Intelligent employee performance evaluation method

Abstract

The invention relates to an intelligent employee performance evaluation method, which comprises the following steps of: capturing various information of work through an integrated enterprise internal information system comprising ERP (Enterprise Resource Planning) and HRM (High Rate Management), an external social platform and a work log multi-dimensional data source of employees, and updating the information in real time; a natural language processing NLP technology is applied to intelligently screen, clean and standardize various data formats; integrating the data through a unified evaluation framework, calculating and summarizing multi-dimensional indexes in real time, and forming a preliminary performance evaluation model; analyzing historical data of a company by using a machine learning algorithm, and identifying influence factors of different posts and different service periods on performance evaluation; the weights of evaluation indexes are adjusted in combination with business requirements and work contents of employees, and the fairness and dynamic adaptability of evaluation are enhanced; by analyzing interaction records between the employees and colleagues, leaders and clients, emotion information is extracted, and the emotion states of the employees and the influence on work are recognized.

Description

Technical Field

[0001] The invention relates to an intelligent employee performance evaluation method. Background Art

[0002] While current intelligent employee performance evaluation methods have made some progress in improving efficiency and reducing human bias, they still face significant shortcomings and drawbacks, primarily in areas such as data quality, algorithm transparency, fairness, sentiment analysis accuracy, and employee privacy protection. First, data quality remains a key challenge in intelligent employee performance evaluation. Systems rely on multiple data sources, such as work logs, sales performance, and customer feedback. However, these data are often inconsistent, missing, or of low quality, potentially leading to inaccurate evaluation results. Irregular data collection and organization can even compromise the effectiveness of the entire evaluation system. Second, while machine learning algorithms can process big data, they often suffer from a "black box" problem, making many evaluation processes and results difficult to understand and interpret. This lack of transparency can lead to employee and manager mistrust of evaluation results, undermining employee trust and work motivation, and even leading to unfairness in the evaluation process, especially in the absence of clear standards or unified metrics.

[0003] Furthermore, while some intelligent assessment methods incorporate bias-removal techniques, their effectiveness is often limited. Algorithms are insufficient in eliminating potential biases based on gender, age, and ethnicity, and many implicit biases cannot be effectively identified and eliminated. Existing systems are particularly unable to completely eliminate some implicit workplace biases (such as gender role bias) and team culture bias. The accuracy of sentiment analysis is another issue. While sentiment analysis can identify emotional fluctuations through channels such as employee work logs, emails, and meeting minutes, emotional expression is often influenced by various external factors, making it difficult to fully and accurately capture an employee's emotional state, thus affecting the accuracy of assessment results. Furthermore, sentiment analysis models themselves have limitations, such as an inability to effectively understand complex emotions or situations where emotions are unclearly expressed. Privacy is another significant challenge facing intelligent assessment. Many intelligent assessment systems collect personal employee data, including emails and social media content, which may infringe on employee privacy. Despite data desensitization and encryption, the risk of information leakage still exists, potentially undermining employee trust in the assessment system and impacting their work performance.

[0004] Finally, while existing intelligent assessment methods can provide big data-based analytical results, they still fall short in providing personalized management and development recommendations. Employees' individual needs and career paths vary widely, but existing systems typically only offer general assessment recommendations, making it difficult to truly help employees achieve personalized career growth. For example, for sales personnel, the focus of evaluation may be on sales performance, while for R&D personnel, the emphasis is more on innovation and project management. Existing systems struggle to provide refined, personalized feedback and development recommendations tailored to these diverse characteristics, thus impacting their long-term career development. In summary, while intelligent employee performance assessment methods offer significant advantages in improving efficiency and accuracy, they still face challenges in areas such as data quality, algorithm transparency, fairness assurance, sentiment analysis accuracy, and privacy protection, and urgently require further optimization and improvement. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent employee performance evaluation method to solve some of the drawbacks and shortcomings pointed out in the background technology.

[0006] The present invention solves the above-mentioned technical problems by adopting the following technical solution, which includes the following steps:

[0007] S1. Intelligent data collection and comprehensive evaluation system:

[0008] S1.1. Capture all kinds of work information and update it in real time through multi-dimensional data sources including ERP, HRM, integrated internal enterprise information systems, external social platforms, and employee work logs;

[0009] S1.2. Use natural language processing (NLP) technology to intelligently filter, clean, and standardize various data formats;

[0010] S1.3. Integrate data through a unified evaluation framework, calculate and summarize multi-dimensional indicators in real time, and form a preliminary performance evaluation model;

[0011] S2. Dynamic weight adjustment mechanism based on machine learning:

[0012] S2.1. Use machine learning algorithms to analyze company historical data and identify factors that influence performance evaluation across different positions and business cycles.

[0013] S2.2. Adjust the weights of evaluation indicators based on business needs and employee work content to enhance the fairness and dynamic adaptability of the evaluation;

[0014] S3. Sentiment Analysis and Employee Behavior Prediction:

[0015] S3.1. Analyze employee interactions with colleagues, leaders, and customers to extract emotional information and identify employees' emotional states and their impact on work.

[0016] S3.2. Predict employee work attitudes and behavior trends based on sentiment analysis models and adjust performance scoring models;

[0017] S3.3. By continuously tracking employees' emotional changes, we can adjust the dynamic weight of the assessment in real time and provide recommendations to management;

[0018] S4. Fairness and transparency guarantee mechanism:

[0019] S4.1. Use debiasing techniques to analyze historical assessment data and identify and eliminate potential biases in assessments, including those related to gender and age.

[0020] S4.2. System evaluation criteria and scoring basis should be made public to maintain transparency of the evaluation process;

[0021] S4.3. The system provides administrator privileges and conducts real-time review and adjustment of performance data; management is allowed to conduct review.

