive data with a first-dimension feature score ≥8, the priority of high-security algorithms (such as AES-256 and SM4) in the array is increased by 30% through weight adjustment, while the candidate weight of lightweight algorithms is reduced; for low-sensitivity data with a first-dimension feature score ≤3, the weight is adjusted in the opposite direction to increase the priority of lightweight algorithms (such as RC4) and reduce the occupation of system resources by high-intensity algorithms; the deformation process is achieved through dynamic sorting of the algorithm array. In the deformed array, the algorithm that matches the data sensitivity level will be placed in the front. For example, in the deformed array corresponding to highly sensitive data, the AES-256 algorithm will be ranked first.

[0070] The target encryption algorithm is selected based on the second dimension characteristics. This step combines the structural characteristics of the data to select the single algorithm with the best adaptability from the transformed algorithm array. The specific selection logic includes: For text data, give priority to algorithms that support variable-length data block encryption (such as AES-256-CBC mode) to avoid padding redundancy caused by fixed block lengths. For multimedia data such as voice and images, select an algorithm that supports streaming encryption (such as RC4) to improve encryption efficiency; For large amounts of data (such as historical behavior records exceeding 1GB), exclude algorithms with extremely high computational complexity (such as RSA-4096) and choose algorithms that balance efficiency and security (such as AES-128).

[0071] For example, for data with a first-dimension feature score of 9 and a second-dimension feature of "text + 1,000 words + daily updates", the AES-256-GCM algorithm with "support for variable-length text + medium efficiency + high strength" is selected from the transformation array as the target encryption algorithm.

[0072] Finally, the target encryption algorithm is tuned and encrypted storage is performed based on the third-dimensional characteristics. The algorithm parameters are optimized according to the data usage scenario to ultimately achieve precise encrypted storage.

[0073] In one embodiment, the tuning content includes: Key management optimization: For emergency access scenarios (e.g., where the third dimension feature is "emergency intervention"), the key validity period is shortened from the default 7 days to 24 hours, and key distribution is subject to multiple approvals. Encryption mode adjustment: For frequently accessed data (such as daily work status data), a hybrid "encryption + caching" mode is adopted. After the initial encryption, a temporary decryption cache (valid for one hour) is retained to reduce the computational overhead of repeated encryption. Additional security mechanism: For administrator access scenarios, an access log encryption function is added based on the target algorithm to record the time, IP address and other information of the access behavior and encrypt and store them to ensure that the operation is traceable.

[0074] The tuned encryption algorithm is used to encrypt associated data. The encryption results are stored in association with the data's unique identifier (such as employee ID + timestamp). At the same time, an encryption parameter log (including algorithm type, key version, and tuning parameters) is generated to provide a basis for subsequent decryption and security audits.

[0075] In one embodiment, when the encrypted data blocks are dispersedly stored in three physically isolated storage nodes, the three physically isolated storage nodes form a dynamic topology network, the connection relationship between the storage nodes is randomly adjusted at preset intervals, and the node IP addresses are regularly updated using dynamic camouflage technology; A topological location tag is attached to each encrypted data block when it is stored, recording its logical location in the current topology structure. The topological location tag is automatically updated as the topology network is adjusted; When calling data, you must first parse the current node connection relationship through the preset topology key, and then locate the data block according to the topological location label.

[0076] In this example, three storage nodes are first deployed in different data centers with a geographical distance of ≥500 km. Each node is equipped with independent power supply, network link, and security protection system to ensure that a failure or attack at a single physical location will not affect other nodes. A quantum random number generator (QRNG) is used to generate a node connection matrix every 5 minutes to define the logical connection relationship between nodes. The connection matrix contains the following parameters: Master node (responsible for receiving external data access requests); Data flow path (e.g., node A → node C → node B); Backup link priority (automatically switches when the primary link delay is greater than 100ms); Each time a new topology is generated, the node connection probability is adjusted through the small-world network model (Watts-Strogatz model) in graph theory, so that the network maintains both local clustering characteristics and global connectivity, and the average path length is controlled between 2-3 hops.

