Talent value dynamic evaluation system based on price-increased gyroscope value model

Abstract

The invention belongs to the technical field of talent evaluation, provides a talent value dynamic evaluation system based on a price-increased gyroscope value model, and aims to solve the problems of lag, oscillation, insufficient adaptive capacity and the like caused by fixed setting of a noise covariance Q and an observation noise covariance R in a Kalman filtering process in traditional evaluation. The system comprises a basic adjustment factor module for obtaining a multi-source environment data construction skill popularity index and generating a basic adjustment factor of Q; the weight optimization module is used for establishing an observation data credibility model and calibrating and optimizing the weight of R; and the dynamic optimization module is used for realizing self-adaptive adjustment of the process noise covariance Q / observation noise covariance R by comparing the first-order difference between the skill heat index and the skill weight, analyzing the Kalman filtering residual error and dynamically balancing the process noise covariance Q and the observation noise covariance R, so that the accuracy and adaptability of talent evaluation are improved, and the accuracy and the adaptability of talent evaluation are improved. The industrial trend and the enterprise demand are met.

Description

Technical Field

[0001] This invention belongs to the field of talent evaluation technology, specifically a dynamic talent value evaluation system based on the price-increasing gyroscope value model. Background Technology

[0002] In dynamic talent value assessment, the Kalman filter algorithm is often used to adjust skill weights in real time. However, it relies on the reasonable setting of process noise covariance Q and observation noise covariance R, and there are many problems in practical applications.

[0003] If the process noise covariance Q is set too small, the system will over-rely on historical weights and fail to track real changes in skill requirements in a timely manner, leading to assessment lag. If the process noise covariance Q is set too large, the system will become overly sensitive to noise, resulting in decreased stability. A similar dilemma exists in setting the observation noise covariance R. If the observation noise covariance R is set too small, it will over-rely on observation data, amplifying noise in the data and causing weight oscillations. If the observation noise covariance R is set too large, it will reduce the role of observation data and affect the timeliness of the assessment.

[0004] Meanwhile, the uncertainty of skill weights in talent assessment and the dynamic nature of observation noise mean that fixing the process noise covariance Q / observation noise covariance R will lead to insufficient system adaptability.

[0005] Therefore, there is an urgent need for a dynamic talent value assessment system that can adaptively adjust the process noise covariance Q / observation noise covariance R and has low model complexity.

[0006] To this end, the present invention provides a dynamic evaluation system for talent value based on a price-increasing gyroscope value model. Summary of the Invention

[0007] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] In a first aspect, this invention provides a dynamic talent value assessment system based on a price-increasing gyroscope value model, comprising:

[0010] Basic Adjustment Factor Module: Acquire multi-source environmental anchor data, construct a skill popularity index, and dynamically generate a basic adjustment factor for the process noise covariance Q based on the skill popularity index;

[0011] Weight optimization module: Establish an observation data credibility assessment model, calibrate the credibility assessment model, and dynamically optimize the weight allocation of the observation noise covariance R;

[0012] Dynamic optimization module: Obtains the first-order difference of the skill popularity index and the skill weights output by the Kalman filter, as well as the Kalman filter prediction and actual values, and dynamically balances the process noise covariance Q and the observation noise covariance R.

[0013] As a further improvement of one embodiment of the present invention, the specific process of acquiring multi-source environmental anchor point data is as follows:

[0014] It can crawl industry reports, corporate strategy documents, and job requirement keywords from recruitment platforms in real time.

[0015] In one embodiment of the present invention, the process of constructing the skill popularity index is as follows:

[0016] By integrating industry reports, corporate strategy documents, and job requirement keywords, a skills popularity index is constructed. The process of constructing the skills popularity index is as follows:

[0017] Skill Popularity Index S: Among them, W i For the weight of the data source, F i T represents the frequency of skill keywords. i is the time decay coefficient, and N is the normalization factor, which normalizes the skill popularity index S to the range of 0 to 100.