[0022] Furthermore, the method of the dynamic weight adjustment mechanism based on machine learning includes:

[0023] Through machine learning modeling, we identify factors influencing employee performance based on historical data. To quantify these influences, we construct a dynamic weighting function W(t) to describe the weight of each performance evaluation indicator at a specific point in time t. This function is dynamically adjusted based on the employee's position type and business cycle, and is expressed as a function of time:

[0024]

[0025] in:

[0026] W(t) represents the comprehensive weight of the evaluation at time point t. i (X i ) is the performance function of the i-th evaluation index, X i Represents the i-th item of data. α i Is the decay factor of each evaluation indicator, reflecting the rate at which each performance indicator decays over time. represents the weighted average of time decay, and γ is the overall time decay factor, which gradually reduces the impact of the evaluation data.

[0027] Furthermore, the method of the dynamic weight adjustment mechanism based on machine learning includes:

[0028] Using reinforcement learning algorithms, the weight of performance evaluation is dynamically adjusted based on business needs and employees' individual work content. By setting the objective function, the weight of each evaluation indicator is updated in real time to maximize the effectiveness of the evaluation model.

[0029]

[0030] in:

[0031] W adjusted (t) represents the performance evaluation weight after dynamic adjustment. It represents the rate of change (gradient) of the weight at time point t, reflecting the change of the evaluation index over time. is the time derivative of the performance indicator adjustment factor δ(t). α and β are adjustment coefficients that control the change gradient during the evaluation process and the rapid adjustment amplitude in the time domain, respectively.

[0032] Furthermore, the method of the dynamic weight adjustment mechanism based on machine learning includes:

[0033] Construct a fairness correction mechanism based on historical data analysis to detect and eliminate bias in the evaluation process; the core of the mechanism is to introduce a fairness correction function S fair (t), dynamically adjust for sources of bias in assessments;

[0034]

[0035] in:

[0036] S fair (t) represents the fairness correction term at time point t. bias is a measure of potential bias related to employee personal characteristics, reflecting the impact of these factors on performance evaluation. γ is the adjustment coefficient of bias influence; λ is the time decay factor. The integral term in the denominator Used to weight time, ensuring that the influence of historical data gradually decreases, while real-time data dominates the bias correction.

[0037] Furthermore, the sentiment analysis and employee behavior prediction include:

[0038] We collect employee sentiment information through various data channels, including work logs, emails, instant messages, meeting minutes, and social media. Using natural language processing technology, we analyze employee sentiment changes, work attitudes, and reactions to company policies and teamwork. We determine whether employee sentiment fluctuations impact performance. Based on these sentiment analysis results, combined with employee historical behavior and work performance data, we use time series analysis to model employee sentiment changes and predict future work attitude and behavior trends.

[0039]

[0040] in:

[0041] S(t): represents the employee's emotional trend score at time t, reflecting the employee's emotional fluctuation trend; α: emotional sensitivity coefficient, indicating the degree of influence of emotional changes on the prediction results. -λt : Attenuation factor, controlling time effect. f(E t ): Sentiment scoring function, which describes the intensity and nature of emotions based on the current employee’s emotional state.

[0042] Furthermore, the sentiment analysis and employee behavior prediction include:

[0043] Provide a new dimension for dynamically adjusting the weight of performance evaluation, and adjust the weight of each evaluation indicator through analysis of employee emotions;

[0044] The core formula of the dynamic weight adjustment mechanism is:

[0045] W adjusted (t) = W0·(1β·|S(t)|)

[0046] in:

[0047] W adjusted (t): Performance evaluation weight adjusted by sentiment analysis. W0: Initial standard evaluation weight. β: Sentiment fluctuation sensitivity coefficient to evaluation weight. |S(t)|: Absolute value of employee sentiment change, reflecting the intensity of employee sentiment fluctuation.

[0048] Furthermore, the sentiment analysis and employee behavior prediction include:

[0049] By continuously tracking employees' emotional states and changes, we maintain real-time awareness of their emotional state. We can identify changes in employee emotional trends, uncover potential mood swings or shifts in work attitudes, generate development recommendations, and help management adjust management strategies and decisions based on employee emotional states.

[0050] The formula for calculating personalized feedback and development suggestions is:

[0051]

[0052] in:

[0053] F personal (t): personalized feedback and suggestion scores at time t; γ: weight coefficient of personalized feedback, which controls the influence of emotional changes on personalized feedback suggestions. -δt : time decay factor; f recommend(S(t)): Personalized feedback function generated based on sentiment analysis and employee sentiment trajectory.

[0054] The present invention has the following beneficial effects:

[0055] By combining sentiment analysis, machine learning, a dynamic weighting mechanism, and a fairness correction mechanism, the intelligent employee performance evaluation method significantly improves the accuracy, fairness, and real-time nature of employee performance evaluations. First, through multi-dimensional data collection and natural language processing technology, the system can comprehensively collect employee work data, emotional fluctuations, and behavioral trends, track changes in employees' emotional states in real time, promptly identify potential mood swings or changes in work attitudes, and dynamically adjust the weights of various performance evaluation indicators based on these changes. Second, based on historical data analysis, the system can automatically identify and eliminate potential biases in the evaluation process, such as gender and age, thereby ensuring the fairness and objectivity of the evaluation results.

[0056] Furthermore, sentiment analysis, combined with employee behavior prediction, can effectively predict employees' future work attitudes and performance trends, helping management implement targeted interventions at the early stages of employee mood swings and preventing the impact of negative emotions on work efficiency and teamwork. Through personalized feedback and development recommendations, the system provides each employee with a tailored career development plan, improving their job satisfaction and loyalty. Overall, the method of the present invention not only optimizes the performance evaluation process but also provides more refined and scientific decision-making support for employee development and company management, thereby enhancing the effectiveness and fairness of employee performance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of the intelligent employee performance evaluation method of the present invention.

[0058] Figure 2 This is a flow chart of the dynamic weight adjustment mechanism based on machine learning in the present invention.

