Method for evaluating dynamic emotional state based on multi-dimensional data analysis of employees

Abstract

This invention provides a dynamic emotional state assessment method based on multidimensional employee data analysis, comprising: collecting multidimensional employee data, including basic physiological data, work behavior data, and environmental correlation data; performing timestamp synchronization processing on the multidimensional data; extracting the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and performing weighted fusion to obtain an initial emotional feature vector; acquiring an employee historical emotional baseline database; wherein, the employee historical emotional baseline database contains the emotional feature benchmark values ​​of employees in different work scenarios; calculating the difference between the initial emotional feature vector and the historical emotional feature benchmark values ​​in the corresponding work scenarios, and combining it with a preset emotional influence weight matrix to output a real-time emotional state assessment result. This invention overcomes the current shortcomings of accurately and comprehensively assessing employee emotions.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for assessing dynamic emotional states based on multidimensional data analysis of employees. Background Technology

[0002] In modern enterprise management, employee emotional state directly impacts work efficiency, teamwork quality, and overall corporate performance. Therefore, accurate assessment and dynamic monitoring of employee emotions have become a core requirement of corporate human resource management. However, current employee emotional state assessment technologies still face numerous bottlenecks, making it difficult to meet the practical needs of refined enterprise management.

[0003] Traditional methods of employee mood assessment often rely on manual interviews, questionnaires, or subjective observation. These methods are not only time-consuming and labor-intensive with limited coverage, but also susceptible to strong subjective bias. For example, employees may conceal their true emotions due to privacy concerns or workplace relationships, leading to significant discrepancies between the assessment results and their actual emotional state. Furthermore, manual assessments are often conducted periodically (e.g., monthly or quarterly), making real-time dynamic monitoring impossible and failing to capture short-term fluctuations and sudden changes in employee emotions, thus missing the optimal opportunity for timely intervention.

[0004] With the development of data collection technology, some solutions have begun to attempt emotion assessment based on single-dimensional data (such as only physiological data or only work behavior data). For example, smart bracelets can be used to collect physiological data such as employee heart rate and steps, or office systems can be used to record work behavior data such as task completion time and email sending frequency, and then emotional state can be judged based on preset thresholds. However, such single-dimensional assessment solutions have significant drawbacks: on the one hand, employee emotions are the result of the combined effects of multiple factors such as physiology, behavior, and environment. Relying solely on single-dimensional data cannot fully reflect the complex causes of emotions and is prone to misjudgment (such as misjudging "abnormal behavior frequency caused by a surge in workload" as "efficiency decline caused by negative emotions").

[0005] In summary, current methods for assessing employee emotional state suffer from problems such as strong subjectivity, poor real-time performance, and limited dimensions, making it difficult to achieve accurate, dynamic, and comprehensive assessment of employee emotions. Summary of the Invention

[0006] The main objective of this invention is to provide a dynamic emotional state assessment method based on multidimensional data analysis of employees, aiming to overcome the current shortcomings of accurately and comprehensively assessing employee emotions.

[0007] To achieve the above objectives, this invention provides a dynamic emotional state assessment method based on multidimensional data analysis of employees, comprising the following steps:

[0008] Collect multidimensional data from employees, including basic physiological data, work behavior data, and environmental data, and perform timestamp synchronization processing on the multidimensional data;

[0009] The fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data are extracted and weighted and fused to obtain the initial emotion feature vector.

[0010] Obtain the employee historical emotion baseline database; the employee historical emotion baseline database contains the baseline values ​​of employees' emotional characteristics in different work scenarios;

[0011] The difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario is calculated, and combined with the preset emotion influence weight matrix, the real-time emotion state assessment result is output.

[0012] Furthermore, basic physiological data includes heart rate data and skin conductance data; work behavior data includes task interaction behavior and document editing pause duration; and environmental correlation data includes workstation lighting intensity and office area sound pressure level.

[0013] Furthermore, after outputting the real-time emotional state assessment results, it includes:

[0014] When an employee's real-time emotional state assessment result is negative over a continuous period of time, an analysis of the sources of emotional influencing factors is triggered. By calculating the Pearson correlation coefficient between the data of each dimension and the real-time emotional state assessment result, the core influencing factors are identified and dynamic adjustment suggestions are generated.

[0015] Furthermore, the fluctuation range of the basic physiological data is extracted, including:

[0016] For heart rate data in basic physiological data, the data is divided into segments by a preset time interval. The difference between the maximum and minimum heart rate values ​​in each data segment is calculated, and the average of the differences of multiple consecutive data segments is used as the heart rate fluctuation amplitude.

[0017] For the electrodermal activity data, the number of times the peak value of the electrodermal signal occurs per unit time is calculated, and the absolute value of the difference between the number of times and the preset normal range is taken as the fluctuation amplitude of the electrodermal activity.

[0018] The weighted average of the heart rate fluctuation amplitude and the skin electrical activity fluctuation amplitude is calculated as the fluctuation amplitude of the baseline physiological data.

[0019] Furthermore, the extraction of frequency anomalies in work behavior data includes:

[0020] The frequency of task interaction behavior in statistical work behavior data is used to establish a normal distribution model of the frequency of task interaction of employees in the same working period within the past 30 days. The deviation between the current task interaction frequency and the model mean is calculated, and then the deviation is divided by the model standard deviation to obtain the standardized deviation.

[0021] If the absolute value of the standardized deviation is greater than the threshold, then the absolute value of the standardized deviation is used as the anomaly of the task interaction frequency.

[0022] The frequency anomaly of document editing pause duration is calculated, and the frequency anomalies of task interaction and document editing pause duration are weighted and fused to obtain the frequency anomaly of work behavior data.

[0023] Furthermore, the extraction of interference coefficients from environmental correlation data includes:

[0024] The workstation lighting intensity data in the environmental correlation data is compared with the preset comfortable lighting intensity range to calculate the lighting intensity interference coefficient.

[0025] For the sound pressure level data in the office area, the sound pressure level interference coefficient is obtained by using a preset decibel as the benchmark value and combining it with calculation.

[0026] The interference coefficient of the environmental correlation data is obtained by weighting and fusing the light intensity interference coefficient and the sound pressure level interference coefficient.

[0027] Furthermore, the difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario is calculated, and combined with the preset emotion influence weight matrix, the real-time emotion state assessment result is output, including:

[0028] Retrieve historical emotional characteristic baseline values ​​from the employee historical emotional baseline database that are consistent with the current work scenario and the same time period; among them, the historical emotional characteristic baseline values ​​include historical physiological fluctuation amplitude baseline, historical behavioral abnormality baseline, and historical environmental interference coefficient baseline;

[0029] Physiological deviation rate is calculated based on the fluctuation range of basic physiological data and the historical physiological fluctuation range benchmark; behavioral deviation rate is calculated based on the frequency abnormality of work behavior data and the historical behavioral abnormality benchmark; environmental deviation rate is calculated based on the interference coefficient of environmental correlation data and the historical environmental interference coefficient benchmark.

[0030] A preset emotion influence weight matrix is ​​invoked, which includes basic weights and scene correction coefficients; the basic weights are corrected based on the scene correction coefficients corresponding to the current work scene to obtain the corrected weights;

[0031] After adjusting the weights, the physiological deviation rate, behavioral deviation rate, and environmental deviation rate are weighted and calculated to obtain the weighted total deviation score.

[0032] By comparing the weighted total deviation score with the emotion level threshold, a real-time emotion state assessment result is obtained, which includes emotion level, sub-labels, deviation rate of each dimension, and core influence dimension.

[0033] Furthermore, before outputting the real-time emotional state assessment results, based on a pre-defined emotional influence weight matrix, the following steps are included:

[0034] Using the fluctuation range of basic physiological data, the frequency abnormality of work behavior data, and the interference coefficient of environmental correlation data as variables, the correlation coefficient between any two variables is calculated to form a correlation matrix.

[0035] If the correlation coefficient between any two groups of variables reaches the threshold, they are determined to be strongly correlated dimension groups. The weights of the corresponding two groups of variables are increased by a preset proportion based on the original emotion influence weight matrix, while the weights of other variables are reduced. If the correlation coefficients between all variables are less than the threshold, the original emotion influence weight matrix remains unchanged.

[0036] In the output of the real-time emotion state assessment results, strongly correlated dimension groups and their corresponding correlation coefficients are labeled.

[0037] This invention also provides a dynamic emotional state assessment device based on multidimensional data analysis of employees, comprising:

[0038] The data acquisition unit is used to collect multidimensional data of employees, including basic physiological data, work behavior data and environmental correlation data, and to perform timestamp synchronization processing on the multidimensional data.

[0039] The extraction unit is used to extract the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and perform weighted fusion to obtain an initial emotion feature vector;

[0040] The acquisition unit is used to acquire the employee historical emotion baseline database; wherein, the employee historical emotion baseline database contains the emotional characteristic benchmark values ​​of employees in different work scenarios;

[0041] The output unit is used to calculate the difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario, and output the real-time emotion state assessment result by combining the preset emotion influence weight matrix.

[0042] This invention provides a dynamic emotional state assessment method based on multidimensional employee data analysis, comprising: collecting multidimensional employee data, including basic physiological data, work behavior data, and environmental correlation data; performing timestamp synchronization processing on the multidimensional data; extracting the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and performing weighted fusion to obtain an initial emotional feature vector; acquiring an employee historical emotional baseline database; wherein, the employee historical emotional baseline database contains the emotional feature benchmark values ​​of employees in different work scenarios; calculating the difference between the initial emotional feature vector and the historical emotional feature benchmark values ​​in the corresponding work scenarios, and combining it with a preset emotional influence weight matrix to output a real-time emotional state assessment result. In this invention, multidimensional employee data is collected, and the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data are extracted respectively, and weighted fusion analysis is performed. Then, combined with the emotional influence weight matrix, a real-time emotional state assessment result is output. Through multidimensional data fusion analysis, the current shortcomings in accurately and comprehensively assessing employee emotions are overcome. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the steps of a dynamic emotional state assessment method based on multidimensional data analysis of employees in one embodiment of the present invention;

[0044] Figure 2 This is a structural block diagram of a dynamic emotional state assessment device based on multidimensional data analysis of employees, according to an embodiment of the present invention.