[0077] Dynamic IP masquerading mechanisms include: Quantum Key Distribution (QKD): Session keys are regularly exchanged between storage nodes through quantum entanglement channels, with each key update cycle lasting 90 seconds, ensuring unconditional security of communication content. IP dynamic change protocol: Each node is configured with a pool of 1024 pre-generated virtual IP addresses, which change every 30 seconds according to the following rules: a pseudo-random number is generated using the Mersenne Twister algorithm as the IP index; the new IP address must have a Hamming distance of ≥ 16 bits from the previous IP address; during the change process, identity continuity is proven to other nodes using zero-knowledge proof (ZKP); Traffic obfuscation strategy: During the IP change period, fake traffic packets (accounting for 30% of the total traffic) are continuously sent to unrelated IP addresses. The size, frequency, and protocol type of the fake packets mimic real data access patterns, interfering with external traffic analysis.

[0078] Topological location labeling system: Multi-dimensional coordinate encoding: The topological location label of each encrypted data block contains four-dimensional coordinates: Timestamp (accurate to nanoseconds, recording the time when the data block was generated); Node ID (the quantum fingerprint hash value of the current storage node); Link vector (records the shortest path from the main node to this node); Version number (incremented each time the topology is adjusted to prevent old version tags from being misused).

[0079] Dynamic label update: When the topology is adjusted, each node synchronously updates the data block label through the distributed consensus algorithm (PBFT): The master node broadcasts the new topology and timestamp; Each node calculates a new link vector for the locally stored data block; The legitimacy of the update operation is verified through elliptic curve digital signature (ECDSA).

[0080] Tag encryption storage: Topology location tags are encrypted using the same dynamic encryption algorithm as data blocks. The key contains the hash value of the current topology structure to ensure the binding integrity of the tag and the topology.

[0081] Data calling and positioning process: Topological key negotiation: When a client initiates a data access request, it must first obtain the current topological key through the quantum key distribution protocol. The key contains: Merkle tree root hash of the latest topology; Public key certificate chain of each node; Valid time window (accurate to microseconds).

[0082] Location tag parsing: The client uses the topology key to decrypt the topology location tag of the data block and verifies its validity through the following steps: Check whether the timestamp is within the validity period of the current topology; Verify whether the link vector conforms to the current topology connection relationship; Compare the version number with the latest topology version to see if it is consistent; Multi-path data retrieval: Based on the parsed link vectors, the client sends parallel retrieval requests to multiple possible nodes simultaneously, using the following strategies to optimize response speed: Prioritize the path with the lowest network latency; When a path response times out (>200ms), it automatically switches to an alternative path; Perform integrity verification on the returned data block (such as SHA-3 hash comparison); Through the above mechanism, the secure storage and efficient retrieval of encrypted data blocks in dynamic topology networks are achieved, making the data storage location uncertain in the time dimension (topology changes every 5 minutes) and obfuscated in the spatial dimension (dynamic camouflage of IP addresses), significantly improving the data's anti-attack capabilities and privacy protection level.

[0083] In one embodiment, when the encrypted data blocks are dispersedly stored in three physically isolated storage nodes, a unique virtual storage address is assigned to each encrypted data block. There is no fixed mapping relationship between the virtual storage address and the physical storage node, and the virtual storage address is updated once every preset time based on a random number generator. The update record is only retained in the encrypted block of the blockchain. Storage nodes are transmitted through obfuscated fake data blocks, of which 30% are fake data blocks. They have the same format as real data blocks but fake content. When an abnormal access attempt is detected, the fake data blocks are automatically returned to the attacker, and the real data blocks are silently migrated to the backup storage node.

[0084] In this embodiment, to dynamically decouple encrypted data blocks from physical storage nodes, each encrypted data block is assigned a unique virtual storage address, which has no fixed association with a physical node. Virtual addresses are generated by integrating multi-dimensional random sources, including quantum physics noise, precise data generation timestamps, and device hardware characteristics. These addresses are expanded into unpredictable address sequences using a chaotic algorithm, ensuring that the addresses themselves do not carry any physical location information.

[0085] Virtual addresses are automatically updated at preset intervals (e.g., 90 seconds). The update process is driven by a random number generator, and the new address after each update has no logical connection to the previous address. All address change records are encrypted and stored only in encrypted blocks on the blockchain. Blockchain blocks are maintained using a distributed consensus mechanism. Only authorized nodes can verify the validity of updates using a specific key, and historical records cannot be tampered with or forged.

[0086] The above design prevents external attackers from inferring the physical storage location of data through virtual addresses. Even if they obtain address information at a certain moment, it will become invalid due to the regular update of the address, fundamentally severing the static association between the virtual identifier and the physical location.