[0059] Figure 3 This is the flow chart of sentiment analysis and employee behavior prediction in the present invention. DETAILED DESCRIPTION

[0060] Combined with attachment Figure 1 The process of intelligent employee performance evaluation method is characterized by including the following steps:

[0061] The S1 intelligent employee performance evaluation method uses a variety of technical means to achieve intelligent assessments of employee performance, ensuring fairness, accuracy, and real-time performance. First, within the intelligent data collection and comprehensive evaluation system, the system integrates internal enterprise information systems such as ERP (Enterprise Resource Planning) and HRM (Human Resource Management), as well as external data sources such as social media platforms and employee work logs to comprehensively capture employee work information. The ERP system provides data on employee work tasks, productivity, and project progress, while the HRM system records basic data such as employee attendance, work hours, and salary. Data from social media platforms helps the system analyze interactions between employees and colleagues and leaders, further supplementing soft indicators such as their teamwork skills. Work logs provide detailed records of employee work and personal work emotions for evaluation. Throughout this process, natural language processing (NLP) technology is used to intelligently filter, cleanse, and standardize various data formats to ensure data validity and consistency. NLP technology processes natural language content in employee emails, meeting minutes, and instant messages, extracting key information and removing redundant or irrelevant content. After data collation, all data is integrated through a unified evaluation framework. The system calculates and summarizes various multi-dimensional evaluation indicators in real time, including task completion, teamwork, and customer satisfaction, to form a preliminary performance evaluation model. This model can be updated in real time based on different business needs and evaluation cycles, providing accurate employee performance data support.

[0062] S2's dynamic weight adjustment mechanism based on machine learning plays a crucial role. Through machine learning algorithms, the system can conduct in-depth analysis of the company's historical employee performance data and identify the factors that influence employee performance evaluations across different positions and business cycles. For example, during the sales cycle, sales personnel's performance data (such as sales volume and customer feedback) may carry a higher weight, while during the R&D cycle, R&D personnel's innovation capabilities, project delivery speed, and teamwork performance will become important evaluation criteria. Based on these analysis results, combined with business needs and employee work content, the system can intelligently adjust the weights of evaluation indicators. This dynamic adjustment ensures that the evaluation criteria can adapt to different business scenarios, avoiding the use of a single weight to evaluate all employees, thereby improving the fairness and dynamic adaptability of the evaluation. Using machine learning algorithm training, the system can continuously optimize these weight adjustment strategies, making each evaluation more accurate and in line with current business needs.

[0063] In the S3 sentiment analysis and employee behavior prediction phase, the system analyzes employee interactions with colleagues, supervisors, and clients to extract emotional information, identifying employees' emotional states and their impact on work. For example, the system can analyze whether negative emotions are evident in employee interactions with colleagues, or whether employees respond positively to work content and company policies. By applying sentiment analysis models, the system can predict future employee work attitudes and behavioral trends, such as whether emotional fluctuations will affect work efficiency or teamwork. Therefore, sentiment analysis not only provides an immediate assessment of an employee's current emotional state but also, by learning from historical data and behavioral patterns, predicts long-term emotional fluctuations and behavioral changes. Furthermore, the system dynamically adjusts performance scoring models based on sentiment analysis results. If sentiment analysis indicates an employee's mood is becoming unstable, the system can adjust the evaluation weighting to prioritize soft indicators such as teamwork and emotional stability, thereby preventing emotional fluctuations from interfering with other performance indicators. Furthermore, by continuously tracking employees' emotional states, the system can monitor emotional changes in real time and dynamically adjust the weighting used in evaluations. Emotion-based assessments not only make performance evaluations more comprehensive, but also provide management with timely feedback, helping them take targeted management measures, such as providing psychological support or adjusting work content.

[0064] The introduction of the S4 fairness and transparency guarantee mechanism ensures fairness and openness throughout the performance evaluation process. The system utilizes debiasing technology to conduct in-depth analysis of historical evaluation data, identifying and eliminating potential biases, including those based on gender, age, and ethnicity. These biases can lead to unfair performance evaluations, impacting employee motivation and company atmosphere. The system uses an algorithmic model to monitor and correct these biases in real time, ensuring that all data in employee performance evaluations accurately reflects actual performance and is unaffected by external factors. To further enhance the fairness of evaluations, the system also makes evaluation criteria and scoring basis public, ensuring that employees understand the basis and criteria for their evaluations. This open and transparent evaluation process enhances employee trust in performance evaluations and ensures high transparency in management decisions. Furthermore, the system provides administrators with real-time review and adjustment of performance data, allowing management to ensure accuracy and fairness by reviewing the evaluation process. Management can also review performance data as needed to ensure that each evaluation meets company fairness requirements and avoid unfair results caused by human error or manipulation.

[0065] Example 1:

[0066] Combined with attachment Figure 2As shown in the process, a manufacturing company implemented an intelligent employee performance evaluation method. This method uses machine learning algorithms, dynamic weighting functions, sentiment analysis, and other technical means to comprehensively assess employee performance and achieve continuous optimization and refined management. To illustrate the feasibility of this solution, the company used an intelligent evaluation method modeled using machine learning algorithms for its annual performance evaluation. The evaluation dimensions included indicators such as task completion, team collaboration, customer satisfaction, and innovation.

[0067] First, during the historical data collection phase, the system collected employee performance data from the past year through multi-dimensional data sources such as the ERP system, HRM system, and employee work logs. This data includes not only quantitative indicators (such as sales and number of tasks completed), but also qualitative indicators (such as teamwork, customer feedback, and leadership comments). This data is intelligently filtered, cleaned, and standardized using natural language processing (NLP) technology to remove noise and retain key information. For example, data related to customer feedback and teamwork is extracted through sentiment analysis models to obtain employee emotional fluctuations over the past year and their impact on work performance. Then, all data is unified and integrated into an evaluation framework. The system uses machine learning algorithms to learn the correlations between various data and identify the factors that affect employee performance in different positions and different business cycles.