[0045] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0046] The implementation, functional features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative and not intended to limit the invention.

[0048] Reference Figure 1 One embodiment of the present invention provides a dynamic emotional state assessment method based on multidimensional data analysis of employees, including the following steps:

[0049] Step S1: Collect multidimensional data of employees, including basic physiological data, work behavior data and environmental correlation data, and perform timestamp synchronization processing on the multidimensional data;

[0050] Step S2: Extract the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and perform weighted fusion to obtain the initial emotion feature vector;

[0051] Step S3: Obtain the employee historical emotion baseline database; wherein, the employee historical emotion baseline database contains the baseline values ​​of the emotional characteristics of employees in different work scenarios.

[0052] Step S4: Calculate the difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario, and combine it with the preset emotion influence weight matrix to output the real-time emotion state assessment result.

[0053] It is particularly important to note that all technical steps, algorithm applications, and parameter settings in the technical solution of this application have clear technical objectives and application value. They do not utilize complex steps and algorithmic formulas to achieve simple functions. To provide detailed explanations of each step and avoid ambiguity, some conventional algorithms are used for illustration. However, this does not mean that the algorithms and technical features listed herein are the only way to implement the technical solution of this application, nor is it intended to limit the scope of protection of this application. This application is not a combination or stacking of the listed algorithms and technical features; its essence is to exemplify the implementation methods of this application to fully explain it. It does not pursue formal complexity by adding meaningless technical steps, nor does it involve the accumulation of technologies divorced from practical needs; it conforms to the conventional logic of technical improvement and design.

[0054] In this embodiment, as in step S1 above, the core objective is to obtain basic data that can comprehensively reflect the emotional state of employees, and to eliminate the interference of data time deviation on subsequent analysis by synchronizing the timestamps, so as to provide an accurate and consistent data foundation for subsequent emotional feature extraction.

[0055] In the specific implementation process, the first step is to construct a multi-source data collection system, designing collection schemes for three core data categories: basic physiological data, work behavior data, and environmentally related data. For basic physiological data, real-time collection can be achieved through smart devices with biosensor functions worn by employees (such as smart bracelets and smartwatches). The collected data includes, but is not limited to, heart rate variability (reflecting the activity of the autonomic nervous system and directly related to mood fluctuations), skin conductance (an important physiological indicator reflecting emotional stress response), and fingertip oxygen saturation (indirectly reflecting the degree of physical fatigue and thus related to emotional state). For work behavior data, collection is conducted through the enterprise office system's backend data interface and behavior monitoring module, specifically covering task interaction frequency (such as email sending / …). The data includes behavioral indicators directly related to the work process, such as the number of times messages are received, the number of times collaborative documents are edited, the frequency of instant messaging messages sent, the duration of document editing pauses (the cumulative value of the period during which no operation lasts longer than a preset threshold), and the response time for cross-departmental collaboration (the time interval from receiving a collaboration request to the first response). For environmental data, environmental sensing devices deployed in the office area (such as light sensors, sound level meters, and temperature and humidity sensors) are used to collect data, including the light intensity at the workstation (both excessive and insufficient light can affect emotional state), the sound pressure level in the office area (the interference of noise levels on emotional stability), and the temperature and humidity around the workstation (unsuitable temperature and humidity can easily cause irritability).

[0056] After completing various data collection processes, the collected multidimensional data needs to be timestamped and synchronized. Since different data collection devices (such as smart bracelets, office systems, and environmental sensors) may have system time discrepancies, this step uses a unified time base (e.g., the enterprise office system server time) to add a timestamp accurate to the millisecond level to each collected data record. Then, a data calibration algorithm is used to align the timestamps from different data sources. For data with time discrepancies, corrections are made according to preset time discrepancy compensation rules (e.g., the average time discrepancy value calculated based on historical synchronization data). This results in a multidimensional dataset with consistent time dimensions and complete data records, ensuring that subsequent data across all dimensions can be correlated and analyzed at the same time points.

[0057] As described in step S2 above, the aim is to extract key feature indicators that can be directly related to emotional state from the multidimensional dataset obtained in step S1, and to integrate the scattered single-dimensional features into an initial emotional feature vector that can comprehensively reflect the emotional state of employees through a weighted fusion method, so as to lay the feature foundation for subsequent comparison with historical baselines.

[0058] In the multi-dimensional feature weighted fusion stage, based on the degree of influence of each dimension feature on emotion assessment, weights are set for the fluctuation range of basic physiological data, the frequency abnormality of work behavior data, and the interference coefficient of environmental correlation data (e.g., 35%, 45%, 20%, the weights can be calibrated based on the correlation analysis between historical emotion assessment data and actual emotion feedback). The feature values ​​of the three dimensions are multiplied by their corresponding weights and then summed to obtain the comprehensive emotion feature value. At the same time, in order to retain the original information of each dimension feature for the convenience of subsequent traceability analysis, the comprehensive emotion feature value is integrated with the original extraction results of each dimension feature and the corresponding timestamp information to construct a three-dimensional vector structure of timestamp-each dimension feature value-comprehensive emotion feature value. This vector is the initial emotion feature vector, which can comprehensively and concisely carry the core feature information of employee emotion correlation within the current assessment period.

[0059] The core function of step S3 above is to establish an emotional assessment benchmark that can reflect individual differences among employees and differences in scenarios, providing a scientific reference for subsequent judgment on whether the current emotional state is abnormal, and avoiding assessment bias caused by using a uniform standard.

[0060] The Employee Historical Emotion Baseline Database is a structured database built upon the long-term accumulation and analysis of employee emotion-related data. Its core features lie in individual targeting and scenario adaptability: In terms of individual targeting, the database establishes independent data profiles for each employee, avoiding benchmark confusion caused by differences in personality, work habits, and emotional thresholds among different employees; in terms of scenario adaptability, the database divides employees' work scenarios into several typical categories (such as independent office scenarios, team discussion scenarios, customer communication scenarios, project tackling scenarios, and routine work scenarios), and matches corresponding emotional characteristic benchmark values ​​for each scenario, ensuring that evaluation standards that fit the scenario can be used in different work scenarios.

[0061] The baseline values ​​for emotional characteristics stored in the database are obtained by analyzing and calculating multidimensional data of employees over a relatively long period (e.g., 3-6 months). For each typical work scenario, multidimensional data on employees in a normal emotional state (annotated through historical emotional feedback surveys, stable performance data, etc.) are extracted. Following the feature extraction method in step S2, the mean and reasonable fluctuation range of basic physiological data fluctuations, the mean and reasonable fluctuation range of abnormal frequency of work behavior data, and the mean and reasonable fluctuation range of interference coefficients of environmental correlation data are calculated for that scenario. These mean values ​​and ranges are then used as the baseline values ​​for emotional characteristics in that scenario (e.g., the mean of physiological fluctuations in an independent office scenario is X, and the reasonable range is X±Y). Simultaneously, the database has a data update mechanism that periodically (e.g., monthly) incorporates the latest normal emotional state data of employees, iteratively optimizing the baseline values ​​for emotional characteristics in each scenario to ensure that the baseline values ​​always remain consistent with the current emotional state patterns of employees, avoiding the invalidation of baseline values ​​due to changes in employees' work status or lifestyle habits.

[0062] During this step, the historical emotional baseline database corresponding to the employee is accessed through the employee's identity identifier (such as employee ID or account). Based on the current work scenario of the employee (determined by office system scenario marking, task type identification, etc., such as determining that the employee is participating in a video conference, it is determined to be a team discussion scenario), the emotional characteristic benchmark value corresponding to the scenario is selected, providing an accurate reference standard for the deviation comparison analysis in step S4.

[0063] Step S4, as described above, is the core execution step of the entire evaluation method. By comparing the current emotional characteristics with historical scenario benchmarks and combining them with a weight matrix, the emotional state is quantitatively evaluated, and the final output is a real-time emotional state evaluation result that can directly guide management decisions.

[0064] In the multidimensional deviation calculation process, the baseline value of emotional characteristics in the corresponding work scenario selected in step S3 is used as a reference to perform deviation analysis on the initial emotional feature vector generated in step S2: For the fluctuation range of basic physiological data, the absolute difference between the current value and the baseline mean is calculated, and then the difference is divided by the baseline mean to obtain the physiological deviation rate (e.g., if the current physiological fluctuation range is A and the baseline mean is B, then the physiological deviation rate = |AB| / B). This indicator reflects the degree of deviation of the current physiological level emotional fluctuation from the historical normal state. Similarly, the behavioral deviation rate of the frequency anomaly of work behavior data and the environmental deviation rate of the interference coefficient of environmental correlation data are calculated respectively. At the same time, the difference between the comprehensive value of emotional characteristics in the initial emotional feature vector and the baseline comprehensive value in the corresponding scenario (calculated by the mean of each dimension of the baseline according to the weight in step S2) is calculated to obtain the comprehensive deviation value. This value can intuitively reflect the magnitude of the difference between the current overall emotional state and the historical normal state.

[0065] In the weighted matrix adaptation and fusion process, a preset emotion influence weighted matrix is ​​first invoked. This matrix contains the basic weights of each dimension's deviation rate in emotion assessment (e.g., physiological deviation rate 35%, behavioral deviation rate 45%, environmental deviation rate 20%). Considering the varying degrees of influence of each dimension on emotion in different work scenarios, the matrix also includes scenario correction coefficients. For example, in a customer communication scenario, work behaviors (e.g., communication efficiency, response attitude) have a more significant impact on emotion; therefore, the weight of the behavioral deviation rate is increased by 15%, while the weights of the physiological and environmental deviation rates are correspondingly reduced. In an independent office scenario, physiological state (e.g., focus, fatigue) has a greater impact on emotion; therefore, the weight of the physiological deviation rate is increased by 10%. After adjusting the weights of each dimension's deviation rate according to the current work scenario, each dimension's deviation rate is multiplied by the adjusted weights and summed to obtain a weighted deviation score. If the absolute value of the overall deviation value is greater than a preset threshold (e.g., 0.5), the weighted deviation score is secondarily calibrated (e.g., multiplied by a calibration coefficient of 1.2) to amplify significant deviations from normal emotional states and avoid overly conservative assessment results.

