Competency extraction and evaluation method based on employee occupational behavior data driving

Abstract

The invention discloses a competency extraction and evaluation method based on employee occupational behavior data driving, which comprises the following steps: firstly collecting employee occupational behavior data and extracting behavior characteristics related to competency to obtain a characteristic set, then constructing a competency factor model, defining a hierarchical structure of competency factors, and performing fine evaluation by a fine-grained feedback model. The method comprises the following steps of: selecting a total competency factor set of a hierarchical structure, optimizing by applying a self-reflection mechanism and a reinforcement learning algorithm, dynamically adjusting the total competency factor set of the hierarchical structure, and finally evaluating the competency of each employee in real time based on a real-time competency factor model and a fine-grained feedback model. According to the invention, the accuracy and consistency of employee competency evaluation results are ensured, and the problems of inaccurate employee competency evaluation results and poor evaluation efficiency at present are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of human resource data processing, and in particular to a competency extraction and evaluation method driven by employee professional behavior data. Background Art

[0002] Traditional competency assessments rely too much on the subjective experience of experts and rigid assessment systems, which often lead to assessment results that are inconsistent with employees' actual behavior. This method is highly subjective and difficult to measure with specific values, and cannot meet the needs of companies for accurate talent matching. With the development of digital management tools, companies are now able to collect employee career behavior data through various systems, such as work logs and project feedback, providing a data foundation for data-driven competency factor extraction.

[0003] Machine learning algorithms can mine patterns and trends from massive amounts of data, which to a certain extent enhances the accuracy of assessments. However, these machine learning models often require experts to perform tedious feature engineering, which is not only inefficient, but also often cannot effectively handle complex nonlinear relationships. As the latest advance in the field of artificial intelligence, large language models have demonstrated unprecedented capabilities in natural language processing and understanding, bringing new possibilities to competency assessment. However, the decision-making process of these models lacks transparency and is highly sensitive to the quality and quantity of training data, which can easily amplify biases and errors in the data.

[0004] In addition, existing technologies fail to fully consider the specific needs of corporate culture, making it difficult to accurately assess the competence of employees in a specific cultural context. The lack of an alignment mechanism for corporate cultural preferences makes it difficult to fully reflect the competence of employees in a specific cultural context. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a competency extraction and evaluation method driven by employee professional behavior data, which solves the current problems of inaccurate employee competency evaluation results and poor evaluation efficiency.

[0006] The technical solution of the present invention is:

[0007] A competency extraction and assessment method based on employee professional behavior data drive includes the following steps:

[0008] (1) Collect employee career behavior data and perform data preprocessing to obtain data set F;

[0009] (2) For the data set F, a clustering algorithm is used to identify the potential patterns of employees’ professional behavior. The number of clusters is set to k, and the clustering result is C = {c1, c2, …, c i ,…,c k}, then extract the features that have a significant impact on competency assessment to form a feature set F′, so that

[0010] (3) Integrate the dataset F to form a comprehensive dataset Among them, D i Denotes the dataset D integrated The i-th data source in, m is the data set D integrated The number of data sources in

[0011] (4) Construct a competency factor model and use the analytic hierarchy process and hierarchical semantic ID to define the total competency factor set E of each employee as a hierarchical structure of competency factors;

[0012] (5) Conduct refined evaluation through a fine-grained feedback model, iteratively optimize the output of the fine-grained feedback model through a self-reflection mechanism, and further iteratively optimize the fine-grained feedback model using a reinforcement learning algorithm. Finally, in each evaluation cycle, dynamically adjust the total set of competency elements in the hierarchy based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm.

[0013] (6) Based on the dynamically adjusted competency factor model and the real-time optimized fine-grained feedback model, each employee’s competency is evaluated in real time to obtain the employee’s real-time competency evaluation results.

[0014] The steps of collecting employee occupational behavior data and performing data preprocessing to obtain data set F are as follows:

[0015] S11. First, collect the career behavior data of each employee. Each employee corresponds to an initial data set D = {d1, d2, ..., d i ,…,d n}, where each d i represents an employee's professional behavior record, and n is the number of employee's professional behavior records;

[0016] S12. Preprocess the initial data set D of each employee to obtain a data set F; wherein the preprocessing process includes missing value processing, outlier detection and data standardization processing.