[0068] In order to quantify these impacts, the system designed a dynamic weighting function W(t) to describe the comprehensive weight of each indicator in the performance evaluation at a certain point in time t. Specifically, this function will be dynamically adjusted according to the employee's job type and the current business cycle, and will be reflected in the form of a function of time. Assume that employee A holds a sales position in 2024 and experiences a sales peak and off-season in the middle of the year. These changes will affect the weight distribution of evaluation indicators. During peak sales season, the weight of sales and customer feedback will increase, while in the off-season, employees' teamwork and innovation capabilities will be given higher weights. The mathematical expression of this dynamic weighting function is:

[0069]

[0070] Where W(t) is the comprehensive evaluation weight at time point t, representing the employee's overall performance evaluation score at a specific time point. To illustrate a specific application, assume that during the evaluation of employee A, the system sets the following parameters:

[0071] f1(X1)=80 represents employee A's sales score, and X1 is the sales data.

[0072] f2(X2)=70 represents the teamwork score of employee A, and X2 is the quantitative data of teamwork.

[0073] f3(X3)=75 represents the customer satisfaction score, and X3 is the customer feedback score.

[0074] f4(X4)=85 represents employee A’s score in innovation ability, and X4 is the completion status of the innovation project.

[0075] In addition, the setting system sets a different attenuation factor α for each indicator i , these decay factors reflect the rate at which each performance indicator gradually weakens over time. The specific parameter settings are as follows:

[0076] α1 = 0.1, the weight of sales volume decays faster because sales performance is more dependent on short-term goals.

[0077] α2 = 0.05, the attenuation factor of teamwork is small because teamwork is often the result of long-term performance.

[0078] α3=0.08, the attenuation factor of customer satisfaction is between the two.

[0079] α4=0.12, the attenuation factor of innovation capability is relatively large, reflecting that innovation results may be gradually replaced over time.

[0080] The denominator of this formula, that is, the integral term represents the weighted average of time decay, and γ is the overall time decay factor, which controls the influence of historical data on the final evaluation results. In practice, setting γ = 0.05 means that the time decay rate of all evaluation indicators is a fixed low value. The value calculated by this integral term is:

[0081]

[0082] Assuming t is the evaluation time point, the calculation result will give an attenuation weighted value corresponding to the time period.

[0083] The dynamic weighting function W(t) combines different job positions, business cycles, and employees' emotional states to calculate a comprehensive employee evaluation score in real time at a specific point in time. For example, during peak sales season, the system prioritizes employee sales and customer satisfaction data, giving these data a higher weight. Meanwhile, during off-season, the system prioritizes teamwork and innovation. This adjustment ensures the evaluation model's adaptability, allowing it to dynamically adjust its focus based on the employee's specific situation, preventing a single metric from overly influencing an employee's performance score.

[0084] The system continuously optimizes this evaluation framework through machine learning algorithms, identifying potentially important factors influencing employee performance, making the evaluation more accurate and aligned with actual business needs. Throughout the evaluation process, by adjusting the weights of different indicators, the system ensures that each indicator has a reasonable impact on employee performance at different time points, thereby ensuring the fairness and accuracy of the evaluation results.

[0085] As the company implements intelligent employee performance evaluation methods, a reinforcement learning algorithm is introduced to dynamically adjust the weights of various performance evaluation indicators. Specifically, through the reinforcement learning model, the system can adjust the performance evaluation weights in real time based on the employee's individual work content and current business needs. This dynamic adjustment mechanism ensures more accurate evaluation results and maximizes the effectiveness of the evaluation model. For example, suppose an employee assumes dual responsibilities of sales and project management during a certain period, and the company conducts a large-scale product promotion during this period. In this case, the evaluation system will dynamically adjust the weights related to sales and project management based on the employee's specific work content, ensuring the fairness and adaptability of the evaluation criteria.

[0086] The core of the system is to update the weights of performance evaluation indicators by setting an objective function. The objective function uses a reinforcement learning algorithm to adjust the performance evaluation model so that the final evaluation results can best reflect the employee's performance within a specific time period. During implementation, the system set the following mathematical formula to describe this dynamic adjustment mechanism:

[0087]

[0088] W adjusted (t): represents the performance evaluation weight adjusted by the reinforcement learning algorithm at time point t. This adjusted weight reflects the optimization adjustment of the evaluation indicator based on current business needs and employee work content.

[0089] The gradient of the performance evaluation weight at time point t reflects the change trend of each evaluation indicator over time and modifies the evaluation results at different time points.

[0090] The performance index adjustment factor δ(t) represents the rate of change over time. This factor fine-tunes the performance evaluation weight by examining changes in employee work content, ensuring that the evaluation model responds immediately to actual work conditions.

[0091] α and β are adjustment coefficients that control the gradient of weight changes and the magnitude of temporal adjustments during the evaluation process, respectively. Specifically, α controls the influence of the indicator's rate of change, while β reflects the evaluation system's response speed to rapid changes in the temporal domain.

[0092] To more intuitively demonstrate the application of this solution, let's assume that Employee A participated in a sales assignment in mid-2024 and transitioned into a project management role in the second half of the year. To ensure fair evaluation, the system dynamically adjusted the weights of the sales and project management evaluation indicators based on the specific work content of the two positions.

[0093] set up:

[0094] During the sales task, employee A's sales index f1(X1)=100, teamwork index f2(X2)=80, customer satisfaction f3(X3)=85, and innovation ability f4(X4)=70.

[0095] During the project management phase, employee A’s innovation ability received more attention, and innovation ability f4(X4) was set to 90, while customer satisfaction decreased slightly due to the impact of the project completion phase, and customer satisfaction f3(X3) was set to 75.