[0018] As a further improvement of one embodiment of the present invention, the process of adjusting the basic adjustment factor of the noise covariance Q in the dynamic generation process is as follows:

[0019] Based on the skill popularity index, the process of dynamically generating the basic adjustment factor of the process noise covariance Q is as follows: Perform a first-order difference operation on the skill popularity index to generate the basic adjustment factor of the process noise covariance Q: Q base =k1×max(0,ΔH)+k0, where k1 is the sensitivity coefficient and k0 is the minimum value to prevent Q from being too small and causing complete rigidity.

[0020] As a further improvement of one embodiment of the present invention, the specific process of establishing the observation data credibility assessment model is as follows:

[0021] The observed data was divided into three categories and assigned weights: the first category was project delivery data automatically recorded by the system; the second category was the ratings from direct supervisors; and the third category was the ratings from colleagues across departments. The first category was marked as high-confidence data, the second category as medium-confidence data, and the third category as low-confidence data.

[0022] A confidence assessment model for observation data is established by quantifying high-confidence, medium-confidence, and low-confidence data, classifying the observation data and assigning numerical values.

[0023] As a further improvement of one embodiment of the present invention, the specific process of calibrating the credibility assessment model is as follows:

[0024] By combining historical observation data and corresponding actual results, deviation calculation is performed, the deviation rate of each type of data is calculated, the deviation rate of any type of data is statistically analyzed, and the mean deviation rate is calculated.

[0025] Based on the confidence value of any type of data, a calibration strategy is designed to adjust the confidence value in real time. The average real-time deviation rate is calculated using the most recent N periods of data. If the deviation rate is greater than the baseline deviation rate for N / 2 consecutive periods, the confidence value of the data is lowered.

[0026] In one embodiment of the present invention, the specific process of dynamically optimizing the weight allocation of the observation noise covariance R is as follows:

[0027] For single observation data, the observation noise covariance R is calculated based on the confidence level of its source: R = R0 × (1-C), where R0 is the baseline noise value of this type of data and C is the confidence level value.

[0028] In one embodiment of the present invention, the process of the dynamic equilibrium process noise covariance Q and the observation noise covariance R includes:

[0029] The first-order difference ΔH of the skill popularity index generated by the basic adjustment factor module is obtained, and the analysis is performed using the first-order difference ΔH of the skill popularity index and the first-order difference ΔW of the skill weights output by the Kalman filter.

[0030] As a further improvement of one embodiment of the present invention, the process of dynamic equilibrium process noise covariance Q and observation noise covariance R further includes:

[0031] If the first difference of the skill popularity index and the skill weight output by the Kalman filter increase in the same direction for M consecutive cycles, then the adjustment range of the current process noise covariance Q will be maintained.

[0032] If the first difference of the skill popularity index does not move in the same direction as the skill weights output by the Kalman filter for M consecutive cycles, the sensitivity of the process noise covariance Q is temporarily increased to force faster tracking of external trends.

[0033] As a further improvement of one embodiment of the present invention, the process of dynamic equilibrium process noise covariance Q and observation noise covariance R further includes:

[0034] For each observation, calculate the residual value |e| between the Kalman filter prediction and the actual value. t |: Among them, Z tLet be the observation value at time t. Let be the predicted value of the Kalman filter at time t;

[0035] If the residual value is greater than the residual value threshold for M consecutive periods, the confidence value C of this type of data will be lowered to reduce reliance on this data.

[0036] If the residual value is less than or equal to the residual value threshold for M consecutive periods, the confidence value C of this type of data will be increased to enhance the reliance on this data.

[0037] Secondly, this invention provides a method for dynamic evaluation of talent value based on a price-increasing gyroscope value model, including:

[0038] S1: Acquire multi-source environmental anchor data, construct a skill popularity index, and dynamically generate a basic adjustment factor for the process noise covariance Q based on the skill popularity index;

[0039] S2: Establish a reliability assessment model for observation data, calibrate the reliability assessment model, and dynamically optimize the weight allocation of the observation noise covariance R.