[0087] When data is transmitted between storage nodes, a fixed ratio (30%) of fake data blocks is mixed in. These fake data blocks use the same formatting specifications as real data blocks, including header identifiers, encryption methods, and checksum structures. Their content only contains invalid information that mimics the distribution characteristics of real data. The fake data blocks are generated based on a training model based on the statistical characteristics of real data, ensuring that they are indistinguishable from real data blocks in terms of format, length, and encryption strength, thereby confusing attackers from identifying valid information.

[0088] When abnormal access attempts are detected (such as continuous illegal signature requests, access frequency or source that does not conform to normal behavior patterns), the defense mechanism will be automatically triggered: carefully constructed false data blocks are returned to the request source. These false data blocks are not only complete in format, but also contain seemingly reasonable but actually invalid content to mislead the attacker's analysis direction.

[0089] At the same time, a silent migration process for the real data blocks is initiated. This migration is conducted over an encrypted channel to a pre-deployed backup storage node. Upon completion, the original node's real data is completely erased, leaving only the fake data blocks. This entire migration process is completed without user awareness, preventing attackers from accessing valid data while ensuring continued access for authorized users.

[0090] When accessing data, the connection relationship between the current storage nodes must first be resolved using the preset topology key. The topology key contains the node association rules of the current network and is generated and distributed encrypted regularly to ensure that only authorized entities can obtain valid connection information between nodes.

[0091] After determining the node connectivity, the specific data block is located based on the topological location tag attached to the encrypted data block during storage. The topological location tag records the logical location of the data block within the current dynamic network topology. This tag automatically updates as the storage node connectivity adjusts, ensuring consistency with the actual storage state of the data block. By combining the topological key with the topological location tag, authorized users can accurately trace the current storage location of the data block, enabling legitimate access and reassembly of the data. However, attackers, unable to access real-time topological information, would be unable to locate the entire data block even if they intercepted a portion of it.

[0092] In one embodiment, the associated data is encrypted and stored using a dynamic encryption algorithm, including: Dynamically updating the encryption algorithm based on the associated data; Generate a first ternary tree structure based on the updated algorithm parameters of the encryption algorithm; generating a second ternary tree structure based on the data features of the associated data; Superimposing the first ternary tree structure and the second ternary tree structure, and sequentially comparing the overlapping nodes; and filtering out a target node based on the comparison results of the nodes according to a predetermined rule; An encryption key is generated based on the character data on each target node; and the associated data is encrypted and stored based on the encryption key.

[0093] In this embodiment, the core parameters of the underlying encryption algorithm are dynamically adjusted by analyzing changes in the characteristics of the associated data in real time, achieving real-time adaptation of the encryption logic to the data characteristics. Specifically, the system continuously monitors fluctuations in the associated data volume (e.g., the number of new records added per day), changes in feature dimensions (e.g., new voice feature parameters in sentiment data), and sensitivity level shifts (e.g., from level 2 to level 1), and quantifies these changes as algorithm adjustment factors.

[0094] For example, if the proportion of highly sensitive fields (such as raw scores on psychological assessments) in the associated data increases by more than 20%, the key length of the basic encryption algorithm will be automatically expanded from 128 bits to 256 bits, and a more complex block cipher mode (such as switching from CBC mode to GCM mode) will be enabled. If the data update frequency drops to less than 10 times per day, the number of round function iterations of the algorithm will be appropriately simplified to optimize encryption efficiency while ensuring security. The algorithm update process is synchronized to all storage nodes via an encrypted channel, and the update records are written to an unalterable audit log to ensure the traceability of algorithm iterations.

[0095] The first ternary tree structure is a structured mapping of the core parameters of the updated encryption algorithm, used to convert abstract algorithm features into computable tree nodes. Key parameters of the updated encryption algorithm, including key length, number of iterations, block size, and hash function type, are extracted. Each parameter is converted into a character sequence according to pre-set rules (e.g., "256-bit key" is converted to "K256" and "16 iterations" is converted to "I16").

[0096] A ternary tree is then constructed based on these character sequences: the root node is the algorithm's unique identifier (e.g., the algorithm name + the hash value of the update timestamp). Each parent node derives three child nodes according to the logic of "parameter type - parameter value - checksum." The depth of the child nodes is positively correlated with the importance of the parameter (e.g., the key length parameter corresponds to a node with a depth of 3, and the hash function type corresponds to a node with a depth of 2). Each node in the tree structure is appended with a SHA-256 checksum to ensure that the parameter conversion and tree structure generation process have not been tampered with.