[0017] The missing value processing is to set the missing value ratio p. After the missing value processing, the integrity of the data set is expressed as D′=D\{d i ∣d i is missing};

[0018] The outlier detection is performed by using statistical methods, setting a threshold z, and defining the outlier value as |z i|>z;

[0019] The data normalization process is performed using the formula To transform the original data x, we get the standardized data x′; where μ represents the mean and σ represents the standard deviation.

[0020] The dataset D integrated Divide into training set and test set, set the training set ratio to r, then the training set is D train =D integrated ×r, the test set is D test =D integrated ×(1-r).

[0021] The specific steps of extracting features that have a significant impact on competency assessment and forming a feature set F′ are as follows:

[0022] S21. Collect the employee career behavior data after data preprocessing and construct the set P =

[0023] {p1,p2,…,p n}, where n is the number of employee professional behavior records. Then, the text data of employee professional behavior records is analyzed through natural language processing technology to extract keywords and phrases related to competency elements, thus forming a data set F = {f1, f2, …, f m}, where m is the number of features in the dataset F;

[0024] S22. Use machine learning algorithms to evaluate the contribution of each feature of the dataset F to the competency assessment results, by setting the feature importance threshold t f , thus selecting the feature set F′={f j ∣importance(f j )>t f}, f j is a feature in the feature set F′.

[0025] In the competency factor model, each competency factor in the initial total competency factor set corresponds to the collected employee professional behavior data. The total competency factor set E of each employee is defined as a hierarchical structure of competency factors by using the hierarchical analysis method and hierarchical semantic ID. The specific steps are as follows:

[0026] S41. Define the hierarchical structure of the total competency factor set E: The total competency factor set E is divided into different levels, each level E j They all contain a set of elements, as shown in the following formula (1):

[0027]

[0028] In formula (1), J represents the number of levels, E j represents the set of competency factors at the jth level, and ∪ is the union symbol;

[0029] S42. Generate hierarchical semantic IDs of competency elements. Each competency element e i A hierarchical semantic ID is assigned to each competency element, which clarifies the position of the competency element in the competency hierarchy. See the following formula (2) for details:

[0030]

[0031] In formula (2), L represents the level depth, ID l (e i ) represents competency factor e i Hierarchical semantic ID at level l;

[0032] S43. Competency elements i The weight w i The analytic hierarchy process is used to determine the specific formula (4):

[0033]

[0034] In formula (4), p il It is a competency element i Compared with other competency elements l Ratings;

[0035] S44, Inter-level Correlation C jl It is calculated by considering the correlation between competency elements at different levels to assess the importance and mutual influence of different levels, as shown in the following formula (5):

[0036]

[0037] In formula (5), E l represents the set of competency elements at level l, e k represents the competency elements in the competency element set of the first level, w k Representative competency elements k The weight of correlation(E j ,E l ) represents the correlation between the j-th and l-th level competency factor sets.

[0038] The generation of the hierarchical semantic ID is based on the function f l , the expression is shown in the following formula (3):

[0039] ID l (e i )=fl (e i , P(e i )) (3);

[0040] In formula (3), e i is the competency factor, P(e i ) is the competency factor e i Attributes.

[0041] The fine-grained feedback model is used for refined evaluation, and the output of the fine-grained feedback model is iteratively optimized through a self-reflection mechanism. The fine-grained feedback model is further iteratively optimized using a reinforcement learning algorithm. Finally, in each evaluation cycle, the total set of competency elements of the hierarchical structure is dynamically adjusted based on the score of the fine-grained feedback model and the optimization result of the reinforcement learning algorithm. The specific steps are:

[0042] S51. For each employee’s career behavior data x, the fine-grained feedback model f p (x, y) corresponds to the scores of multiple dimensions, as shown in the following formula (6):

[0043] f p (x, y)={Logical Correctness, Factuality, Insightfulness,

[0044] Communication Skills, Collaboration, Adaptability} (6);

[0045] In formula (6), Logical Correctness represents logical correctness, Factuality represents factuality, Insightfulness represents insight, Communication Skills represents communication skills, Collaboration represents collaboration skills, and Adaptability represents adaptability;

[0046] Each dimension is scored as -1, 0, or 1, with -1 representing failure, 0 representing medium, and 1 representing success;

[0047] S52, through the fine-grained feedback model f p (x,y) generates a response set D y , response set D y Each response y in i All will be fed back through the fine-grained feedback model f p (x, y) is scored, as shown in the following formula (7):