[0040] S3: Obtain the skill weights and Kalman filter outputs of the first-order difference of the skill popularity index, as well as the Kalman filter prediction and actual values, and dynamically balance the process noise covariance Q and the observation noise covariance R.

[0041] The beneficial effects of this invention are as follows:

[0042] To address the issues of lag, oscillation, and insufficient adaptability caused by the fixed setting of the noise covariance Q and observation noise covariance R in traditional talent assessment, the basic adjustment factor module constructs a skill popularity index through multi-source environmental data and generates a basic adjustment factor for Q. This enables Q to dynamically track external skill demand trends, achieving a deep integration of skill assessment with industry trends and corporate strategies, thus avoiding lag.

[0043] The weight optimization module classifies and calibrates the credibility of the observed data, and dynamically optimizes the weight allocation of R. This preserves the decision-making value of high-quality data while suppressing noise interference from low-quality data, reducing oscillations in the evaluation results. The dynamic optimization module dynamically balances Q and R through trend comparison and residual analysis, enabling both to adjust adaptively. This improves the accuracy, stability, and robustness of talent evaluation and enhances its adaptability to dynamic changes. In summary, this system significantly improves the quality of dynamic talent value evaluation and is more suitable for practical application scenarios. Attached Figure Description

[0044] The invention will now be further described with reference to the accompanying drawings.

[0045] Figure 1This is a system module diagram of the dynamic talent value assessment system based on the price-increasing gyroscope value model of this invention;

[0046] Figure 2 This is a flowchart illustrating the steps of the dynamic talent value assessment method based on the price-increasing gyroscope value model of this invention. Detailed Implementation

[0047] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0048] Example 1

[0049] like Figure 1 As shown in the embodiment of the present invention, the dynamic talent value assessment system based on the price-increasing gyroscope value model includes:

[0050] Basic Adjustment Factor Module: Acquire multi-source environmental anchor data, construct a skill popularity index, and dynamically generate a basic adjustment factor for the process noise covariance Q based on the skill popularity index;

[0051] In the basic adjustment factor module, the first specific process of obtaining multi-source environmental anchor data is to crawl industry reports, corporate strategic documents, and job demand keywords from recruitment platforms in real time, and construct a skill popularity index to reflect the speed of change in skill demand.

[0052] For example, real-time crawling of industry reports, such as obtaining quarterly Gartner Hype Cycles and IDC industry white papers by subscribing to relevant official APIs, corporate strategy documents, such as identifying core skills and strategic documents mentioned in corporate annual OKRs through OCR recognition technology, and job requirement keywords from recruitment platforms, such as obtaining the frequency of occurrence of skill keywords by connecting with the official API of recruitment platforms and performing deduplication and filtering.

[0053] In the basic adjustment factor module, the second specific one is to integrate industry reports, corporate strategic documents, and job requirement keywords to construct a skills popularity index, which reflects the speed of change in skills demand.

[0054] The process of constructing a skill popularity index is as follows:

[0055] Skill Popularity Index S: Among them, W i For the weight of the data source, F i T represents the frequency of skill keywords. i is the time decay coefficient, and N is the normalization factor, which normalizes the skill popularity index S to the range of 0 to 100.

[0056] It should be noted that Wi The data source weight is used to measure the authenticity of the data source. It refers to the weight set for industry reports, corporate strategy documents, and job requirement keywords. It can be set according to specific circumstances. This invention does not limit it. For example, the weight of industry reports can be set to 0.5, corporate strategy documents to 0.3, and keyword settings to 0.2.

[0057] In some embodiments, the weight of job requirement keywords can be set to 0.6 to quickly capture emerging professional skills. If the technical route is tracked over a long period, the weight of industry reports can be set to 0.6 to focus on long-term periodic evaluation of the skills of the technical route.