[0097] The second ternary tree structure is used to capture the characteristic attributes of the linked data itself, achieving a structured association between data characteristics and encryption algorithms. Multi-dimensional features are extracted from the linked data, including data format distribution (such as the proportion of text, voice, and numerical data), timestamp sequence (such as the distribution of data generation time in the past seven days), and the frequency of sensitive fields (such as the number of occurrences of keywords such as "anxiety" and "stress"). These features are quantified into standardized character labels (such as "60% text" is converted to "T60" and "50 records generated per day" is converted to "D50").

[0098] Based on these labels, a ternary tree is constructed: the root node uniquely identifies the batch of associated data. Each parent node derives three child nodes according to the logic of "feature category - feature value - feature weight" (for example, the child nodes of the "sensitive feature" category include "negative emotion word frequency" and "number of psychological interventions"). The node level corresponds to the degree of impact of the feature on data security (for example, features directly related to privacy are assigned to deeper nodes). Similar to the first ternary tree, each node includes a checksum to ensure the accuracy of data feature mapping.

[0099] The first and second ternary trees are superimposed by aligning their root nodes, so that the nodes of the two trees form overlapping or independent node sets in the same spatial coordinate system. During the superposition process, the character labels of the nodes are compared with the checksum to identify overlapping nodes in the two trees and unique non-overlapping nodes in each tree.

[0100] Overlapping nodes are then evaluated according to pre-set filtering rules: If the semantic correlation between the character labels of two overlapping nodes in the tree is ≥80% (for example, "K256" and "T60" are strongly correlated because they both associate with highly sensitive data), and the node checksum matches ≥95%, the nodes are marked as candidate nodes. The candidate nodes are further screened by depth, retaining nodes with a depth of ≥3 (to ensure the core characteristics of the correlation), and finally determining the target node. The filtering rules are solidified through smart contracts, allowing for dynamic adjustment of the correlation threshold based on the data security level (for example, raising the threshold for Level 1 sensitive data to 90%).

[0101] The character data on the target node contains a fusion of encryption algorithm parameters and data characteristics. This character data is serialized and concatenated from deep to shallow node levels (for example, characters from nodes with a depth of 5 are concatenated first) to form the original key material. Multiple rounds of hash operations are performed on this material to generate an encryption key of a fixed length (for example, 256 bits). The key generation process uses quantum random numbers as salts to ensure that the same character sequence generates different keys.

[0102] Encryption keys are distributed through a key management system and are only briefly decrypted during encryption operations. They are destroyed immediately after use to prevent the keys from lingering in memory. Before storage, the associated data is encrypted using this encryption key. The encrypted ciphertext is stored across distributed nodes according to a data sharding strategy. Each shard's header is appended with a key index and encryption timestamp to ensure accurate identification of the corresponding key version during decryption.

[0103] Reference Figure 2 In another embodiment of the present invention, a data processing device for an integrated virtual employee care SaaS platform is provided, comprising: An acquisition unit is used to acquire employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; a tuning unit, configured to tune a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; An evaluation unit, configured to evaluate the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically push care plan information when the employee mental state evaluation result triggers a warning threshold; The storage unit is used to encrypt and store the associated data using a dynamic encryption algorithm.

[0104] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0105] Reference Figure 3In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0106] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0107] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0108] In summary, the data processing method and device of the integrated virtual employee care SaaS platform provided by the present invention in the embodiment of the present invention include: obtaining employee related data through the integrated virtual employee care SaaS platform; the related data include historical behavior data, work status data and emotional fluctuation data; tuning the preset psychological state recognition algorithm based on the related data to obtain a tuned psychological state recognition algorithm; evaluating the related data based on the tuned psychological state recognition algorithm to generate an employee psychological state evaluation result, and automatically pushing care plan information when the employee psychological state evaluation result triggers an early warning threshold; and encrypting and storing the related data using a dynamic encryption algorithm. In the present invention, related data of multiple dimensions of employees are obtained through the integrated virtual employee care SaaS platform, and the psychological state recognition algorithm is tuned, thereby improving the accuracy of psychological state evaluation; and the related data are encrypted and stored using a dynamic encryption algorithm to ensure data security.

[0109] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0110] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0111] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A data processing method for an integrated virtual employee care SaaS platform, characterized in that: The following steps are involved: Obtain employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; Tuning a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; Evaluate the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically push care plan information when the employee mental state evaluation result triggers a warning threshold; The associated data is encrypted and stored using a dynamic encryption algorithm.

2. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The integrated virtual employee care SaaS platform integrates mental health support, career planning advice, social interaction services and life guidance functional modules into the same SaaS platform. The SaaS platform receives employee service requests through a unified interface and automatically matches the corresponding functional modules according to the request type.

3. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The emotional fluctuation data includes the emotional tendency value of the text chat record and the intonation feature parameters of the voice interaction.

4. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: Tuning a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm includes: Extracting personalized strong correlation features from the correlation data; where, for the same employee in the same scenario, the Pearson correlation coefficient between the negative indicator in the emotional fluctuation data and the task delay rate in the work status data is ≥0.7, and it has occurred stably ≥4 times in the past six months, this is marked as a personalized strong correlation feature; Constructing a correction factor based on the personalized strong correlation feature; The correction factor is embedded in the feature extraction layer of the preset mental state recognition algorithm, and the feature weight of the data matched with personalized strong correlation features in the algorithm is automatically increased. At the same time, the weight of the general features of the group that are not related to the employees is frozen, thereby obtaining a tuned mental state recognition algorithm.

5. The data processing method of the integrated virtual employee care SaaS platform according to claim 4 is characterized in that: The calculation formula of the correction factor is: Correction factor = (individual characteristic value - group mean) / group standard deviation × 0.6 + characteristic occurrence frequency × 0.

4.

6. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The method further comprises: A points incentive mechanism is established based on employees' frequency of service usage and interactive participation on the platform. Points can be exchanged for professional psychological assessment services or career planning courses, and points can be shared and experiences can be shared among employees through the social interaction module.

7. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The associated data is encrypted and stored using a dynamic encryption algorithm, including: Classifying the sensitivity of the associated data; Determine a corresponding splitting rule based on the level corresponding to each associated data, split the corresponding associated data into 512-byte encrypted data blocks based on the splitting rule, and append a unique check code to each encrypted data block; The encrypted data blocks are stored in three physically isolated storage nodes, and the encrypted data blocks stored in any one storage node do not exceed 40% of the total; When calling the associated data, the calling conditions need to be met: The IP address of the calling node is in the preset whitelist; Obtain dynamic authorization tokens from three storage nodes, with token generation intervals within the preset time, and the tokens contain the version number of the data block reorganization algorithm; Through real-time facial feature verification of employees; After all three conditions are met, the reassembly algorithm is automatically triggered to splice the complete data according to the check code sequence.

8. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The associated data is encrypted and stored using a dynamic encryption algorithm, including: Calculate the data activity index of the associated data using the following formula: (access frequency within a preset time period × 0.5) + (number of associated warning triggers × 0.3) + (data update frequency × 0.2); When the data activity index is greater than a first preset value, the data is encrypted and stored using the AES-256-GCM algorithm, with an additional quantum-resistant hash lock, and automatically re-encrypted every preset time. When the data activity index is less than a second preset value, the core data in the associated data is retained and the original detail data is deleted, and encrypted storage is performed using a lightweight encryption algorithm; the first preset value is greater than the second preset value; When the data activity index is not greater than a first preset value and not less than a second preset value, a preset encryption algorithm is used for encrypted storage.

9. The data processing method of the integrated virtual employee care SaaS platform according to claim 1 is characterized in that: The associated data is encrypted and stored using a dynamic encryption algorithm, including: Acquire multi-dimensional features of the associated data and classify them to obtain first dimensional features, second dimensional features, and third dimensional features; Obtain an encryption algorithm array; wherein the encryption algorithm array is an array including a plurality of encryption algorithms; Based on the first dimensional feature, the encryption algorithm array is deformed to obtain a deformed encryption algorithm array; Screening the array of deformed encryption algorithms based on the second dimensional feature to obtain a target encryption algorithm; The target encryption algorithm is tuned based on the third dimension feature to obtain a tuned encryption algorithm; and the associated data is encrypted and stored based on the tuned encryption algorithm.

10. A data processing device for an integrated virtual employee care SaaS platform, characterized in that: include: An acquisition unit is used to acquire employee-related data through the integrated virtual employee care SaaS platform; the related data includes historical behavior data, work status data, and emotional fluctuation data; a tuning unit, configured to tune a preset mental state recognition algorithm based on the associated data to obtain a tuned mental state recognition algorithm; An evaluation unit, configured to evaluate the associated data based on the tuned mental state recognition algorithm to generate an employee mental state evaluation result, and automatically push care plan information when the employee mental state evaluation result triggers a warning threshold; The storage unit is used to encrypt and store the associated data using a dynamic encryption algorithm.

Images