[0048]

[0049] In formula (7), is based on the current fine-grained feedback model f p (x,y) parameters The generated policy function, r(x,y i ) is the reward value given by the preference model;

[0050] Select candidate response set D y The highest scoring response As a starting point for reflection and optimization:

[0051] Based on fine-grained feedback model Generate self-reflection and improve responses to Conduct self-reflection and generate new and improved responses j , see the following formula (8) for details:

[0052]

[0053] In formula (8), Indicates input combined with feedback information; D z represents the improved response set;

[0054] S53. After obtaining the improved response set D z Afterwards, the reinforcement learning algorithm DPO is applied to further iteratively optimize the fine-grained feedback model to obtain R;

[0055] Objective function of DPO optimization See the following formula (9):

[0056]

[0057] In formula (9), y + and - are positive and negative sample responses, respectively, ref is the reference strategy, β is the hyperparameter controlling the weight; σ represents the sigmoid activation function, E D In the test set D test The objective function is averaged over all samples in to evaluate the performance of the fine-grained feedback model on the dataset;

[0058] S54. In each evaluation cycle, based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm, the total set of competency elements of the hierarchical structure is dynamically adjusted, as shown in the following formula (10):

[0059] E t+1 =g(E t , Rt ) (10);

[0060] In formula (10), E t+1 is the hierarchical competency factor set at time t+1, g is the update function, E t is the hierarchical set of competency factors at time t, R t is the feedback information at time t, that is, the result obtained by further iteratively optimizing the fine-grained feedback model using the reinforcement learning algorithm DPO at time t.

[0061] The competency assessment result of each employee is the total competency score S, which is the sum of all competency factors. i The weighted sum of the corresponding feature scores;

[0062] S61. First, the weight of the association matrix A and the score of the fine-grained feedback model are used to adjust the weight of each competency factor, as shown in the following formula (11):

[0063]

[0064] In formula (11), w i It is a competency element i The weight of A ij is feature f in feature set F′ j Competency elements i The correlation degree of ; m represents the number of features in the feature set F′; I represents the number of competency elements in the total competency element set;

[0065] S62. After each competency assessment, the weight w is adjusted according to the scoring results of the fine-grained feedback model. i Update the weight w i The update formula of is shown in the following formula (12):

[0066]

[0067] In formula (12), α is the learning rate, which is used to control the update speed of the weights; f p (x,y i ) is a fine-grained feedback model for competency elements e i Ratings; is the competency factor e at time t i The weight of

[0068] S63. The final total competency score of each employee is output as the evaluation result, as shown in the following formula (13):

[0069]

[0070] In formula (13), It is the final weight after T feedback adjustments, m represents the number of features in the feature set F′, and I represents the number of competency elements in the total competency element set.

[0071] Advantages of the present invention:

[0072] (1) The present invention constructs a set of competency factors by collecting employees' professional behaviors, and identifies key competency factors that are consistent with corporate values, behavioral norms, and work environment, thereby clarifying the evaluation objectives and specific needs of corporate culture;

[0073] (2) The present invention constructs a competency factor model, uses the hierarchical analysis method and hierarchical semantic ID, and defines the total competency factor set E of each employee as a hierarchical structure of competency factors, so that the competency factors are divided into different levels, each level reflects different aspects of the organizational structure, capability level or professional knowledge field, and sets a hierarchical ID to clarify the position of the competency factor in the competency hierarchy;

[0074] (3) The present invention uses a fine-grained feedback model to perform refined evaluation and iteratively optimizes the output of the fine-grained feedback model through a self-reflection mechanism to solve the problem of inconsistent preferences of human evaluators, so as to dynamically align corporate culture preferences to ensure the accuracy and consistency of employee competency evaluation results;

[0075] (4) The present invention applies a reinforcement learning algorithm to further iteratively optimize the fine-grained feedback model. By optimizing the positive and negative sample pairs, the fine-grained feedback model can effectively improve the alignment with the corporate culture preferences.