[0066] In the evaluation result generation and output process, the first step is to set the threshold for the emotional level corresponding to the weighted total deviation score. For example, when the weighted total deviation score is in the range of 0-0.2, it is judged as a positive emotional state, corresponding to sub-emotional labels such as pleasure, focus, and efficiency (determined based on the positive or negative direction of the deviation rate of each dimension; for example, a negative behavioral deviation rate indicates better behavioral frequency, which may correspond to the focus label). When it is in the range of 0.2-0.4, it is judged as a stable emotional state, corresponding to sub-labels such as calm, normal, and no significant fluctuations. When it is greater than 0.4, it is judged as a negative emotional state, corresponding to sub-labels such as anxiety, fatigue, and irritability (for example, a positive physiological deviation rate with a large value may correspond to the fatigue label). Then, the information such as emotional level, sub-labels, deviation rates of each dimension, and core influencing dimensions (the deviation dimension with the highest weight; for example, if the behavioral deviation rate has the highest weight, it is judged as behavior-driven emotional change) is integrated to generate a structured real-time emotional state evaluation report. The report will also link the timestamp of the current assessment period, work scenario information, and data collection sources to ensure that the assessment results are traceable and verifiable. Finally, the assessment report will be pushed to the human resources management department or direct supervisor through the enterprise management platform to provide accurate data support for subsequent emotional interventions (such as adjusting work tasks, providing psychological support, and optimizing the office environment).

[0067] In one embodiment, basic physiological data includes heart rate data and skin conductance data; work behavior data includes task interaction behavior and document editing pause duration; and environmental correlation data includes workstation light intensity and office area sound pressure level.

[0068] In one embodiment, after outputting the real-time emotional state assessment result, the process includes:

[0069] When an employee's real-time emotional state assessment result is negative over a continuous period of time, an analysis of the sources of emotional influencing factors is triggered. By calculating the Pearson correlation coefficient between the data of each dimension and the real-time emotional state assessment result, the core influencing factors are identified and dynamic adjustment suggestions are generated.

[0070] In this embodiment, when the real-time emotional state assessment results of employees are all negative over a continuous period of time, the source analysis is automatically triggered to avoid false triggering due to short-term fluctuations.

[0071] Multidimensional data from the three time windows preceding the trigger were selected, low-reliability data were removed and standardized to eliminate interference from dimensionality. Pearson correlation coefficients were calculated for each sub-indicator of basic physiological data, work behavior data, and environmental data, along with the total weighted deviation score of emotion, to quantify the strength of linear correlation. Highly correlated factors were screened using an absolute correlation coefficient value ≥ 0.6 as a threshold, and the top 2-3 factors were selected based on their coefficients to pinpoint the core influencing factors. Actionable suggestions were generated based on the type of core influencing factors (relaxation guidance for physiological factors, task allocation adjustment for behavioral factors, and optimized office conditions for environmental factors).

[0072] In this embodiment, abstract negative emotions are transformed into specific influencing factors, avoiding subjective judgment and upgrading emotion assessment from a monitoring tool to a decision support tool, thus improving the closed loop of the assessment system and helping to alleviate negative emotions in a timely manner.

[0073] In one embodiment, extracting the fluctuation range of the basic physiological data includes:

[0074] For heart rate data in basic physiological data, the data is divided into segments by a preset time interval. The difference between the maximum and minimum heart rate values ​​in each data segment is calculated, and the average of the differences of multiple consecutive data segments is used as the heart rate fluctuation amplitude.

[0075] For the electrodermal activity data, the number of times the peak value of the electrodermal signal occurs per unit time is calculated, and the absolute value of the difference between the number of times and the preset normal range is taken as the fluctuation amplitude of the electrodermal activity.

[0076] The weighted average of the heart rate fluctuation amplitude and the skin electrical activity fluctuation amplitude is calculated as the fluctuation amplitude of the baseline physiological data.

[0077] In this embodiment, the fluctuation range of basic physiological data is the core physiological indicator reflecting the emotional fluctuation of employees. Its extraction process focuses on heart rate data and skin conductance data. Through multidimensional calculation and weighted fusion, a quantitative indicator that can comprehensively reflect the emotional changes at the physiological level is finally obtained.

[0078] Heart rate data changes dynamically with emotional state (e.g., heart rate increases when anxious, and remains stable when calm). By calculating the amplitude of heart rate fluctuations, the severity of heart rate changes can be quantified, indirectly reflecting emotional stability.

[0079] First, a preset time interval (e.g., 5 minutes) is set to divide the continuously collected heart rate data into several data segments of equal duration. The basis for setting this time interval is: too short (e.g., 1 minute) is easily affected by momentary interference (e.g., accidental getting up and moving around), causing data fluctuations and distortion; too long (e.g., 10 minutes) makes it difficult to capture heart rate changes caused by short-term emotions. A 5-minute interval can achieve a balance between anti-interference and sensitivity.

[0080] For each data segment, extract the maximum and minimum heart rate values ​​within that segment, and calculate the difference between them (e.g., if the maximum heart rate in a 5-minute segment is 85 beats / minute and the minimum is 70 beats / minute, the difference is 15 beats / minute). This difference directly reflects the range of heart rate variation within a single time period; the larger the difference, the more drastic the heart rate fluctuations during that period.

[0081] Select multiple consecutive data segments (e.g., 10 segments, corresponding to a total duration of 50 minutes), calculate the arithmetic mean of the differences between these data segments, and use this mean as the heart rate fluctuation amplitude. Using multiple segment means instead of single segment differences is to further reduce the impact of transient interference and ensure that the results reflect the heart rate fluctuation trend over a period of time, rather than occasional short-term fluctuations.

[0082] Electrodermal activity (such as changes in skin resistance) is controlled by the activity of the human sympathetic nervous system. Emotional stress (such as tension and irritability) significantly increases the frequency of peak electrodermal signals. Therefore, the amplitude of fluctuations in electrodermal activity can be directly correlated with the degree of emotional stress.

[0083] Set a unit time (e.g., 1 minute) and count the number of times the peak value of the skin conductance signal (the waveform peak that exceeds a preset signal strength threshold) occurs within that time. For example, the number of peak values ​​per unit time is usually 2-3 in a calm state, while it may rise to 5-6 in an anxious state;

[0084] Beforehand, by using employees' historical normal emotional data, the normal range of peak frequency for electrodermal signal (EDS) is determined (e.g., 2-4 times / minute). The absolute value of the deviation between the current peak frequency per unit time and this range is calculated. If the current frequency is 5 times, exceeding the upper limit of the normal range by 1 time, the absolute value of the deviation is 1; if the current frequency is 1 time, falling below the lower limit of the normal range by 1 time, the absolute value of the deviation is also 1. This absolute value of deviation directly quantifies the degree of deviation between the current EDS activity and the normal emotional state. The larger the value, the stronger the emotional stress response. This value is ultimately used as the amplitude of EDS activity fluctuation.

[0085] Heart rate fluctuations and skin conductance fluctuations reflect emotional states from different physiological dimensions (heart rate is related to the overall activity of the autonomic nervous system, while skin conductance focuses on the sympathetic nervous system stress response). They need to be weighted and integrated into a single indicator to simplify subsequent calculations of emotional characteristics.

[0086] Based on the strength of the correlation between physiology and emotion, the weight of heart rate fluctuation amplitude (e.g., 60%) is set higher than that of skin conductance activity fluctuation amplitude (e.g., 40%). The basis for this weighting is that heart rate fluctuation is more stable to be affected by emotions and covers a wider range of emotional types (e.g., pleasure, fatigue, and anxiety can all cause heart rate changes), while skin conductance activity fluctuation focuses more on stress-related emotions (e.g., tension, irritability). Therefore, a higher weight is given to heart rate fluctuation amplitude to ensure that the overall indicator can more comprehensively reflect various emotional changes.

[0087] Multiply the heart rate fluctuation amplitude by its weight (60%) and the skin conductance activity fluctuation amplitude by its weight (40%), and the sum of the two results is the fluctuation amplitude of the basic physiological data, which can be directly used as the core indicator of the physiological dimension in the subsequent initial emotion feature vector, providing a quantitative basis for emotion assessment.

[0088] In one embodiment, the extraction of frequency anomalies in work behavior data includes:

[0089] The frequency of task interaction behavior in statistical work behavior data is used to establish a normal distribution model of the frequency of task interaction of employees in the same working period within the past 30 days. The deviation between the current task interaction frequency and the model mean is calculated, and then the deviation is divided by the model standard deviation to obtain the standardized deviation.

[0090] If the absolute value of the standardized deviation is greater than the threshold, then the absolute value of the standardized deviation is used as the anomaly of the task interaction frequency.

[0091] The frequency anomaly of document editing pause duration is calculated, and the frequency anomalies of task interaction and document editing pause duration are weighted and fused to obtain the frequency anomaly of work behavior data.

[0092] In this embodiment, the extraction of frequency anomalies in work behavior data is primarily achieved by quantifying the degree of deviation in two key behaviors: task interaction and document editing pauses. This indirectly captures the impact of employee emotions on work performance (e.g., negative emotions are often accompanied by reduced interaction and increased pauses). The specific steps are as follows:

[0093] Task interaction behaviors (such as sending and receiving emails, collaborative document operations, and sending instant messages) are a direct reflection of employees' work collaboration and engagement. Abnormal frequency of these behaviors can reflect changes in work status caused by emotions. Calculations need to rely on historical data modeling and standardized analysis to ensure the objectivity and personalization of anomaly judgments.