[0096] The attenuation factors are set to α1 = 0.1, α2 = 0.05, α3 = 0.08, and α4 = 0.12 to reflect the rate of change of each evaluation indicator over time.

[0097] During the sales mission, the system enhanced the weights of sales, teamwork, and customer satisfaction indicators, setting α = 0.2, thereby increasing the influence weights of these indicators; while in the project management stage, the system increased the weights of indicators related to innovation capabilities and reduced the weights related to customer satisfaction.

[0098] According to the formula, the change of the calculated performance weight over time can be expressed as:

[0099]

[0100] For example, during the peak sales period of employee A, the system will automatically increase the weight of sales and customer satisfaction, and update the evaluation model in real time to reflect this change. and By adjusting the formula, we can conclude that at time point t, during the sales task period, the comprehensive weight of performance evaluation is:

[0101]

[0102] This calculated weight adjustment indicates that during peak sales periods, the performance evaluation model will be more sensitive to sales-related indicators than during project management, ensuring the maximum effectiveness of the evaluation model during a specific business cycle.

[0103] When employee A transfers to the project management role, the system will optimize the evaluation indicators again by adjusting the weights. The evaluation weight of innovation ability is set to be increased during this stage, while the weight of customer satisfaction is reduced due to the changes in the project. Assuming that during the project management period, innovation ability f4(X4)=90, the system will update the performance weight again to reflect the employee's performance in the new role. During this period, set and The calculated performance weight is:

[0104]

[0105] Through this dynamic adjustment mechanism, the system can track employee work content and position changes in real time and optimize evaluation weights based on business needs. This dynamic adjustment solution, based on reinforcement learning, not only maximizes the effectiveness of the evaluation model but also ensures that each employee's performance at different stages of their work is fairly evaluated.

[0106] In this manufacturing company, the intelligent improvement of the employee performance evaluation system is not only reflected in dynamic weighting and sentiment analysis, but also involves the introduction of a fairness correction mechanism to ensure that the evaluation process is not affected by any potential bias. This mechanism is based on historical data analysis and uses the fairness correction function S fair (t) ensures fairness by detecting and eliminating biases based on gender, age, and race that may exist during the evaluation process. Consider a company whose performance evaluation process has previously seen unfair impacts on employee performance scores due to gender or age. This is particularly true in the sales department, where female or older employees may receive lower performance evaluation scores than male or younger employees with equivalent performance. To correct for this bias, the system incorporates a fairness correction mechanism that leverages historical data and a time decay function to correct for biases in the evaluation process.

[0107] The mathematical expression of the fairness correction function is:

[0108]

[0109] Formula explanation:

[0110] S fair (t): represents the fairness correction term in the evaluation process at time point t. This correction term adjusts the evaluation process based on historical data and the characteristics of the current employee to reduce the impact of bias.

[0111] Xbias : It is a measure of potential bias related to employee personal characteristics, reflecting the impact of these characteristics (such as gender, age, race, etc.) on performance evaluation. For example, if an employee shows high leadership ability in project management due to his or her age, but is still underestimated in performance evaluation, then X bias It is a positive value, indicating the influence of this bias.

[0112] γ: This is the bias adjustment coefficient, which controls the strength of the bias's impact on the evaluation results. The value range of γ is set to 0.05≤γ≤0.2. The larger the value, the greater the impact of bias on the correction.

[0113] λ is a time decay factor that controls the impact of historical data on fairness corrections. Setting λ = 0.1 means that the influence of bias will gradually decrease over time, giving more weight to corrections based on current data.

[0114] Integral term in the denominator Time-weighted data is used to ensure that the influence of historical data on the correction gradually decreases, and real-time data plays a greater role in bias correction. This design is to ensure that the system can respond quickly to new data.

[0115] Practical application scenarios:

[0116] In 2024, employee B excels in project management. Despite her leadership role and the completion of multiple cross-departmental collaborative projects, her performance evaluation scores are lower than those of other male leadership employees due to gender bias. Through the fairness correction mechanism, the system first identifies a possible bias factor: gender. The system calculates X based on the employee's historical data. bias , reflecting the unfair impact that employee B may have suffered in past evaluations due to his gender, and setting this value to 0.2. Then, the system calculates the correction factor S fair (t), and dynamically adjust the evaluation criteria. Assuming that the system is corrected during the evaluation process in March 2024, with values ​​of λ = 0.1 and γ = 0.15, then according to the above formula, the fairness correction term is:

[0117]

[0118] Through actual calculation, the denominator integral calculation result is set to 2.7, and the numerator integral calculation result is 0.5, and the correction term S is obtained. fair The result of (3) is:

[0119]

[0120] This correction indicates that the impact of gender bias on employee B's performance rating has been reduced. By adjusting the evaluation results, the system increased recognition of her teamwork and leadership skills, thereby improving her final rating.

[0121] Furthermore, the system gradually reduces the influence of past data based on the time decay factor λ, and relies more on real-time data. For example, after employee B completes a new project and receives positive feedback from a client, the system continuously tracks her work results and sentiment changes, updating her evaluation score to more closely reflect her current performance, rather than being overly influenced by historical biases.

[0122] This process not only ensures evaluation fairness but also improves the system's adaptability to employees' dynamic performance. In the scenario described above, employee B's evaluation score improved after undergoing gender bias correction, fully demonstrating the effectiveness of the fairness correction mechanism in performance evaluation. This not only ensures a fair evaluation of employee B's performance, but also enhances the transparency and fairness of the system, making internal performance management more scientific and humane. In this way, the practical application of the fairness correction mechanism ensures the transparency, accuracy, and fairness of employee evaluations, eliminates potential bias, and increases employee trust and satisfaction with the performance management system.