[0076] (5) The present invention conducts real-time evaluation of each employee's competency based on a dynamically adjusted competency factor model and a real-time optimized fine-grained feedback model, so that new employee career behavior data and evaluation results are continuously incorporated into the evaluation system. The fine-grained feedback model adjusts the competency factors according to the scores of different dimensions in each evaluation cycle, and finally obtains a set of competency factors that are closely aligned with the corporate culture. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] See Figure 1 , a competency extraction and assessment method based on employee professional behavior data, specifically including the following steps:

[0080] (1) Collect employee career behavior data and perform data preprocessing to obtain data set F;

[0081] S11. First, collect the career behavior data of each employee. Each employee corresponds to an initial data set D = {d1, d2, ..., d i ,…,d n}, where each d i represents an employee's professional behavior record, and n is the number of employee's professional behavior records;

[0082] S12, preprocessing the initial data set D of each employee to obtain a data set F; wherein the preprocessing process includes missing value processing, outlier detection and data standardization processing;

[0083] Missing value processing is to set the missing value ratio p. After missing value processing, the integrity of the data set is expressed as D′=D\{d i ∣d i is missing};

[0084] The outlier detection is performed by using statistical methods, setting a threshold z, and defining the outlier value as |z i |>z;

[0085] The data normalization process is performed using the formula To transform the original data x, we get the standardized data x′; where μ represents the mean and σ represents the standard deviation;

[0086] (2) For the data set F, a clustering algorithm is used to identify the potential patterns of employees’ professional behavior. The number of clusters is set to k, and the clustering result is C = {c1, c2, …, c i ,…,c k}, then extract the features that have a significant impact on competency assessment to form a feature set F′, so that

[0087] S21. Collect the employee career behavior data after data preprocessing and construct the set P =

[0088] {p1,p2,…,p n}, where n is the number of employee professional behavior records. Then, the text data of employee professional behavior records is analyzed through natural language processing technology to extract keywords and phrases related to competency elements, thus forming a data set F = {f1, f2, …, f m}, where m is the number of features in the dataset F;

[0089] S22. Use machine learning algorithms to evaluate the contribution of each feature of the dataset F to the competency assessment results, by setting the feature importance threshold t f , thus selecting the feature set F′={f j ∣importance(f j )>t f}, f j is the feature in the feature set F′;

[0090] (3) Integrate the dataset F to form a comprehensive dataset Among them, D i Denotes the dataset D integrated The i-th data source in, m is the data set D integrated The number of data sources in the dataset D integrated Divide into training set and test set, set the training set ratio to r, then the training set is D train =D integrated ×r, the test set is D test =D integrated ×(1-r), used for the training and verification of the subsequent competency factor model and fine-grained feedback model;

[0091] (4) Construct a competency factor model and use the analytic hierarchy process and hierarchical semantic ID to define the total competency factor set E of each employee as a hierarchical structure of competency factors;

[0092] S41. Define the hierarchical structure of the total competency factor set E: The total competency factor set E is divided into different levels, each level E j They all contain a set of elements, as shown in the following formula (1):

[0093]

[0094] In formula (1), J represents the number of levels, E j represents the set of competency factors at the jth level, and ∪ is the union symbol;

[0095] S42. Generate hierarchical semantic IDs of competency elements. Each competency element e i A hierarchical semantic ID is assigned to each competency element, which clarifies the position of the competency element in the competency hierarchy. See the following formula (2) for details:

[0096]

[0097] In formula (2), L represents the level depth, ID l(e i ) represents competency factor e i Hierarchical semantic ID at level l; the generation of hierarchical semantic ID is based on the function f l , the expression is shown in the following formula (3):

[0098] ID l (e i )=f l (e i , P(e i )) (3);

[0099] In formula (3), e i is the competency factor, P(e i ) is the competency factor e i Attributes;

[0100] S43. Competency elements i The weight w i The analytic hierarchy process is used to determine the specific formula (4):

[0101]

[0102] In formula (4), p il It is a competency element i Compared with other competency elements l Ratings;

[0103] S44, Inter-level Correlation C jl It is calculated by considering the correlation between competency elements at different levels to assess the importance and mutual influence of different levels and calculate C jl The main purpose is to determine the relationship and influence between competency factors at different levels, which helps to more accurately reflect the importance of each factor and its contribution to employee competency assessment in the competency factor model. It is also used to adjust or optimize the structure of the competency factor model to make it more consistent with the actual organizational structure and capability level. For details, see the following formula (5):

[0104]

[0105] In formula (5), E l represents the set of competency elements at level l, e k represents the competency elements in the competency element set of the first level, w k Representative competency elements k The weight of correlation(E j , E l ) represents the correlation between the j-th level and the l-th level competency element sets;

[0106] (5) Conduct refined evaluation through a fine-grained feedback model, iteratively optimize the output of the fine-grained feedback model through a self-reflection mechanism, and further iteratively optimize the fine-grained feedback model using a reinforcement learning algorithm. Finally, in each evaluation cycle, dynamically adjust the total set of competency elements in the hierarchy based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm.