[0094] First, the frequency of task interactions for the employee during the same working hours (e.g., 9:00-10:00 and 14:00-15:00 daily, ensuring consistency to eliminate interference from time factors) over the past 30 days was statistically analyzed to create a historical frequency dataset. Choosing 30 days as the period ensures a sufficient sample size to support model reliability (avoiding the influence of short-term random data) while also reflecting recent normal behavioral patterns (reducing interference from long-term changes in work status). Based on this dataset, a normal distribution model was constructed, with the core outputs being the mean and standard deviation. The mean represents the normal average level of task interactions during that period, and the standard deviation reflects the normal fluctuation range. Together, they constitute a personalized normal benchmark, avoiding the bias of using a uniform standard to measure different employees.

[0095] Statistically analyze the actual task interaction frequency during the current evaluation period (which must be completely consistent with the historical model period, such as 9:00-10:00 on the current day). First, calculate the deviation value from the model mean (deviation value = current frequency - mean). Since the absolute values ​​of interaction frequencies vary greatly among different employees (e.g., managers average 50 times per day, ordinary employees 20 times per day), directly using the deviation value cannot uniformly determine the degree of abnormality. Therefore, it needs to be divided by the model standard deviation σ to obtain the standardized deviation. This value eliminates the difference in absolute values, making the degree of abnormality comparable across different employees and time periods.

[0096] Set a standardized deviation threshold (usually 1.5 or 2, adjusted according to the company's requirements for anomaly accuracy). If the absolute value of the standardized deviation is greater than the threshold, it indicates that the current frequency significantly deviates from the normal range of the same period in the past 30 days (belonging to a low-probability anomaly event). In this case, the absolute value is used as the anomaly degree of the task interaction frequency. The larger the absolute value, the more obvious the anomaly. If the absolute value is less than or equal to the threshold, it indicates that the frequency is within the normal fluctuation range, and the anomaly degree is set to 0 to avoid misjudging normal behavior.

[0097] The frequency of document editing pauses (the duration of continuous no editing operations, for which a judgment criterion needs to be defined first, such as counting more than 5 minutes as one pause) directly reflects employee focus and work efficiency. Its anomaly calculation logic is consistent with the task interaction frequency, and only needs to be adapted to the statistical characteristics of pause events.

[0098] First, define a pause event as a continuous period of no editing operations exceeding a preset duration (e.g., 5 minutes). Count the frequency of pause events occurring during the same work period for the employee over the past 30 days to create a historical pause frequency dataset. Next, construct a normal distribution model based on the dataset to obtain the mean and standard deviation. Then, calculate the deviation between the actual pause frequency and the mean for the current period, and divide it by the standard deviation to obtain the standardized deviation. Finally, using the same threshold as the task interaction frequency, determine whether the absolute value of the standardized deviation exceeds the threshold. If it exceeds the threshold, the absolute value is taken as the document editing pause duration frequency anomaly; otherwise, it is taken as 0.

[0099] It should be noted that the standardized deviation of task interaction frequency is negative (with positive outlier), which usually indicates reduced interaction and negative emotions; the standardized deviation of document editing pause frequency is positive (with positive outlier), which usually indicates increased pauses, decreased focus, and negative emotions. Both reflect the impact of emotions from different behavioral dimensions.

[0100] Task interaction and document editing pauses are core behaviors in employees' daily work. These two types of anomalies need to be integrated into a single indicator through weighted fusion. This simplifies subsequent calculations of sentiment characteristics while also taking into account the differences in importance between the two types of behaviors.

[0101] First, weights are assigned based on the correlation strength between the two types of behaviors and emotional states (e.g., an equal weight of 5:5, since task interaction reflects collaboration and pauses reflect focus, and both have comparable value in mapping emotions; if the enterprise scenario is special, such as a collaborative role, the weights can be adjusted to 6:4). Then, the frequency anomaly of task interaction is multiplied by its weight, and the frequency anomaly of document editing pause duration is multiplied by its weight; the sum of these two is the frequency anomaly of the work behavior data. This value can be directly used as the core quantitative indicator of the work behavior dimension in the initial emotional feature vector, providing data support for subsequent emotion assessment.

[0102] By establishing personalized benchmarks through historical normal distribution models, we avoid individual biases caused by uniform standards; by standardizing biases, we achieve objective and comparable anomaly measurement; and by weighting and fusion to cover key behavioral dimensions, we ensure that anomalies accurately reflect the impact of emotions on work performance.

[0103] In one embodiment, the extraction of interference coefficients from environmental correlation data includes:

[0104] The workstation lighting intensity data in the environmental correlation data is compared with the preset comfortable lighting intensity range to calculate the lighting intensity interference coefficient.

[0105] For the sound pressure level data in the office area, the sound pressure level interference coefficient is obtained by using a preset decibel as the benchmark value and combining it with calculation.

[0106] The interference coefficient of the environmental correlation data is obtained by weighting and fusing the light intensity interference coefficient and the sound pressure level interference coefficient.

[0107] In this embodiment, light intensity directly affects human visual comfort and physiological rhythms (e.g., excessively weak light can easily lead to visual fatigue and low mood, while excessively strong light can easily cause visual stimulation and irritability). The calculation of its interference coefficient needs to be based on a comparison of the comfort range, and the magnitude of the interference is determined by quantifying the degree of deviation between the actual light intensity and the comfort range.

[0108] Based on ergonomics and office environment optimization standards, and combined with visual comfort feedback from most employees in office settings, a preset comfortable lighting intensity range is established (typically 200-500 lux, which meets the needs of office work such as document reading and screen operation while avoiding eye strain). If there are special job requirements within the company (such as designers requiring higher lighting for color reproduction), the comfortable range can be adjusted accordingly (e.g., 300-600 lux for designers) to ensure that the range is suitable for the actual office environment.

[0109] Real-time data on actual light intensity at employee workstations is collected and compared with preset comfort ranges. Light intensity interference coefficients are calculated for three scenarios:

[0110] If the actual light intensity is less than the lower limit of the comfort range (e.g., actual 150 lux < lower limit 200 lux): Interference coefficient = (lower limit of comfort range - actual light intensity) / lower limit of comfort range. This formula quantifies the degree of interference from insufficient light. The larger the value, the more severe the insufficient light and the stronger the negative interference on emotions.

[0111] If the actual light intensity is greater than the upper limit of the comfort range (e.g., actual 600 lux > upper limit 500 lux): Interference coefficient = (actual light intensity - upper limit of comfort range) / upper limit of comfort range. This quantifies the degree of interference caused by excessive light intensity. The larger the value, the more obvious the excessive light intensity and the stronger the interference.

[0112] If the actual light intensity is within the comfortable range (e.g., 250-450 lux): it indicates that there is no significant interference in the lighting environment, and the interference coefficient is set to 0 to avoid misjudging the normal environment as a interference factor.

[0113] The sound pressure level (noise level) in the office area is a key environmental factor affecting emotional stability (e.g., continuous noise can easily lead to distraction and anxiety). Its interference coefficient is calculated based on a benchmark value comparison, determining the magnitude of the interference by quantifying the deviation of the actual noise from the benchmark value.

[0114] Referring to the daytime noise limits in office areas in the office environment noise control standards, and taking into account the sound needs of employees to concentrate on work, a preset sound pressure level benchmark value is set (usually 50 decibels, at which the ambient sound is soft and does not affect normal communication and focused work).

[0115] Real-time sound pressure level data is collected in the office area (within a 3-meter radius of employee workstations), and the interference coefficient is calculated by comparing it with a preset benchmark value.

[0116] If the actual sound pressure level is greater than the reference value (e.g., actual 65 dB > 50 dB): Interference coefficient = (actual sound pressure level - reference value) / reference value. This formula quantifies the degree of interference caused by excessive noise. The larger the value, the more serious the noise and the stronger the interference on emotions.

[0117] If the actual sound pressure level is less than or equal to the reference value (e.g., 45 dB ≤ 50 dB): it indicates that the noise level is within an acceptable range and there is no significant interference. The interference coefficient is set to 0 to ensure that the normal sound environment is not included in the interference.

[0118] It should be noted that the sound pressure level interference coefficient calculation only considers situations where noise exceeds the standard (actual value > benchmark value). Sounds below the benchmark value (such as slight keyboard typing or normal communication) usually do not have a negative impact on emotions and may instead form background white noise, so they do not need to be included in the interference coefficient.

[0119] Light intensity and sound pressure level are the two most direct and significant factors affecting mood in the office environment. These two types of interference coefficients need to be integrated into a single indicator through weighted fusion, which simplifies subsequent calculations of mood characteristics while also taking into account the difference in the importance of the two factors.

[0120] By combining the weights of the two types of environmental factors on the intensity of their impact on emotions, sound pressure level (noise) has an immediate effect on emotions (e.g., sudden noise can instantly interrupt concentration and cause irritability), and its impact is usually more direct. Light intensity, on the other hand, has a cumulative effect on emotions (e.g., prolonged insufficient light gradually leads to low mood), and its impact is relatively milder. Therefore, the interference coefficient weight for sound pressure level is usually set higher (60%) than that for light intensity (40%). If the enterprise scenario is special (e.g., design positions are more sensitive to light), the weights can be adjusted to 50% for light intensity and 50% for sound pressure level to ensure that the weights are appropriate for the job requirements.

[0121] Multiplying the light intensity interference coefficient by its weight and the sound pressure level interference coefficient by its weight, the sum of the two is the interference coefficient of the environmental correlation data. This value can be directly used as the core quantitative indicator of the environmental dimension in the initial emotion feature vector, providing accurate data support for the degree of environmental interference in subsequent emotion assessment.

[0122] In one embodiment, the difference between the initial emotion feature vector and the historical emotion feature benchmark value under the corresponding work scenario is calculated, and combined with a preset emotion influence weight matrix, a real-time emotion state assessment result is output, including:

[0123] Retrieve historical emotional characteristic baseline values ​​from the employee historical emotional baseline database that are consistent with the current work scenario and the same time period; among them, the historical emotional characteristic baseline values ​​include historical physiological fluctuation amplitude baseline, historical behavioral abnormality baseline, and historical environmental interference coefficient baseline;

[0124] Physiological deviation rate is calculated based on the fluctuation range of basic physiological data and the historical physiological fluctuation range benchmark; behavioral deviation rate is calculated based on the frequency abnormality of work behavior data and the historical behavioral abnormality benchmark; environmental deviation rate is calculated based on the interference coefficient of environmental correlation data and the historical environmental interference coefficient benchmark.