[0123] Example 2:

[0124] Combined with attachment Figure 3 To improve the scientific nature and accuracy of employee performance management, a large technology company implemented an intelligent employee performance evaluation method. This method focuses on dynamically assessing employee work attitudes and behavioral trends through multi-channel data collection and sentiment analysis. The company's human resources department noted that employee emotional fluctuations and reactions to company policies significantly impacted their work performance. This was particularly true during high-pressure project phases and business changes, when employee mood swings could directly impact teamwork and task completion efficiency. Therefore, the company adopted a dynamic performance evaluation method based on sentiment analysis. This method, using natural language processing (NLP) technology, collects and analyzes employee emotional information to promptly understand employee emotional changes and, combined with historical work performance data, predict future work attitudes.

[0125] During implementation, the system collects employee sentiment information through various data channels, including work logs, emails, instant messages, meeting minutes, and social media. Each employee's sentiment data is collated and fed into a sentiment analysis model. The system then uses natural language processing to analyze employee work logs and meeting minutes to identify emotional tendencies. For example, if an employee uses negative sentiment words (such as "bored" or "disappointed") in their work log, the system calculates the employee's current emotional state based on the frequency and intensity of these sentiment words. Furthermore, the system scans instant messages and meeting minutes between employees and colleagues and supervisors to assess their attitudes toward teamwork and their reactions to company policies. This sentiment information is automatically categorized and quantified to generate a sentiment score reflecting the employee's emotional fluctuations.

[0126] For example, let's assume that employee C's emotional state changes significantly while participating in a major product launch project. Initially, employee C demonstrates high enthusiasm and communicates effectively with the team. However, as the project progresses, employee C's work logs begin to display more negative emotions, such as "feeling very stressed" and "I'm unsure if I can complete the task on time." These emotional fluctuations are also reflected in instant messages, where employee C gradually expresses anxiety in his communications with team members and even expresses frustration with the company's project progress. By processing this emotional data, the system analyzes employee C's emotional changes and calculates an emotional score, reflecting the impact of his emotional fluctuations on his work attitude and performance.

[0127] At this point, the system models the employee's emotional trend through the formula S(t):

[0128]

[0129] Formula explanation:

[0130] S(t) represents the employee's emotional trend score at time t, reflecting the employee's emotional fluctuations. In the case of employee C, S(t) will show a gradual shift from initial positive emotions to negative emotions.

[0131] α: Emotional sensitivity coefficient, which indicates the degree to which emotional changes affect the forecast results. Setting α = 0.8 indicates that emotional fluctuations have a significant impact on performance evaluation, especially at critical moments in a project.

[0132] e -λt: A decay factor used to control for time effects, ensuring that the influence of older sentiment data gradually decreases. Setting λ = 0.05 means that the influence of historical sentiment data gradually decreases over time, placing greater emphasis on the impact of current sentiment on the evaluation.

[0133] f(E t ): Emotional scoring function, which describes the intensity and nature of emotions based on the current emotional state of employees. Assume that at time point t0, employee C’s emotional score f(E t ) is calculated as a negative value (e.g., -0.4), which reflects the negative emotions (e.g., stress, anxiety) of employees at the current point in time.

[0134] For example, at the mid-term point t = 3 (i.e., the third month of the project), employee C's emotional score is -0.4, which means that his current emotional fluctuations are in a negative state and have a negative impact on his work attitude. After substituting these data into the formula, the system calculates the corrected emotional trend score S(3) to be -0.2, which means that the negative impact of employee C's emotional fluctuations on his performance is reflected in real time, and the system has adjusted the relevant weights for subsequent evaluations. The emotional impact weight calculated at this time is set to:

[0135] S adjusted =W0·(1+α·S(3))=0.8·(1+0.8·(-0.2))=0.8·(10.16)=0.672

[0136] This means that, based on the correction of emotional fluctuations, employee C's performance evaluation score has been adjusted downward, reflecting the impact of his emotional fluctuations on performance. In this way, the system tracks employees' emotional changes in real time and converts these changes into adjustments to evaluation indicators, ensuring that employees' performance evaluations are based not only on workload and completion, but also on changes in their emotional and psychological state.

[0137] In this example, at a large technology company, an intelligent employee performance evaluation method was applied to optimize the employee work evaluation process, using sentiment analysis to provide refined predictions of employees' work attitudes and behavioral trends. The company discovered that employee emotional fluctuations significantly impacted their work performance, especially during high-pressure project management phases, where these fluctuations could directly impact work efficiency and teamwork. Therefore, performance evaluations no longer rely solely on quantitative metrics such as sales or task completion; they also incorporate employee emotional fluctuations as a key factor in adjusting performance scores. In this case, sentiment analysis, combined with a dynamic weighting adjustment mechanism, became a new dimension in employee performance evaluation. Consider employee D, who experienced significant emotional fluctuations midway through a major project due to difficulties in teamwork, project schedule pressure, and personal emotional fluctuations. To promptly identify and adjust his performance evaluation, the system used sentiment analysis to capture employee D's emotional changes and dynamically adjusted the evaluation weights related to teamwork and task completion.

[0138] The company's dynamic weight adjustment mechanism is based on sentiment analysis results and adjusts performance evaluation weights through the following core formula:

[0139] W adjusted (t) = W0·(1+β·|S(t)|)

[0140] Formula explanation:

[0141] W adjusted (t): Performance evaluation weight adjusted through sentiment analysis. This weight is dynamically adjusted based on employee sentiment changes, reflecting the impact of employee sentiment fluctuations on performance evaluation.

[0142] W0: The initial standard evaluation weight, which represents the employee's performance score when there is no emotional influence.

[0143] β: The sensitivity coefficient of emotional fluctuations to the evaluation weight. The value range of β is set to 0.1≤β≤0.3, which means that the system allows emotional fluctuations to affect performance scores within a certain range. A higher β indicates greater sensitivity.

[0144] |S(t)|: The absolute value of employee sentiment change, reflecting the intensity of employee emotional fluctuations. Using a sentiment analysis model, the system can calculate and quantify employee emotional fluctuations in real time, generating a score reflecting their emotional state.