[0107] S51. For each employee’s career behavior data x, the fine-grained feedback model f p (x, y) corresponds to the scores of multiple dimensions, as shown in the following formula (6):

[0108] f p (x, y)={Logical Correctness, Factority, Insightfulness, CommunicationSkills, Collaboration, Adaptability}

[0109] In formula (6), Logical Correctness represents logical correctness, Factuality represents factuality, Insightfulness represents insight, Communication Skills represents communication skills, Collaboration represents collaboration skills, and Adaptability represents adaptability;

[0110] Each dimension is scored as -1, 0, or 1, with -1 representing failure, 0 representing medium, and 1 representing success;

[0111] S52, through the fine-grained feedback model f p (x,y) generates a response set D y , response set D y Each response y in i All will be fed back through the fine-grained feedback model f p (x, y) is scored, as shown in the following formula (7):

[0112]

[0113] In formula (7), is based on the current fine-grained feedback model f p (x,y) parameters The generated policy function, r(x,y i ) is the reward value given by the preference model;

[0114] Select candidate response set D y The highest scoring response As a starting point for reflection and optimization:

[0115] Based on fine-grained feedback model Generate self-reflection and improve responses to Conduct self-reflection and generate new and improved responses j , see the following formula (8) for details:

[0116]

[0117] In formula (8), Indicates input combined with feedback information; D z represents the improved response set;

[0118] S53. After obtaining the improved response set D z Afterwards, the reinforcement learning algorithm DPO is applied to further iteratively optimize the fine-grained feedback model to obtain z;

[0119] Objective function of DPO optimization See the following formula (9):

[0120]

[0121] In formula (9), y + and - are positive and negative sample responses, respectively, ref is the reference strategy, β is the hyperparameter controlling the weight; σ represents the sigmoid activation function, E D In the test set D test The objective function is averaged over all samples in to evaluate the performance of the fine-grained feedback model on the dataset;

[0122] S54. In each evaluation cycle, based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm, the total set of competency elements of the hierarchical structure is dynamically adjusted, as shown in the following formula (10):

[0123] E t+1 =g(E t , R t ) (10);

[0124] In formula (10), E t+1 is the hierarchical competency factor set at time t+1, g is the update function, E t is the hierarchical set of competency factors at time t, R t is the feedback information at time t, that is, the result obtained by further iteratively optimizing the fine-grained feedback model using the reinforcement learning algorithm DPO at time t;

[0125] (6) Based on the dynamically adjusted competency factor model and the fine-grained feedback model after real-time optimization, each employee's competency is evaluated in real time to obtain the employee's real-time competency evaluation results;

[0126] The competency assessment result of each employee is the total competency score S, which is the sum of all competency factors. i The weighted sum of the corresponding feature scores;

[0127] S61. First, the weight of the association matrix A and the score of the fine-grained feedback model are used to adjust the weight of each competency factor, as shown in the following formula (11):

[0128]

[0129] In formula (11), w i It is a competency element i The weight of A ij is feature f in feature set F′ j Competency elements i The correlation degree is used to measure the relationship between each competency factor and the characteristics of occupational behavior data; m represents the number of features in the feature set F′; I represents the number of competency factors in the total competency factor set;

[0130] S62. After each competency assessment, the weight w is adjusted according to the scoring results of the fine-grained feedback model. i Update the weight w i The update formula of is shown in the following formula (12):

[0131]

[0132] In formula (12), α is the learning rate, which is used to control the update speed of the weights; f p (x,y i ) is a fine-grained feedback model for competency elements e i Ratings; is the competency factor e at time t i The weight of

[0133] S63. The final total competency score of each employee is output as the evaluation result, as shown in the following formula (13):

[0134]

[0135] In formula (13), It is the final weight after T feedback adjustments, m represents the number of features in the feature set F′, and I represents the number of competency elements in the total competency element set.