[0125] A preset emotion influence weight matrix is ​​invoked, which includes basic weights and scene correction coefficients; the basic weights are corrected based on the scene correction coefficients corresponding to the current work scene to obtain the corrected weights;

[0126] After adjusting the weights, the physiological deviation rate, behavioral deviation rate, and environmental deviation rate are weighted and calculated to obtain the weighted total deviation score.

[0127] By comparing the weighted total deviation score with the emotion level threshold, a real-time emotion state assessment result is obtained, which includes emotion level, sub-labels, deviation rate of each dimension, and core influence dimension.

[0128] In this embodiment, comparing and analyzing the initial emotion feature vector with the historical emotion feature benchmark value under the corresponding work scenario, and combining it with the emotion influence weight matrix to output the evaluation result, is the core step in achieving accurate quantification of emotional state.

[0129] Historical emotional baseline values ​​serve as a personalized benchmark for determining whether current emotions are abnormal. Their retrieval must strictly match the current scenario and time period to avoid assessment bias due to baseline mismatch: From the employee's historical emotional baseline database, priority is given to selecting baseline data that is completely consistent with the current work scenario. Work scenarios are categorized by core task attributes (e.g., independent office scenarios, customer communication scenarios, team discussion scenarios, project tackling scenarios) to ensure that the baseline values ​​match the employee's current work situation (e.g., if the current scenario is customer communication, only historical customer communication scenario baseline data should be retrieved); simultaneously, the condition of matching the same time period must be met to eliminate the influence of differences in work status at different times on the baseline.

[0130] The retrieved historical emotional feature benchmark values ​​must correspond one-to-one with the dimensions of the initial emotional feature vector. Specifically, this includes three core indicators: historical physiological fluctuation amplitude benchmark (i.e., the mean and reasonable range of fluctuation amplitude of basic physiological data under the same historical scenario and time period, such as mean 8.5 and range 6.0-11.0), historical behavioral anomaly benchmark (the mean and reasonable range of frequency anomalies of work behavior data under the same historical scenario and time period, such as mean 0.8 and range 0.3-1.3), and historical environmental interference coefficient benchmark (the mean and reasonable range of interference coefficients of environmentally related data under the same historical scenario and time period, such as mean 0.2 and range 0.0-0.4). These benchmarks provide direct reference for subsequent multidimensional deviation calculations.

[0131] The calculation is based on the fluctuation range of basic physiological data and the historical average physiological fluctuation range, using the relative deviation formula: Physiological deviation rate = |Current physiological fluctuation range - Historical average physiological fluctuation range| / Historical average physiological fluctuation range. This value reflects the proportion of deviation of the current physiological emotional fluctuation from the historical normal state; the larger the value, the more abnormal the physiological emotional response.

[0132] Similarly, based on the frequency of abnormality in work behavior data and the historical baseline mean of abnormality (e.g., a historical mean of 0.8), the same relative deviation formula is used to calculate: Behavioral Deviation Rate = |Current Behavioral Deviation - Historical Baseline Mean| / Historical Baseline Mean. This value quantifies the difference between the current degree of abnormality in work behavior and the historical normal state; the larger the value, the more significant the influence of emotions on behavior.

[0133] Based on the interference coefficient of environmental data and the historical average environmental interference coefficient, the environmental deviation rate is calculated as follows: Environmental Deviation Rate = |Current Environmental Interference Coefficient - Historical Average Environmental Interference Coefficient| / Historical Average Environmental Interference Coefficient. For example, if the current value is 0.5 and the historical average is 0.2, then the environmental deviation rate = |0.5 - 0.2| / 0.2 = 1.5 (i.e., 150%). This value reflects the proportion of deviation of the current level of environmental interference from the historical normal environment, providing a basis for judging whether the environment is a trigger for emotional abnormalities.

[0134] The emotion influence weight matrix is ​​used to allocate the contribution of each dimension's deviation rate to the final evaluation result. It achieves dynamic adaptation through basic weights and scenario correction coefficients to avoid the problem that fixed weights cannot match scenario differences. The preset emotion influence weight matrix contains two core components: one is the basic weight, which is the weight ratio of each dimension under the default scenario, reflecting the general influence of each dimension on the emotion evaluation; the other is the scenario correction coefficient, which is the weight adjustment ratio set for different work scenarios.

[0135] Based on the current work scenario of the employee, the corresponding scenario correction coefficient is applied to adjust the basic weights. For example, in a customer communication scenario, the basic weights are 35% physiological, 45% behavioral, and 20% environmental. After adjustment using the scenario correction coefficient (behavior +15%, physiological -10%, environmental -5%), the corrected weights are 25% physiological, 60% behavioral, and 15% environmental. The corrected weights better reflect the actual impact of each dimension on emotions in the current scenario, ensuring the scenario-appropriateness of the assessment results.

[0136] The weighted deviation score is a core indicator that comprehensively measures the degree of current emotional abnormality. By integrating the deviation rates of each dimension according to adjusted weights, it transforms multi-dimensional differences into a single quantitative score.

[0137] The specific calculation method is as follows: Weighted Deviation Score = Physiological Deviation Rate × Physiological Correction Weight + Behavioral Deviation Rate × Behavioral Correction Weight + Environmental Deviation Rate × Environmental Correction Weight. This score directly quantifies the overall degree of deviation of the current emotion from the historical normal state; the higher the score, the more significant the emotional abnormality.

[0138] The assessment results should be graded based on the weighted total deviation score, and supplemented with dimensional details to provide a clear basis for subsequent interventions: Emotional level determination: Preset the emotional level threshold corresponding to the weighted total deviation score (calibrated based on corporate emotional management goals and historical data), for example:

[0139] 0-0.2 (0-20 points): Positive emotion level, corresponding to sub-labels such as pleasure, focus, and efficiency (judgment criteria: low deviation rate in each dimension, no significant abnormalities);

[0140] 0.2-0.4 (20-40 points): Stable emotional level, corresponding to sub-labels such as calm, normal, no obvious fluctuations (judgment basis: slight deviations in individual dimensions, but no abnormalities overall).

[0141] >0.4 (>40 points): Negative emotion level, corresponding to sub-labels such as anxiety, fatigue, and irritability (judgment criteria: significant overall deviation, or excessively high deviation in key dimensions).

[0142] Taking the above weighted deviation total score of 0.8975 (89.75 points) as an example, it is judged as a negative emotion level. Combined with the highest behavioral deviation rate (125%), the sub-label can be defined as irritability (behavior-driven).

[0143] The final output of the real-time emotional state assessment should include four core types of information: First, the emotional level (e.g., negative) and sub-labels (e.g., irritability), which intuitively reflect the emotional state; second, the deviation rate of each dimension (e.g., physiological 41%, behavioral 125%, environmental 30%), showing the specific degree of abnormality in each dimension; third, the core influencing dimension (e.g., behavioral dimension), which is the dimension with the highest weight and the largest deviation rate, clarifying the dominant factors of emotional abnormality; and fourth, the current work scenario and assessment time (e.g., customer communication scenario, Wednesday morning 10:00-11:00), to ensure that the results are traceable and verifiable.

[0144] In one embodiment, before outputting the real-time emotional state assessment result by combining a preset emotion influence weight matrix, the process includes:

[0145] Using the fluctuation range of basic physiological data, the frequency abnormality of work behavior data, and the interference coefficient of environmental correlation data as variables, the correlation coefficient between any two variables is calculated to form a correlation matrix.

[0146] If the correlation coefficient between any two groups of variables reaches the threshold, they are determined to be strongly correlated dimension groups. The weights of the corresponding two groups of variables are increased by a preset proportion based on the original emotion influence weight matrix, while the weights of other variables are reduced. If the correlation coefficients between all variables are less than the threshold, the original emotion influence weight matrix remains unchanged.

[0147] In the output of the real-time emotion state assessment results, strongly correlated dimension groups and their corresponding correlation coefficients are labeled.

[0148] In this embodiment, before outputting the real-time emotional state assessment result by combining the preset emotional influence weight matrix, the inherent correlation between basic physiological data, work behavior data, and environmental related data is identified to optimize the rationality of weight allocation and avoid assessment bias caused by ignoring the coupling effect between dimensions. At the same time, correlation information is marked in the results to improve the targeting of intervention. The specific steps are as follows:

[0149] The analysis sample consists of the fluctuation range of basic physiological data, the frequency anomaly of work behavior data, and the interference coefficient of environmental correlation data from the current assessment period and several previous consecutive assessment periods. This ensures that the sample size is sufficient to support the reliability of the correlation coefficient calculation and avoids random biases in the coefficients due to insufficient sample size. The sample data is standardized by converting the data values ​​of each dimension to a uniform numerical range to eliminate the interference of differences in units of measurement (such as different units of measurement for different dimensions) on the correlation coefficient calculation, ensuring that the three are correlated on the same order of magnitude.

[0150] The Pearson correlation coefficient is used as a quantitative indicator of the correlation coefficient (the Pearson correlation coefficient ranges from negative one to positive one; the closer the absolute value of the coefficient is to one, the stronger the linear correlation between the two dimensions of data; the closer it is to zero, the weaker the correlation). The correlation coefficient between any two dimensions of data is calculated as follows:

[0151] Calculate the correlation coefficient between the fluctuation range of basic physiological data and the frequency abnormality of work behavior data: This is used to measure the degree of correlation between physiological emotional response and work behavior abnormality. For example, if the coefficient is at a high level, it indicates that the greater the fluctuation range of basic physiological data, the higher the frequency abnormality of work behavior data. The two are strongly positively correlated, which may reflect the linkage effect of physiological stress causing a decline in behavioral efficiency.

[0152] The correlation coefficient between the fluctuation range of basic physiological data and the interference coefficient of environmental data is calculated: it is used to measure the degree of correlation between physiological emotional response and environmental interference. For example, if the coefficient is at a high level, it indicates that the larger the interference coefficient of environmental data, the higher the fluctuation range of basic physiological data. The two are strongly positively correlated, reflecting the linkage effect of physiological stress caused by environmental discomfort.