[0058] It should also be noted that T i This is a time decay coefficient used to filter out fake and deliberately hyped job postings. It can be adjusted according to the actual situation, and this invention does not limit it. For example, a gradient time decay coefficient table can be set. For instance, in this invention, the decay coefficient can be set to 1 within 7 days of job posting, 0.7 within 7 to 30 days, 0.2 within 30 to 90 days, and 0.1 beyond 90 days.

[0059] In the basic adjustment factor module, the third specific step, based on the skill popularity index, is to dynamically generate the basic adjustment factor for the process noise covariance Q. This involves performing a first-order difference operation on the skill popularity index to generate the basic adjustment factor for the process noise covariance Q: Q base = k1×max(0,ΔH)+k0, where k1 is the sensitivity coefficient. If the industry is in a period of change, the sensitivity coefficient can be set to 0.8. If the industry is in a stable period, the sensitivity coefficient can be set to 0 or 3. The sensitivity coefficient can be adjusted according to the actual situation. This invention does not limit the sensitivity coefficient. k0 is the minimum value to avoid Q being too small and causing complete stagnation. The minimum value can be set according to the specific situation. In this invention, the minimum value is not limited.

[0060] The basic adjustment factor module provides an adjustment basis for the process noise covariance Q value of Kalman filtering through three-dimensional perception of multi-source data, precise quantification of dynamic models, and intelligent adjustment of adaptive factors. It not only solves the problem of lag and oscillation of the process noise covariance Q of traditional fixed Kalman filtering, but also realizes the deep binding of skill requirements with corporate strategy and industry trends, enabling dynamic talent assessment to have the core capability of moving from passive response to proactive prediction.

[0061] Weight optimization module: Establish an observation data credibility assessment model, calibrate the credibility assessment model, and dynamically optimize the weight allocation of the observation noise covariance R;

[0062] In the weight optimization module, the first specific step is to establish an observation data credibility assessment model, which divides the observation data into three categories and assigns weights to them. The first category is project delivery data automatically recorded by the system; the second category is the ratings from direct superiors; and the third category is the ratings from colleagues across departments. The first category of data is marked as high credibility data, the second category as medium credibility data, and the third category as low credibility data.

[0063] A confidence assessment model for observation data is established by quantifying high-confidence, medium-confidence, and low-confidence data, classifying the observation data and assigning numerical values.

[0064] It should be noted that the observation data includes: Category I data, Category II data, and Category III data;

[0065] For example, high-confidence data is assigned a value of 0.9, medium-confidence data is assigned a value of 0.8, and low-confidence data is assigned a value of 0.5. It should be noted that in this invention, the values ​​assigned to high-confidence data, medium-confidence data, and low-confidence data are not limited, and can be set according to the actual situation.

[0066] In the weight optimization module, the second specific step is to calibrate the credibility assessment model, combine historical observation data and corresponding actual results, perform deviation calculation, calculate the deviation rate of each type of data, and statistically analyze the deviation rate of any type of data (first type of data, second type of data, and third type of data) to calculate the mean deviation rate.

[0067] For example, calculate the deviation rate for each type of data:

[0068] Based on the confidence value of any type of data (e.g., 0.9 for high confidence data, 0.8 for medium confidence data, and 0.5 for low confidence data), a calibration strategy is designed to adjust the confidence value in real time. The average real-time deviation rate is calculated using the most recent N (N ≤ 6) periods of data. If the deviation rate is greater than the baseline deviation rate for N / 2 consecutive periods, the confidence value of the data is reduced by 10%, and the dynamic adjustment mechanism of the observation noise covariance R is triggered.

[0069] It should be noted that the benchmark deviation rate is a reference value set by technical personnel in this industry based on historical experience, and there is no limit to the percentage of the confidence value that can be reduced.