[0136] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A competency extraction and assessment method based on employee professional behavior data, characterized by: The specific steps include: (1) Collect employee career behavior data and perform data preprocessing to obtain data set F; (2) For the data set F, a clustering algorithm is used to identify the potential patterns of employees’ professional behavior. The number of clusters is set to k, and the clustering result is C = {c1, c2, ..., c i , ..., c k }, then extract the features that have a significant impact on competency assessment to form a feature set F′, so that (3) Integrate the dataset F to form a comprehensive dataset Among them, D i Denotes the dataset D integrated The i-th data source in, m is the data set D integrated The number of data sources in (4) Construct a competency factor model and use the analytic hierarchy process and hierarchical semantic ID to define the total competency factor set E of each employee as a hierarchical structure of competency factors; (5) Conduct refined evaluation through a fine-grained feedback model, iteratively optimize the output of the fine-grained feedback model through a self-reflection mechanism, and further iteratively optimize the fine-grained feedback model using a reinforcement learning algorithm. Finally, in each evaluation cycle, dynamically adjust the total set of competency elements in the hierarchy based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm. (6) Based on the dynamically adjusted competency factor model and the real-time optimized fine-grained feedback model, each employee’s competency is evaluated in real time to obtain the employee’s real-time competency evaluation results.

2. The method for extracting and evaluating competency based on employee professional behavior data driven according to claim 1, characterized in that: The steps of collecting employee occupational behavior data and performing data preprocessing to obtain data set F are as follows: S11. First, collect the career behavior data of each employee. Each employee corresponds to an initial data set D = {d1, d2, ..., d i , ..., d n }, where each d i represents an employee's professional behavior record, and n is the number of employee's professional behavior records; S12. Preprocess the initial data set D of each employee to obtain a data set F; wherein the preprocessing process includes missing value processing, outlier detection and data standardization processing.

3. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 2, characterized in that: The missing value processing is to set the missing value ratio p. After the missing value processing, the integrity of the data set is expressed as D′=D\{d i |d i is missing}; The outlier detection is performed by using statistical methods, setting a threshold z, and defining the outlier value as |z i |>Z; The data normalization process is performed using the formula To transform the original data x, we get the standardized data x′; where μ represents the mean and σ represents the standard deviation.

4. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 1, characterized in that: The dataset D integrated Divide into training set and test set, set the training set ratio to r, then the training set is D train =D integrated ×r, the test set is D test =D integrated ×(1-r).

5. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 2, characterized in that: The specific steps of extracting features that have a significant impact on competency assessment and forming a feature set F′ are as follows: S21. Collect the employee career behavior data after data preprocessing and construct a set P = {p1, p2, ..., p n }, where n is the number of employee professional behavior records. Then, the text data of employee professional behavior records is analyzed through natural language processing technology to extract keywords and phrases related to competency elements, thus forming a data set F = {f1, f2, ..., f m }, where m is the number of features in the dataset F; S22. Use machine learning algorithms to evaluate the contribution of each feature of the dataset F to the competency assessment results, by setting the feature importance threshold t f , thus selecting the feature set F′={f j |importance(f j )>t f }, f j is a feature in the feature set F′.

6. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 5, characterized in that: In the competency factor model, each competency factor in the initial total competency factor set corresponds to the collected employee professional behavior data. The total competency factor set E of each employee is defined as a hierarchical structure of competency factors by using the hierarchical analysis method and hierarchical semantic ID. The specific steps are as follows: S41. Define the hierarchical structure of the total competency factor set E: The total competency factor set E is divided into different levels, each level E j They all contain a set of elements, as shown in the following formula (1): In formula (1), J represents the number of levels, E j represents the set of competency factors at the jth level, and ∪ is the union symbol; S42. Generate hierarchical semantic IDs of competency elements. Each competency element e i A hierarchical semantic ID is assigned to each competency element, which clarifies the position of the competency element in the competency hierarchy. See the following formula (2) for details: In formula (2), L represents the level depth, ID l (e i ) represents competency factor e i Hierarchical semantic ID at level l; S43. Competency elements i The weight w i The analytic hierarchy process is used to determine the specific formula (4): In formula (4), p il It is a competency element i Compared with other competency elements l Ratings; S44, Inter-level Correlation C jl It is calculated by considering the correlation between competency elements at different levels to assess the importance and mutual influence of different levels, as shown in the following formula (5): In formula (5), E l represents the set of competency elements at level l, e k represents the competency elements in the competency element set of the first level, w k Representative competency elements k The weight of correlation(E j , E l ) represents the correlation between the j-th and l-th level competency factor sets.

7. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 6, characterized in that: The generation of the hierarchical semantic ID is based on the function f l , the expression is shown in the following formula (3): ID l (it i )=f l (it i ,On i )) (3); In formula (3), e i is the competency factor, P(e i ) is the competency factor e i Attributes.

8. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 6, characterized in that: The fine-grained feedback model is used for refined evaluation, and the output of the fine-grained feedback model is iteratively optimized through a self-reflection mechanism. The fine-grained feedback model is further iteratively optimized using a reinforcement learning algorithm. Finally, in each evaluation cycle, the total set of competency elements of the hierarchical structure is dynamically adjusted based on the score of the fine-grained feedback model and the optimization result of the reinforcement learning algorithm. The specific steps are: S51. For each employee’s career behavior data x, the fine-grained feedback model f p (x, y) corresponds to the scores of multiple dimensions, as shown in the following formula (6): f p (x,y)={Logical Correctness,Factuality,Insightfulness,CommunicationSkills,Collaboration,Adaptability} (6); In formula (6), Logical Correctness represents logical correctness, Factuality represents factuality, Insightfulness represents insight, Communication Skills represents communication skills, Collaboration represents collaboration skills, and Adaptability represents adaptability; Each dimension is scored as -1, 0, or 1, with -1 representing failure, 0 representing medium, and 1 representing success; S52, through the fine-grained feedback model f p (x, y) generates a response set D y , response set D y Each response y in i All will be fed back through the fine-grained feedback model f p (x, y) is scored, as shown in the following formula (7): In formula (7), is based on the current fine-grained feedback model f p (x, y) parameters The generated policy function, r(x, y i ) is the reward value given by the preference model; Select candidate response set D y The highest scoring response As a starting point for reflection and optimization: Based on fine-grained feedback model Generate self-reflection and improve responses to Conduct self-reflection and generate new and improved responses j , see the following formula (8) for details: In formula (8), Indicates input combined with feedback information; D z represents the improved response set; S53. After obtaining the improved response set D z Afterwards, the reinforcement learning algorithm DPO is applied to further iteratively optimize the fine-grained feedback model to obtain R; Objective function of DPO optimization See the following formula (9): In formula (9), y + and - are positive and negative sample responses, respectively, ref is the reference strategy, β is the hyperparameter controlling the weight; σ represents the sigmoid activation function, E D In the test set D test The objective function is averaged over all samples in to evaluate the performance of the fine-grained feedback model on the dataset; S54. In each evaluation cycle, based on the score of the fine-grained feedback model and the optimization results of the reinforcement learning algorithm, the total set of competency elements of the hierarchical structure is dynamically adjusted, as shown in the following formula (10): E t+1 =g(E t ,R t ) (10); In formula (10), E t+1 is the hierarchical competency factor set at time t+1, g is the update function, E t is the hierarchical set of competency factors at time t, R t is the feedback information at time t, that is, the result obtained by further iteratively optimizing the fine-grained feedback model using the reinforcement learning algorithm DPO at time t.

9. The method for extracting and evaluating competency based on employee professional behavior data drive according to claim 8, characterized in that: The competency assessment result of each employee is the total competency score S, which is the sum of all competency factors. i The weighted sum of the corresponding feature scores; S61. First, the weight of the association matrix A and the score of the fine-grained feedback model are used to adjust the weight of each competency factor, as shown in the following formula (11): In formula (11), w i It is a competency element i The weight of A ij is feature f in feature set F′ j Competency elements i The association degree; m represents the number of features in the feature set F′; I represents the number of competency factors in the total competency factor set; S62. After each competency assessment, the weight w is adjusted according to the scoring results of the fine-grained feedback model. i Update the weight w i The update formula of is shown in the following formula (12): In formula (12), α is the learning rate, which is used to control the update speed of the weights; f p (x,y i ) is a fine-grained feedback model for competency elements e i Ratings; is the competency factor e at time t i The weight of S63. The final total competency score of each employee is output as the evaluation result, as shown in the following formula (13): In formula (13), It is the final weight after T feedback adjustments, m represents the number of features in the feature set F′, and I represents the number of competency elements in the total competency element set.

Images