[0153] The correlation coefficient between the frequency anomaly of work behavior data and the interference coefficient of environmental data is calculated. This coefficient measures the degree of correlation between work behavior anomalies and environmental interference. For example, if the coefficient is at a medium level, it indicates that the larger the interference coefficient of environmental data, the higher the frequency anomaly of work behavior data. The two are moderately positively correlated, which may reflect the linkage effect of environmental noise causing a decrease in behavioral focus.

[0154] Formation of the correlation matrix: The three sets of calculated correlation coefficients are integrated in matrix form to form a third-order correlation matrix (the diagonal elements are the correlation coefficients of each dimension's data itself, with a value of one, and the off-diagonal elements are the correlation coefficients between pairs of dimensions). This matrix can intuitively present the correlation between the three dimensions of data, providing structured data basis for subsequent determination of strongly correlated dimension groups.

[0155] A preset correlation coefficient threshold is set (this threshold is determined by combining the accuracy requirements of emotion assessment and historical data calibration). If the absolute value of the correlation coefficient between any two sets of dimensional data reaches or exceeds the threshold, the two sets of dimensional data are determined to constitute a strongly correlated dimensional group. If multiple sets of correlation coefficients reach or exceed the threshold at the same time (such as the correlation coefficients between the fluctuation amplitude of basic physiological data and the frequency abnormality of work behavior data, and the correlation coefficients between the fluctuation amplitude of basic physiological data and the interference coefficient of environmental correlation data all meet the conditions), then they are simultaneously determined to be multiple strongly correlated dimensional groups.

[0156] For strongly correlated dimension groups, an adjustment strategy of weight increase followed by compensatory decrease is adopted to ensure that the sum of the weights of each dimension remains 100% after the adjustment. The specific adjustment method is as follows:

[0157] Preset weight increase ratio (this ratio is determined based on the dimensional correlation strength and evaluation fairness calibration, and is usually within a reasonable medium range): If two sets of dimensional data are determined to be strongly correlated dimensional groups, the basic weights of the two sets of dimensional data in the original emotion influence weight matrix are increased by a preset ratio.

[0158] The sum of the weights of the two sets of dimension data is deducted from the base weights of the third set of dimension data outside the strongly correlated dimension group; if there are multiple strongly correlated dimension groups, the weights to be deducted are allocated according to the proportion of the base weights of the third set of dimension data.

[0159] For example, if the basic physiological data and work behavior data are determined to be a strongly correlated dimension group, then the weights of the basic physiological data dimension and the work behavior data dimension will be increased by a preset ratio, and the total increased weight will be deducted from the basic weight of the environmental correlation data dimension. If the weight of the environmental correlation data dimension is too low or negative after deduction, the preset increase ratio can be appropriately reduced to keep the weight of the environmental correlation data dimension above a reasonable minimum level, ensuring that each dimension has a weight percentage.

[0160] If the correlation coefficients between all dimensions of data are lower than the preset threshold (e.g., none of the three sets of correlation coefficients reach the threshold), it indicates that there is no significant correlation between the data in each dimension. The original sentiment influence weight matrix should be kept unchanged to avoid over-adjustment that could lead to weight distortion.

[0161] Verification of the effectiveness of the adjusted weights: Calculate the product of the adjusted weights of each dimension and the standard deviation of the corresponding dimension data. If the sum of the products of strongly correlated dimension groups accounts for a proportion of the total sum of the products of all dimensions that meets the preset reasonable proportion (this proportion must ensure that the influence of strongly correlated dimensions is fully reflected), it indicates that the weight adjustment effectively reflects the influence of strongly correlated dimensions. If this proportion does not meet the preset value, readjust and increase the proportion until the requirements are met, ensuring that the influence of strongly correlated dimensions is fully reflected in the weights.

[0162] In this embodiment, a new dimension correlation annotation module is added to the output real-time emotional state assessment results, presenting strongly correlated dimension groups and their corresponding correlation coefficients, providing a basis for dimensional coupling relationships for subsequent emotional intervention:

[0163] Clearly label the strongly correlated dimension combinations; label the specific values ​​of the correlation coefficients of the strongly correlated dimension groups to intuitively reflect the strength of the correlation; based on the positive and negative directions and magnitudes of the correlation coefficients, briefly explain the correlation effect (e.g., basic physiological data and work behavior data are strongly positively correlated: an increase in the fluctuation range of basic physiological data is accompanied by a significant increase in the frequency of abnormalities in work behavior data, indicating a linkage effect of physiological stress leading to a decrease in behavioral efficiency).

[0164] By identifying the coupling effect between dimensions through correlation analysis, the limitations of fixed weights in traditional single-dimensional systems are overcome, enabling dynamic adaptation of weights. By presenting correlations through result labeling, emotion assessment is upgraded from single-dimensional attribution to multi-dimensional linkage analysis, further enhancing the scientific nature of assessment results and the targeted nature of interventions, and improving the technical depth and practicality of the dynamic emotion assessment system.

[0165] In one embodiment, the method further includes:

[0166] Obtain the employee's employee ID; the employee ID includes English letters and numeric characters; decode the English letters using a preset encoding table to obtain the corresponding decoded numbers; combine the numeric characters in the employee ID to obtain multiple different combined numbers;

[0167] Obtain a preset key string, and modify the key string based on each of the combined numbers to obtain an adjusted string;

[0168] Obtain a preset array, which includes multiple array positions; sequentially insert the characters in the adjustment string into each array position one by one from the starting position of the preset array until all characters are inserted into the preset array, thus obtaining a character array;

[0169] A mutation curve is generated based on the decoded numbers, and the character array is mutated and adjusted based on the mutation curve to obtain an adjusted array;

[0170] A graph is generated based on the attribute features of the multidimensional data, and an encryption key is generated based on the graph and the adjustment array for encrypted storage of the multidimensional data.

[0171] In this embodiment, the employee ID serves as a unique identifier within the company, possessing individual uniqueness. This step extracts basic digital information that can be used for subsequent key mutation through structured processing of the employee ID, providing personalized initial data support for encryption. The specific operation logic is as follows:

[0172] Employee ID Acquisition and Structure Recognition: Through the enterprise human resource management system interface or employee identity authentication module, the unique employee ID of the current data is obtained, and the English letters (such as DEV" and "HR") and numeric characters (such as "20230512") in the employee ID are automatically recognized. The segment positions of the two types of characters are clearly defined (such as prefix letters + suffix numbers, letters and numbers are interspersed), laying the foundation for subsequent classification processing.

[0173] English letter decoding and conversion: The preset encoding table (built based on ASCII code or custom mapping rules) is called to decode the English letters in the employee number one by one, converting each letter into the corresponding decimal decoded number; if there are consecutive letters in the employee number, the decoded numbers are concatenated in alphabetical order to form a unique letter-decoded number string for that employee, ensuring that the decoding result for each employee is unique.

[0174] Multiple combinations of numeric characters are generated: Numeric characters in the employee ID are processed in multiple dimensions to generate multiple different combinations of numbers, avoiding insufficient randomness caused by single number combinations. Combination rules include: segmented combination by number of digits, split combination by parity, and sliding combination by fixed length; each combination rule generates 1-3 combinations of numbers, ultimately forming a set containing multiple different combinations of numbers, providing multiple sets of differentiated adjustment parameters for subsequent key mutation.

[0175] The preset key string serves as the basic carrier for encryption. This step modifies and adjusts it using a combination of numbers generated from the employee ID, breaking the static nature of the fixed key and enhancing its dynamic randomness. A preset initial key string (typically 16-32 characters long, randomly generated from uppercase and lowercase letters, numbers, and special symbols) is retrieved from the secure storage module of the enterprise data encryption management system (such as a hardware encryption machine or encrypted database). This initial key string serves only as a basic template and is not directly used for encryption; it needs to be modified and adjusted to form a unique key.

[0176] The generated set of combined numbers is associated with the character positions of the key string. For example, the sum of the digits of the first combined number (e.g., the sum of the digits of the combined number "2023" is 2+0+2+3=7) is taken as the starting position of the first mutation in the key string; the last digit of the second combined number (e.g., the last digit of the combined number "0512" is 2) is taken as the interval number of character replacement; the parity of the combined number (e.g., "2305" is odd) determines the mutation operation type (odd numbers correspond to character replacement, even numbers correspond to character order reversal).

[0177] Based on the aforementioned association mapping rules, multiple rounds of mutation operations are performed on the initial key string: at the specified mutation start position, the original character is replaced with a character whose ASCII code value is offset by the "last digit of the combined number" (e.g., the original character "K" has an ASCII code of 75, and the last digit of the combined number is 3, so it is replaced with N corresponding to 75+3=78); for character segments divided at specified intervals, the character order of some segments is reversed. Preset special symbols are inserted at positions where the combined number is a prime number.

[0178] After multiple rounds of mutation, an adjustment string with a difference of ≥40% from the initial key string is generated to ensure that the adjustment string of each employee is unique and random.

[0179] The preset array serves as the structured carrier of the key. By inserting the adjusted string into the array through a cyclic insertion method, the limitations of linear keys are broken, and a two-dimensional structured character array is constructed, which increases the difficulty of key cracking.

[0180] Retrieve the preset array (this array is designed as a two-dimensional matrix with a fixed number of rows and columns based on encryption requirements, such as 4 rows and 6 columns, or 5 rows and 5 columns. The array positions are numbered according to the "row priority" or "column priority" rule, such as positions 1-24 for a 4 row and 6 column array), parse the number of rows, columns, position numbering rules, and starting position of the array (usually the top left corner of the array, numbered 1), and ensure that the insertion rules are compatible with the array structure.

[0181] Starting from the first character of the adjusted string, characters are inserted sequentially from the starting position into the array positions: characters of the adjusted string are inserted in ascending order of array position number; if the length of the adjusted string is less than the total number of array positions (e.g., the string length is 18 and the total number of array positions is 24), then after the string traversal is completed, the insertion is repeated from the first character until all array positions are filled (e.g., the first character of the string is filled into the 19th position, the second character is filled into the 20th position, and so on).