[0145] Suppose employee D, while participating in a critical project, experiences significant emotional fluctuations due to external pressure and internal team conflict. The system's sentiment analysis indicates that employee D's mid-project sentiment score, S(t), is -0.5, indicating low mood and significant emotional fluctuations during this period, potentially negatively impacting work performance. The sentiment score reflects the intensity of an employee's emotional shifts; larger values ​​indicate more intense emotional fluctuations. The system then adjusts the employee's performance evaluation weight based on this emotional fluctuation.

[0146] Employee D's initial performance evaluation weight, W0, is set to 0.75, indicating that his performance in project management and teamwork is initially assessed as average, in line with expectations. However, considering employee D's emotional fluctuations during the project, the system uses β = 0.2 (representing a moderate sensitivity of emotional fluctuations to the evaluation results) and the score S(t) = -0.5 derived from the sentiment analysis (i.e., a negative sentiment score, reflecting the employee's emotional fluctuations at that time) to calculate the dynamically adjusted performance evaluation weight:

[0147] W adjusted (t) = 0.75·(1+0.2·|-0.5|)

[0148] W adjusted (t)=0.75·(1+0.2·0.5)=0.75·1.1=0.825

[0149] The performance evaluation weight adjusted by sentiment analysis is 0.825. This means that due to employee D's emotional fluctuations, the system appropriately increased the weights associated with teamwork and task completion in his performance score, reflecting the impact of emotional fluctuations on performance. The impact of employee D's emotional state on the evaluation is reflected in the increased scores for teamwork and task completion.

[0150] At the end of the project, assume that employee D's mood has recovered and the emotional score S(t) has turned positive to 0.3, indicating that the emotional fluctuations have stabilized. The system will adjust the evaluation weight accordingly. At this stage, the emotional sensitivity coefficient β remains unchanged, and the system will use the new emotional score to recalculate the performance weight. The calculation at this time is:

[0151] W adjusted (t) = 0.75·(1+0.2·|0.3|)

[0152] W adjusted (t)=0.75·(1+0.2·0.3)=0.75·1.06=0.795

[0153] Therefore, if employee D's emotions recover, the adjusted performance weight is 0.795, slightly lower than the previous adjustment value, but still higher than the initial assessment value, reflecting the actual impact of emotional fluctuations on the assessment. As the employee's emotions further stabilize, the system will continue to track emotional changes and adjust the assessment weight, ensuring that the final performance evaluation is more fair and accurate.

[0154] This combination of sentiment analysis and dynamic weighting adjustments not only makes performance evaluations more flexible but also reflects employee emotional fluctuations in real time, thereby optimizing the employee work environment and performance management. By adjusting performance weights based on emotional fluctuations, the evaluation system more accurately reflects employees' actual performance, rather than relying solely on fixed indicators or historical data, thereby improving employee job satisfaction and performance.

[0155] As this large technology company implements an intelligent employee performance evaluation approach, sentiment analysis not only assesses current employee performance but also plays a crucial role in real-time monitoring and predicting changes in employee emotional states. The company has recognized that employee emotional fluctuations can directly impact their work attitude, productivity, and teamwork. Therefore, it applies sentiment analysis to its ongoing employee tracking and generates personalized feedback and development recommendations based on these changes. This approach not only allows the company to detect changes in employee emotional states in real time but also provides personalized recommendations based on these changes, helping management make more targeted management decisions. Consider a large-scale product development project where employee E's emotional state has a significant impact on project progress. Initially, employee E was emotionally stable, had a positive work attitude, and was highly productive. However, as the project progressed, employee E began to experience emotional fluctuations due to heavy workloads and high pressure. Using a sentiment analysis system, employee E's emotional fluctuations were monitored and analyzed in real time. The system then calculated personalized feedback and development recommendations based on his emotional state, helping management make appropriate adjustments. The company employed the following formula for personalized feedback and development recommendations based on sentiment analysis:

[0156]

[0157] Formula explanation:

[0158] F personal (t): The employee's personalized feedback and suggestion score at time point t. This score reflects the employee's current emotional state and the impact of emotional changes on their work attitude and behavioral trends, generating personalized feedback and development suggestions.

[0159] γ: The weighting factor for personalized feedback, controlling the degree to which sentiment changes influence feedback suggestions. Setting the value of γ to 0.1 ≤ γ ≤ 0.3 means that the influence of sentiment changes on feedback can be adjusted within a reasonable range. A higher γ will make sentiment fluctuations account for a greater proportion of feedback suggestions.

[0160] e -δt : A time decay factor, indicating that as time passes, the influence of data from more distant time points on the current feedback gradually decreases. Setting δ = 0.05 means that the influence of sentiment data decays relatively steadily over time, ensuring that real-time data dominates feedback suggestions.

[0161] f recommend (S(t)): A personalized feedback function generated based on sentiment analysis and employee emotional trajectory. It generates corresponding feedback suggestions based on the employee's emotional state at a specific time point, reflecting the impact of employee emotional fluctuations on work performance. In the case of employee E, the system obtains employee E's emotional score S(t) = 0.2 at the beginning of the project, and the feedback function generated based on the emotional state is f recommend (S(t)) = 0.9, indicating that when the affective state is more positive, employees’ work attitude and productivity remain at a higher level.

[0162] As the project enters the middle stage, employee E's emotional score drops sharply due to increased work pressure and task complexity. The emotional score S(t) reaches -0.3, reflecting negative emotional fluctuations, which has an adverse impact on the employee's work attitude. At this stage, the system calculates personalized feedback suggestions and provides feedback based on the formula:

[0163]

[0164] Set γ = 0.2 and f recommend (S(t)) = 0.6, indicating that emotional fluctuations have a greater negative impact on work attitudes, and the strength of feedback suggestions is also correspondingly stronger. The calculation results are:

[0165]

[0166] The feedback indicated that changes in employee E's emotional state significantly impacted their work performance. The system then provided appropriate adjustments, such as reducing task burdens, providing psychological support, and encouraging teamwork. These results indicate that emotional fluctuations gradually increase the impact on assessments and development recommendations, ensuring that management can make appropriate adjustments in a timely manner to help employees recover and maintain their work efficiency.