[0070] In the weight optimization module, the second specific aspect is the dynamic optimization of the weight allocation of the observation noise covariance R. For a single observation data, the observation noise covariance R is calculated based on the confidence value of its source: R = R0 × (1-C), where R0 is the baseline noise value of this type of data and C is the confidence value.

[0071] It should be noted that R0 is the baseline noise value for this type of data. For example, the observation noise covariance R of high confidence data can be set to only 10% of the baseline noise value to reduce the sensitivity to noise; the observation noise covariance R of low confidence data can be set to 50% of the baseline value to avoid subjective bias affecting weight adjustment; the confidence value can be adjusted according to the actual situation, and this invention does not limit it.

[0072] The observed data are divided into three confidence levels—high, medium, and low—based on their source characteristics. This reflects the objective differences in data reliability in real-world scenarios and provides a clear classification basis for subsequent weight allocation. It avoids homogenization of data of different quality. By assigning specific numerical values, the abstract confidence level is transformed into a calculable parameter, making the influence of the observed data on the evaluation results quantifiable and comparable. This provides a numerical basis for the dynamic adjustment of R in Kalman filtering and enhances the practicality of the model.

[0073] By combining historical data with actual results to calculate the deviation rate, and by using a calibration strategy that lowers the credibility if the deviation exceeds the benchmark for N / 2 consecutive periods, the deviation of the initial credibility assignment is corrected in real time. This corrects for potential subjective biases in the rating of the direct superior, enabling the credibility assessment model to adaptively iterate as data quality changes, thus avoiding long-term errors caused by fixed values.

[0074] Based on the dynamically updated credibility numerical calculation R, differentiated processing is achieved, with low noise weights for high credibility data and high noise weights for low credibility data. This preserves the decision-making value of high-quality data, such as project data automatically recorded by the system, while suppressing noise interference from low-quality data, such as subjective rating biases from colleagues across departments, thus reducing fluctuations in evaluation results.

[0075] Dynamic optimization module: Obtains the skill weights from the first difference and Kalman filter outputs of the skill popularity index, as well as the Kalman filter prediction and actual values, and dynamically balances the process noise covariance Q and the observation noise covariance R.

[0076] In the dynamic optimization module, the first specific step is to obtain the first difference ΔH of the skill popularity index generated by the basic adjustment factor module, and then analyze it using the first difference ΔH of the skill popularity index and the first difference ΔW of the skill weights output by the Kalman filter.

[0077] If the first difference of the skill popularity index and the skill weight output by the Kalman filter increase in the same direction for M consecutive periods (M is greater than 3), then the adjustment range of the current process noise covariance Q will be maintained.

[0078] If the first difference of the skill popularity index does not move in the same direction as the skill weight output by the Kalman filter for M consecutive periods (M is greater than 3), then the sensitivity of the process noise covariance Q is temporarily increased to force the tracking of external trends to accelerate.

[0079] The sensitivity of temporarily increasing the process noise covariance Q can be set according to the actual situation. For example, the sensitivity of the process noise covariance Q can be set as: Q T =Q base ×1.2, in this invention, there is no limitation;

[0080] In the dynamic optimization module, specifically, for each observation data point, the residual value |e| between the Kalman filter prediction and the actual value is calculated. t |: Among them, Z t Let be the observation value at time t. Let be the predicted value of the Kalman filter at time t;

[0081] If the residual value is greater than the residual value threshold for M consecutive periods (M is greater than 3), the confidence value C of this type of data will be lowered to reduce the dependence on this data.

[0082] If the residual value is less than or equal to the residual value threshold for M consecutive periods (M is greater than 3), then the confidence value C of this type of data is increased to enhance the reliance on this data.

[0083] By comparing the first difference of the skill popularity index with the first difference of the skill weights output by the Kalman filter, the stability of Q is maintained when the two are in the same long-term direction, avoiding evaluation oscillations caused by over-adjustment; when the long-term directions are inconsistent, the sensitivity of Q is increased to force tracking of external trends, effectively solving the problem that the evaluation results under the traditional fixed Q value lag behind industry changes, and ensuring that the adjustment of skill weights is closely linked to external demand trends.