[0182] If the length of the string to be adjusted is greater than the total number of positions in the array (e.g., the string length is 30 and the total number of positions in the array is 24), then the first 24 characters of the string are inserted, and the remaining characters are temporarily stored as backup data for subsequent adjustments to ensure that there are no empty positions in the array after insertion.

[0183] After all characters are inserted, a two-dimensional character array is formed, which contains all characters (or the core part of the adjusted string). The validity of the array is verified by character uniqueness check and position mapping check to avoid insertion errors that could lead to subsequent encryption deviations.

[0184] The mutation curve is generated based on the decoded numbers of employee IDs and possesses individual dynamic characteristics. This curve is used to perform secondary adjustments on the character array, further enhancing the array's randomness and dynamism. Specific operations include: constructing a two-dimensional coordinate system (horizontal axis for array position number, vertical axis for adjustment magnitude) using the obtained decoded letter numbers as basic parameters; generating the mutation curve by using each number in the decoded number string as a key node parameter of the curve; connecting the key nodes into a continuous mutation curve using linear interpolation or polynomial fitting algorithms, with the peaks and troughs of the curve corresponding to the adjustment intensity of the array position (peaks indicate the largest adjustment magnitude, troughs the smallest); and converting the curve's adjustment magnitude values ​​into specific array adjustment operations.

[0185] Based on the adjustment rules corresponding to the variation curve, dynamic adjustments are performed on each position of the character array:

[0186] Position adjustment: Swap the character positions at the array positions corresponding to the curve peaks.

[0187] Character adjustment: For the array position corresponding to the trough of the curve, increment the character's ASCII code by 1.

[0188] Special adjustment: For the array position corresponding to the inflection point of the curve, perform character replacement with the pre-selected character.

[0189] The generation and verification of the adjustment array: After all adjustments are completed, an adjustment array with a difference of ≥30% from the original character array is generated, and the curve mapping consistency verification is performed to ensure that the adjustment operation matches the adjustment range of the mutation curve; encryption strength verification: the randomness of the array is detected by a third-party encryption tool to ensure that the randomness score is ≥85 points, verifying the security of the adjustment array and providing a reliable structured carrier for subsequent key generation.

[0190] Generation of multidimensional data attribute feature graphs: Extract the attribute features of the current multidimensional data to be encrypted (such as data collection duration, number of data dimensions, mean range of each dimension, and data collection device number), and convert these attribute features into graphical parameters:

[0191] Graphic type selection: Select the graphic type according to the number of data dimensions (e.g., 3 dimensions correspond to triangles, 4 dimensions correspond to quadrilaterals, and 5 or more dimensions correspond to polygons or circles).

[0192] Graphical parameter settings: Convert data acquisition duration into graphic side length, convert the mean range of data in each dimension into graphic interior angles, and convert device number into graphic center point coordinates.

[0193] Graphic digitization: The generated graphics are converted into a digital graphic matrix through graphics processing algorithms (such as converting the outline points, interior angles, side lengths and other parameters of the graphics into a decimal number matrix), forming a graphic digital carrier that is strongly associated with multi-dimensional data attributes.

[0194] The fusion of the graphic and adjustment arrays is achieved by performing bitwise operations on matrix elements to merge the graphic digital matrix and the adjustment array.

[0195] Position alignment: Adjust the number of rows and columns of the graphic number matrix to match the adjustment array.

[0196] Bitwise operation: Performs operations on corresponding numbers or characters in two matrices (e.g., adding the number matrix element to the character ASCII code value and taking the last two digits of the result as the merged element; characters are replaced and merged according to a preset mapping table).

[0197] Standardization of fusion results: Convert the elements of the fused matrix into a unified format (such as converting them into characters or decimal numbers) to form a fused matrix.

[0198] Encryption key generation and application: The fusion matrix is ​​concatenated into a string in row-major order, and then hashed using the SHA-256 hash algorithm to generate a fixed-length (e.g., 256-bit) encryption key. This encryption key is strongly bound to the employee's ID number and the attribute characteristics of the multidimensional data, and can only be used to encrypt and store the corresponding multidimensional data of the current employee (the key and data are combined using the AES-256 encryption algorithm, and the encrypted data is stored in the enterprise database). At the same time, the key generation log (including employee ID number, data attributes, and generation time) is stored in the security log system to facilitate subsequent key traceability and data decryption.

[0199] In this embodiment, through a multi-stage design involving unique employee IDs, dynamic key variation, array structuring, and data attribute binding, the encryption key possesses the characteristics of being individual-specific, dynamically random, and data-associated. This avoids the security risks associated with fixed keys while ensuring a strong correlation between the key and the data to be encrypted, effectively preventing data leakage and unauthorized decryption, and safeguarding the storage security of employees' multidimensional data.

[0200] In one embodiment, the method further includes:

[0201] Obtain employee identity information and simultaneously extract the MAC address and first login time of their fixed login device;

[0202] Convert the identity information into a number and perform a modulo operation with the first login time to obtain the modulo operation result. Convert the identity information into a number and sum it with the ASCII code of the MAC address to obtain the sum result.

[0203] The modulo operation result and the summation result are used to generate a feature matrix, and the dimension of the feature matrix is ​​adjusted according to the number of logged-in devices;

[0204] Using employee ID numbers as initial parameters, generate random sequences; embed the random sequences into a feature matrix and output mixed feature values;

[0205] An initial key is generated based on the mixed feature values; the employee's job permission level is obtained, and the initial key is transformed based on the exclusive transformation rules corresponding to the employee's job permission level to generate an encryption key, which is used for encrypted storage of the multidimensional data.

[0206] In this embodiment, firstly, through the enterprise's unified identity authentication system, with the employee's authorization and in compliance with data privacy regulations, the employee's name, department, job title, and de-identified ID number are collected. This information is then uniformly converted into an encoded string to achieve format standardization and avoid character format confusion from affecting subsequent calculations.

[0207] Based on the enterprise terminal management system, extract the MAC addresses of the fixed login devices (excluding temporary visitor devices) that employees use to access work data on a daily basis.

[0208] The system retrieves the time (accurate to the second) when the employee first successfully logs into the data platform on a fixed device from the enterprise login log system. This time is then converted into a Unix timestamp based on the UTC time zone (a decimal integer in seconds) to eliminate time discrepancies caused by time zone differences. This process ultimately forms the basic data set consisting of identity information, trusted device MAC address, and the first login timestamp.

[0209] Next, the standardized identity information string is converted character by character into decimal numbers and concatenated to form a continuous identity number string. The compressed identity number string (let's call it X) and the first login timestamp (let's call it T) are substituted into the modulo operation formula. If T is 0 (in the extreme case that timestamp acquisition fails), it is automatically replaced with the Unix timestamp corresponding to the company's establishment time as the base value. A unique feature value is generated through the modulo operation. Even if two employees have similar identity numbers, the results will be significantly different due to their different first login times.

[0210] The verified MAC address is converted character by character into its corresponding ASCII code value, and the total ASCII code value of the MAC address is obtained by summing them (let's call it M). Then, the identity number string (16 bits) is converted into an integer (let's call it Y), and summed with M. If the summation result exceeds 64 bits, it is compressed to a fixed length, and finally two bound quantized feature values ​​are formed, which provide the data foundation for the construction of the feature matrix.

[0211] Count the total number of fixed login devices of current employees from the enterprise trusted device library (let's call it N), and establish a mapping rule between the number of devices and the matrix dimension: when N=1, set a 2×2 basic matrix; when N=2, set a 3×3 extended matrix; when N≥3, set a (N+1)×(N+1) matrix, to ensure that the matrix dimension expands as the number of devices increases, so as to accommodate more device-related features.

[0212] Fill in the modulo operation results and summation results in a row-major cyclic filling rule until the first N rows of the matrix are filled; if the number of rows of the matrix is ​​(N+1), fill in the last row with the derived feature value of the number of devices (which can be customized) to avoid the matrix having empty elements.

[0213] After filling, the sum of all elements in the matrix is ​​calculated. If the sum is 0 (eigenvalue calculation error), the eigenvalue operation is re-executed. After the verification is passed, the formula (element value - minimum matrix value) / (maximum matrix value - minimum matrix value) is used to standardize all elements to the [0,1] interval, eliminating matrix bias caused by differences in the absolute values ​​of eigenvalues ​​of different employees, and providing a compatible structured carrier for subsequent random sequence embedding.

[0214] Furthermore, using the employee's unique employee ID as the seed for the pseudo-random number generator, a random sequence with the exact same number of elements as the feature matrix is ​​generated. For example, a 3×3 matrix with 9 elements generates 9 random numbers with values ​​in the range [0, 1000].

[0215] During the generation process, the numeric characters in the employee ID are extracted, and the last digit is used as the jump step size for random number generation (e.g., if the last digit is 2, the algorithm parameters are adjusted once every 2 random numbers generated) to ensure that the random sequences generated by different employee IDs are not repeated, thus realizing the unique random sequence driven by the employee ID.

[0216] Next, each random number in the random sequence is matched one by one with the normalized element at the corresponding position in the feature matrix, and the operation of matrix element × (1 + random number / 1000) is performed (which preserves the original matrix features while introducing dynamic interference). After the superposition is completed, the matrix elements are concatenated into a continuous string of numbers in column priority order, and compressed into a 32-bit hexadecimal string by the MD5 hash algorithm. This is the hybrid feature value that integrates the three features, which greatly improves the randomness and uniqueness of the subsequent key.

[0217] Then, an initial key is generated using the mixed feature values ​​as the core. If the feature value length is insufficient for the target key length, exclusive characters are added according to the employee's job permission level to complete the key to 32 bytes. This is then converted to binary format to form the initial key, which is already associated with the identity and device, but further security enhancements are needed based on permissions.