Claims

1. Intelligent employee performance evaluation method, characterized by The following steps are involved: S1. Intelligent data collection and comprehensive evaluation system: S1.

1. Capture all kinds of work information and update it in real time through multi-dimensional data sources including ERP, HRM, integrated internal enterprise information systems, external social platforms, and employee work logs; S1.

2. Use natural language processing (NLP) technology to intelligently filter, clean, and standardize various data formats; S1.

3. Integrate data through a unified evaluation framework, calculate and summarize multi-dimensional indicators in real time, and form a preliminary performance evaluation model; S2. Dynamic weight adjustment mechanism based on machine learning: S2.

1. Use machine learning algorithms to analyze company historical data and identify factors that influence performance evaluation across different positions and business cycles. S2.

2. Adjust the weights of evaluation indicators based on business needs and employee work content to enhance the fairness and dynamic adaptability of the evaluation; S3. Sentiment Analysis and Employee Behavior Prediction: S3.

1. Analyze employee interactions with colleagues, leaders, and customers to extract emotional information and identify employees' emotional states and their impact on work. S3.

2. Predict employee work attitudes and behavior trends based on sentiment analysis models and adjust performance scoring models; S3.

3. By continuously tracking employees’ emotional changes, we can adjust the dynamic weight of the assessment in real time and provide recommendations to management; S4. Fairness and transparency guarantee mechanism: S4.

1. Use debiasing techniques to analyze historical assessment data to identify and eliminate potential assessment biases, including those related to gender and age. S4.

2. System evaluation criteria and scoring basis should be made public to maintain transparency of the evaluation process; S4.

3. The system provides administrator privileges and conducts real-time audit and adjustment of performance data, allowing management to review.

2. The intelligent employee performance evaluation method according to claim 1 is characterized in that The method of the dynamic weight adjustment mechanism based on machine learning includes: Through machine learning algorithm modeling, we identify factors that affect employee performance based on historical data. We quantify the impact and construct a dynamic weighting function W(t) to describe the weight of each indicator in the performance evaluation at a unit time point t. This function is dynamically adjusted based on the employee's position type and business cycle, and expressed as a function of time: in: W(t) represents the comprehensive weight of the evaluation at time point t; f i (X i ) is the performance function of the i-th evaluation index, X i Represents the i-th data; α i is the attenuation factor of each evaluation indicator, reflecting the rate at which each performance indicator decays over time; the integral term in the denominator represents the weighted average of time decay, and γ is the overall time decay factor, which gradually reduces the impact of the evaluation data.

3. The intelligent employee performance evaluation method according to claim 2 is characterized in that The method of the dynamic weight adjustment mechanism based on machine learning includes: Using reinforcement learning algorithms, the weight of performance evaluation is dynamically adjusted according to business needs and employees' individual work content. By setting an objective function, the weight of each evaluation indicator is updated in real time to maximize the effectiveness of the evaluation model. in: W adjusted (t) represents the performance evaluation weight after dynamic adjustment; It represents the rate of change of weight at time point t, reflecting the change of evaluation index over time; is the time derivative of the performance indicator adjustment factor δ(t); α and β are adjustment coefficients, which control the change gradient in the evaluation process and the rapid adjustment amplitude in the time domain respectively.

4. The intelligent employee performance evaluation method according to claim 3 is characterized in that The method of the dynamic weight adjustment mechanism based on machine learning includes: Construct a fairness correction mechanism based on historical data analysis to detect and eliminate bias in the evaluation process; the core of the mechanism is to introduce a fairness correction function S fair (t), dynamically adjust for sources of bias in assessments; in: S fair (t) represents the fairness correction term at time point t; X bias is a measure of potential bias related to employee personal characteristics, reflecting the impact of these factors on performance evaluation; γ is the adjustment coefficient of bias influence; λ is the time decay factor; the integral term in the denominator Used to weight time, the influence of historical data is gradually reduced, while real-time data dominates the bias correction.

5. The intelligent employee performance evaluation method according to claim 1 is characterized in that The sentiment analysis and employee behavior prediction include: Employee emotional information is collected through multiple data channels, including work logs, emails, instant messages, meeting minutes, and social media. Natural language processing technology is used to analyze employees' emotional changes, work attitudes, and reactions to company policies and teamwork. It is determined whether employees' emotional fluctuations will affect performance. Based on these emotional analysis results, combined with employees' historical behavior and work performance data, time series analysis is used to model employees' emotional changes in order to predict their future work attitudes and behavioral trends.

6. The intelligent employee performance evaluation method according to claim 5 is characterized in that The sentiment analysis and employee behavior prediction include: providing a new dimension for dynamically adjusting the weight of performance evaluation, and adjusting the weight of each evaluation indicator through analysis of employee sentiment; The core formula of the dynamic weight adjustment mechanism is: W adjusted (t)=W0·(1β·|S(t)|) in: W adjusted (t) is the performance evaluation weight after adjustment through sentiment analysis; W0 is the initial standard evaluation weight; β is the sensitivity coefficient of sentiment fluctuation to the evaluation weight; |S(t)| is the absolute value of employee sentiment change, reflecting the intensity of employee sentiment fluctuation.

7. The intelligent employee performance evaluation method according to claim 6 is characterized in that The sentiment analysis and employee behavior prediction include: By tracking employees' emotional states and changes, we maintain real-time awareness of their emotional states; identify changes in employees' emotional development trends, discover potential mood swings or changes in work attitudes, generate development suggestions, and calculate management's adjustment of management strategies and decision-making data based on employees' emotional states.

Images