[0084] Based on residual analysis between Kalman filter predictions and actual values, the reliability value C of the observed data is dynamically adjusted, achieving an adaptive response to data quality. When the residual exceeds the threshold for a long period, C is lowered to reduce the interference of low-quality data on the evaluation; when the residual is stable for a long period, C is raised to enhance the decision-making value of high-quality data. This avoids evaluation bias caused by subjective or low-quality data, fully leverages the supporting role of high-quality data, improves the reliability of the evaluation results, maintains the stability of internal evaluation, and improves the accuracy and robustness of dynamic talent value evaluation as a whole.

[0085] The technical solution of this embodiment is as follows: Dynamically generating the basic adjustment factor for the process noise covariance Q; acquiring multi-source environmental anchor point data: real-time crawling of three types of data, including industry reports, corporate strategic documents, and job requirement keywords from recruitment platforms; constructing a skill popularity index: merging the above data to generate a skill popularity index; performing a first-order difference operation on the skill popularity index to generate a basic adjustment factor; dividing the observed data into three categories and assigning initial confidence values: high confidence data, medium confidence data, and low confidence data, with values ​​adjustable as needed; calculating the deviation rate by combining historical observation data with actual results; statistically calculating the average deviation rate for each data category; calculating the real-time average deviation rate; if the deviation rate exceeds the benchmark for N / 2 consecutive periods, then this data category can be... The reliability value is reduced by 10%, triggering the adjustment mechanism of R. For a single observation, R is calculated based on its source reliability value C. The dynamic adjustment of the process noise covariance Q and the observation noise covariance is dynamically balanced. The first difference of the skill popularity index is obtained and compared with the first difference of the skill weight output by the Kalman filter. If the direction is consistent for M consecutive periods, the current adjustment magnitude of Q is maintained. If the direction is inconsistent for M consecutive periods, the sensitivity of Q is temporarily increased to force tracking of external trends. The residual between the Kalman filter prediction value and the actual value is calculated. If the residual exceeds the threshold for M consecutive periods, the reliability C of the corresponding data is reduced and R is increased to reduce its impact. If the residual is less than or equal to the threshold for M consecutive periods, C is increased and R is decreased to strengthen its effect.

[0086] Example 2

[0087] like Figure 2 As shown in Example 1, this invention provides a dynamic talent value assessment method based on a price-increasing gyroscope value model, including:

[0088] S1: Acquire multi-source environmental anchor data, construct a skill popularity index, and dynamically generate a basic adjustment factor for the process noise covariance Q based on the skill popularity index;

[0089] As a further improvement, the specific process for acquiring multi-source environmental anchor point data is as follows:

[0090] It can crawl industry reports, corporate strategy documents, and job requirement keywords from recruitment platforms in real time.

[0091] As a further improvement, the process of constructing the skill popularity index is as follows:

[0092] By integrating industry reports, corporate strategy documents, and job requirement keywords, a skills popularity index is constructed. The process of constructing the skills popularity index is as follows:

[0093] Skill Popularity Index S: Among them, W i For the weight of the data source, F i T represents the frequency of skill keywords.i is the time decay coefficient, and N is the normalization factor, which normalizes the skill popularity index S to the range of 0 to 100.

[0094] As a further improvement, the process for adjusting the basic adjustment factor of the noise covariance Q in the dynamic generation process is as follows:

[0095] Based on the skill popularity index, the process of dynamically generating the basic adjustment factor of the process noise covariance Q is as follows: Perform a first-order difference operation on the skill popularity index to generate the basic adjustment factor of the process noise covariance Q: Q base =k1×max(0,ΔH)+k0, where k1 is the sensitivity coefficient and k0 is the minimum value to prevent Q from being too small and causing complete rigidity.