[0218] The system retrieves employee job permission levels from the enterprise access control system and matches them with corresponding key transformation rules: basic permissions are executed by character reversal and single-position replacement; intermediate permissions are executed by segmented obfuscation and symbol insertion; and advanced permissions are executed by multi-round XOR operations and hash recalculation. The transformation complexity increases with the level of permission, balancing security and management costs.

[0219] The final generated encryption key is used to perform encryption operations on the employee's multidimensional data (basic physiological, work behavior, and environmental related data) using the AES-256 symmetric encryption algorithm, generating ciphertext data which is then stored in the enterprise's encrypted database. At the same time, the key generation log (including identity fragments, MAC address, permission level, and generation time) is stored in the security log system, which can only be queried by authorized administrators, achieving full-link security control and preventing unauthorized decryption and data leakage.

[0220] In the above embodiments, this application incorporates some existing algorithms and technical features for explanation and description to make the specification more detailed, clear, and complete, thus complying with the provisions of the Patent Law. However, this is not achieved by using a series of complex steps and algorithmic formulas, nor by complicating the technical solution, nor by combining or stacking conventional or simple features. The existing algorithms and technical features listed are for the purpose of disclosing the specific implementation methods of each step of this application (not to limit this application) and to avoid situations where this application cannot be implemented.

[0221] Reference Figure 2 In another embodiment of the present invention, a dynamic emotional state assessment device based on multidimensional data analysis of employees is also provided, comprising:

[0222] The data acquisition unit is used to collect multidimensional data of employees, including basic physiological data, work behavior data and environmental correlation data, and to perform timestamp synchronization processing on the multidimensional data.

[0223] The extraction unit is used to extract the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and perform weighted fusion to obtain an initial emotion feature vector;

[0224] The acquisition unit is used to acquire the employee historical emotion baseline database; wherein, the employee historical emotion baseline database contains the emotional characteristic benchmark values ​​of employees in different work scenarios;

[0225] The output unit is used to calculate the difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario, and output the real-time emotion state assessment result by combining the preset emotion influence weight matrix.

[0226] In this embodiment, the specific implementation of each unit in the above device embodiment is described in the above method embodiment, and will not be repeated here.

[0227] Reference Figure 3 This invention also provides a computer device, which can be a server, and its internal structure can 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 provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0228] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0229] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. 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.

[0230] In summary, the dynamic emotional state assessment method based on multidimensional employee data analysis provided in this embodiment of the invention includes: collecting multidimensional employee data, including basic physiological data, work behavior data, and environmental correlation data; performing timestamp synchronization processing on the multidimensional data; extracting the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data, and performing weighted fusion to obtain an initial emotional feature vector; acquiring an employee historical emotional baseline database; wherein, the employee historical emotional baseline database contains the emotional feature benchmark values ​​of employees in different work scenarios; calculating the difference between the initial emotional feature vector and the historical emotional feature benchmark values ​​in the corresponding work scenarios, and combining it with a preset emotional influence weight matrix to output a real-time emotional state assessment result. In this invention, multidimensional employee data is collected, and the fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data are extracted respectively, and weighted fusion analysis is performed. Then, combined with the emotional influence weight matrix, a real-time emotional state assessment result is output. Through multidimensional data fusion analysis, the current shortcomings of accurately and comprehensively assessing employee emotions are overcome.

[0231] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. 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 embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can 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), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0232] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0233] 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 structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

[0234] In this application, all actions to acquire signal information or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization of the owner of the relevant device.

Claims

1. A dynamic emotional state assessment method based on multidimensional data analysis of employees, characterized in that, Includes the following steps: Collect multidimensional data from employees, including basic physiological data, work behavior data, and environmental data, and perform timestamp synchronization processing on the multidimensional data; The fluctuation amplitude of the basic physiological data, the frequency anomaly of the work behavior data, and the interference coefficient of the environmental correlation data are extracted and weighted and fused to obtain the initial emotion feature vector. Obtain the employee historical emotion baseline database; the employee historical emotion baseline database contains the baseline values ​​of employees' emotional characteristics in different work scenarios; The difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario is calculated, and combined with the preset emotion influence weight matrix, the real-time emotion state assessment result is output. When the employee's real-time emotional state assessment result is negative over a continuous period of time, an analysis of the source of emotional influencing factors is triggered. By calculating the Pearson correlation coefficient between the data of each dimension and the real-time emotional state assessment result, the core influencing factors are identified and dynamic adjustment suggestions are generated. Obtain the employee's ID number; the ID number includes English letters and numeric characters; decode the English letters using a preset encoding table to obtain the corresponding decoded numbers; combine the numeric characters in the ID number to obtain multiple different combined numbers; obtain a preset key string, and mutate and adjust the key string based on each of the combined numbers to obtain an adjusted string; obtain a preset array, which is a structured carrier of the key and includes multiple array positions; sequentially insert the characters in the adjusted string into each array position from the starting position of the preset array until all characters are inserted into the preset array to obtain a character array; generate a mutation curve based on the decoded numbers, the mutation curve being constructed using the obtained decoded numbers as the basic parameters in a two-dimensional coordinate system. The horizontal axis represents the array position number, and the vertical axis represents the adjustment amplitude value. Each digit of the decoded number is used as a key node parameter of the curve. A linear interpolation or polynomial fitting algorithm is used to connect the key nodes into a continuous variation curve, and the peaks and troughs of the curve correspond to the adjustment intensity of the array position. The character array is mutated and adjusted based on the variation curve to obtain an adjusted array. A graph is generated based on the attribute features of the multidimensional data, including: converting the attribute features into graph parameters; selecting the graph type according to the number of data dimensions; setting the graph parameters; converting the generated graph into a digital graph matrix through a graph processing algorithm to form a graphic digital carrier strongly associated with the multidimensional data attributes; generating an encryption key based on the graph and the adjusted array for encrypted storage of the multidimensional data.

2. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 1, characterized in that, Basic physiological data includes heart rate data and skin conductance data; work behavior data includes task interaction behavior and document editing pause duration; environmental data includes workstation lighting intensity and office area sound pressure level.

3. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 2, characterized in that, Extracting the fluctuation range of the basic physiological data, including: For heart rate data in basic physiological data, the data is divided into segments by a preset time interval. The difference between the maximum and minimum heart rate values ​​in each data segment is calculated, and the average of the differences of multiple consecutive data segments is used as the heart rate fluctuation amplitude. For the electrodermal activity data, the number of times the peak value of the electrodermal signal occurs per unit time is calculated, and the absolute value of the difference between the number of times and the preset normal range is taken as the fluctuation amplitude of the electrodermal activity. The weighted average of the heart rate fluctuation amplitude and the skin electrical activity fluctuation amplitude is calculated as the fluctuation amplitude of the baseline physiological data.

4. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 2, characterized in that, Extraction of frequency anomalies in work behavior data includes: The frequency of task interaction behavior in statistical work behavior data is used to establish a normal distribution model of the frequency of task interaction of employees in the same working period within the past 30 days. The deviation between the current task interaction frequency and the model mean is calculated, and then the deviation is divided by the model standard deviation to obtain the standardized deviation. If the absolute value of the standardized deviation is greater than the threshold, then the absolute value of the standardized deviation is used as the anomaly of the task interaction frequency. The frequency anomaly of document editing pause duration is calculated, and the frequency anomalies of task interaction and document editing pause duration are weighted and fused to obtain the frequency anomaly of work behavior data.

5. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 2, characterized in that, Extraction of interference coefficients from environmental correlation data includes: The workstation lighting intensity data in the environmental correlation data is compared with the preset comfortable lighting intensity range to calculate the lighting intensity interference coefficient. For the sound pressure level data in the office area, the sound pressure level interference coefficient is obtained by using a preset decibel as the benchmark value and combining it with calculation. The interference coefficient of the environmental correlation data is obtained by weighting and fusing the light intensity interference coefficient and the sound pressure level interference coefficient.

6. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 1, characterized in that, The difference between the initial emotion feature vector and the historical emotion feature benchmark value in the corresponding work scenario is calculated. Combined with a preset emotion influence weight matrix, the real-time emotion state assessment result is output, including: Retrieve historical emotional characteristic baseline values ​​from the employee historical emotional baseline database that are consistent with the current work scenario and the same time period; among them, the historical emotional characteristic baseline values ​​include historical physiological fluctuation amplitude baseline, historical behavioral abnormality baseline, and historical environmental interference coefficient baseline; Physiological deviation rate is calculated based on the fluctuation range of basic physiological data and the historical physiological fluctuation range benchmark; behavioral deviation rate is calculated based on the frequency abnormality of work behavior data and the historical behavioral abnormality benchmark; environmental deviation rate is calculated based on the interference coefficient of environmental correlation data and the historical environmental interference coefficient benchmark. A preset emotion influence weight matrix is ​​invoked, which includes basic weights and scene correction coefficients; the basic weights are corrected based on the scene correction coefficients corresponding to the current work scene to obtain the corrected weights; Based on the adjusted weights, the physiological deviation rate, behavioral deviation rate, and environmental deviation rate are weighted and calculated to obtain the weighted total deviation score. By comparing the weighted total deviation score with the emotion level threshold, a real-time emotion state assessment result is obtained, which includes emotion level, sub-labels, deviation rate of each dimension, and core influence dimension.

7. The dynamic emotional state assessment method based on multidimensional employee data analysis according to claim 1, characterized in that, Before outputting the real-time emotional state assessment results, based on a pre-defined emotional influence weight matrix, the following steps are included: Using the fluctuation range of basic physiological data, the frequency abnormality of work behavior data, and the interference coefficient of environmental correlation data as variables, the correlation coefficient between any two variables is calculated to form a correlation matrix. If the correlation coefficient between any two groups of variables reaches the threshold, they are determined to be strongly correlated dimension groups. The weights of the corresponding two groups of variables are increased by a preset proportion based on the original emotion influence weight matrix, while the weights of other variables are reduced. If the correlation coefficients between all variables are less than the threshold, the original emotion influence weight matrix remains unchanged. In the output of the real-time emotion state assessment results, strongly correlated dimension groups and their corresponding correlation coefficients are labeled.

Images