[0096] S2: Establish a reliability assessment model for observation data, calibrate the reliability assessment model, and dynamically optimize the weight allocation of the observation noise covariance R.

[0097] As a further improvement, the specific process of establishing the observation data credibility assessment model is as follows:

[0098] The observed data was divided into three categories and assigned weights: the first category was project delivery data automatically recorded by the system; the second category was the ratings from direct supervisors; and the third category was the ratings from colleagues across departments. The first category was marked as high-confidence data, the second category as medium-confidence data, and the third category as low-confidence data.

[0099] A confidence assessment model for observation data is established by quantifying high-confidence, medium-confidence, and low-confidence data, classifying the observation data and assigning numerical values.

[0100] As a further improvement, the specific process of calibrating the credibility assessment model is as follows:

[0101] By combining historical observation data and corresponding actual results, deviation calculation is performed, the deviation rate of each type of data is calculated, the deviation rate of any type of data is statistically analyzed, and the mean deviation rate is calculated.

[0102] Based on the confidence value of any type of data, a calibration strategy is designed to adjust the confidence value in real time. The average real-time deviation rate is calculated using the most recent N periods of data. If the deviation rate is greater than the baseline deviation rate for N / 2 consecutive periods, the confidence value of the data is lowered.

[0103] As a further improvement, the specific process of dynamically optimizing the weight allocation of the observation noise covariance R is as follows:

[0104] For single observation data, the observation noise covariance R is calculated based on the confidence level of its source: R = R0 × (1-C), where R0 is the baseline noise value of this type of data and C is the confidence level value.

[0105] S3: Obtain the skill weights and Kalman filter outputs of the first-order difference of the skill popularity index, as well as the Kalman filter prediction and actual values, and dynamically balance the process noise covariance Q and the observation noise covariance R.

[0106] As a further improvement, the process of reconciling the noise covariance Q and the observation noise covariance R in the dynamic equilibrium process includes:

[0107] The first-order difference ΔH of the skill popularity index generated by the basic adjustment factor module is obtained, and the analysis is performed using the first-order difference ΔH of the skill popularity index and the first-order difference ΔW of the skill weights output by the Kalman filter.

[0108] As a further improvement, the process of reconciling the noise covariance Q and the observation noise covariance R in the dynamic equilibrium process also includes:

[0109] If the first difference of the skill popularity index and the skill weight output by the Kalman filter increase in the same direction for M consecutive cycles, then the adjustment range of the current process noise covariance Q will be maintained.

[0110] If the first difference of the skill popularity index does not move in the same direction as the skill weights output by the Kalman filter for M consecutive cycles, the sensitivity of the process noise covariance Q is temporarily increased to force faster tracking of external trends.

[0111] As a further improvement, the process of reconciling the noise covariance Q and the observation noise covariance R in the dynamic equilibrium process also includes:

[0112] For each observation, calculate the residual value |e| between the Kalman filter prediction and the actual value. t |: Among them, Z t Let be the observation value at time t. Let be the predicted value of the Kalman filter at time t;

[0113] If the residual value is greater than the residual value threshold for M consecutive periods, the confidence value C of this type of data will be lowered to reduce reliance on this data.

[0114] If the residual value is less than or equal to the residual value threshold for M consecutive periods, the confidence value C of this type of data will be increased to enhance the reliance on this data.

[0115] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic talent value assessment system based on a price-increasing gyroscope value model, characterized by: include: Basic Adjustment Factor Module: Acquire multi-source environmental anchor data, construct a skill popularity index, and dynamically generate a basic adjustment factor for the process noise covariance Q based on the skill popularity index; Weight optimization module: Establish an observation data credibility assessment model, calibrate the credibility assessment model, and dynamically optimize the weight allocation of the observation noise covariance R; Dynamic optimization module: Obtains the first-order difference of the skill popularity index and the skill weights output by the Kalman filter, as well as the Kalman filter prediction and actual values, and dynamically balances the process noise covariance Q and the observation noise covariance R.

2. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The specific process for acquiring multi-source environmental anchor point data is as follows: It can crawl industry reports, corporate strategy documents, and job requirement keywords from recruitment platforms in real time.

3. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The process of constructing the skill popularity index is as follows: By integrating industry reports, corporate strategy documents, and job requirement keywords, a skills popularity index is constructed. The process of constructing the skills popularity index is as follows: Skill Popularity Index S: Among them, W i For the weight of the data source, F i T represents the frequency of skill keywords. i is the time decay coefficient, and N is the normalization factor, which normalizes the skill popularity index S to the range of 0 to 100.

4. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The process for adjusting the basic adjustment factor of the noise covariance Q in the dynamic generation process is as follows: Based on the skill popularity index, the process of dynamically generating the basic adjustment factor of the process noise covariance Q is as follows: Perform a first-order difference operation on the skill popularity index to generate the basic adjustment factor of the process noise covariance Q: Q base =k1×max(0,ΔH)+k0, where k1 is the sensitivity coefficient and k0 is the minimum value to prevent Q from being too small and causing complete rigidity.

5. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The specific process for establishing the observation data reliability assessment model is as follows: The observed data was divided into three categories and assigned weights: the first category was project delivery data automatically recorded by the system; the second category was the ratings from direct supervisors; and the third category was the ratings from colleagues across departments. The first category was marked as high-confidence data, the second category as medium-confidence data, and the third category as low-confidence data. A confidence assessment model for observation data is established by quantifying high-confidence, medium-confidence, and low-confidence data, classifying the observation data and assigning numerical values.

6. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The specific process for calibrating the credibility assessment model is as follows: By combining historical observation data and corresponding actual results, deviation calculation is performed, the deviation rate of each type of data is calculated, the deviation rate of any type of data is statistically analyzed, and the mean deviation rate is calculated. Based on the confidence value of any type of data, a calibration strategy is designed to adjust the confidence value in real time. The average real-time deviation rate is calculated using the most recent N periods of data. If the deviation rate is greater than the baseline deviation rate for N / 2 consecutive periods, the confidence value of the data is lowered.

7. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The specific process of weight allocation for dynamically optimizing the observation noise covariance R is as follows: For single observation data, the observation noise covariance R is calculated based on the confidence level of its source: R = R0 × (1-C), where R0 is the baseline noise value of this type of data and C is the confidence level value.

8. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 1, characterized in that: The process of the dynamic equilibrium process noise covariance Q and observation noise covariance R includes: The first-order difference ΔH of the skill popularity index generated by the basic adjustment factor module is obtained, and the analysis is performed using the first-order difference ΔH of the skill popularity index and the first-order difference ΔW of the skill weights output by the Kalman filter.

9. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 8, characterized in that: The process of reconciling the noise covariance Q and the observation noise covariance R in the dynamic equilibrium process also includes: If the first difference of the skill popularity index and the skill weight output by the Kalman filter increase in the same direction for M consecutive cycles, then the adjustment range of the current process noise covariance Q will be maintained. If the first difference of the skill popularity index does not move in the same direction as the skill weights output by the Kalman filter for M consecutive cycles, the sensitivity of the process noise covariance Q is temporarily increased to force faster tracking of external trends.

10. The dynamic talent value assessment system based on the price-increasing gyroscope value model according to claim 8, characterized in that: The process of reconciling the noise covariance Q and the observation noise covariance R in the dynamic equilibrium process also includes: For each observation, calculate the residual value |e| between the Kalman filter prediction and the actual value. t |: Among them, Z t Let be the observation value at time t. Let be the predicted value of the Kalman filter at time t; If the residual value is greater than the residual value threshold for M consecutive periods, the confidence value C of this type of data will be lowered to reduce reliance on this data. If the residual value is less than or equal to the residual value threshold for M consecutive periods, the confidence value C of this type of data will be increased to enhance the reliance on this data.

Images