Post competency model optimization method and device based on input correlation function

Abstract

The invention relates to a post competency model optimization method and device based on an input correlation function. The method comprises the following steps: calculating output data of each layer of neural network based on preset input data, network parameters and an interlayer progressive function; calculating a partial derivative of the output data with respect to preset input data; calculating the parameter change rate of each layer of neural network through a parameter change rate function based on the partial derivative and the weight matrix; calculating a correlation component of the previous I layer based on the partial derivative of the previous I layer and the weight matrix; calculating the distribution expectation of the network parameters of the (I + 1) th layer through the output data of the I th layer and the parameters of the (I + 1) th layer; calculating an input correlation function value of the (I + 1) th layer through an input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used for adjusting the optimization network, the input correlation function is introduced in the method, the correlation of the input processed by the neural network is represented, and evaluation and optimization of the optimization network design constructed by the post competency-oriented model are realized.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method and device for optimizing a job competency model based on input-related functions. Background Art

[0002] In the field of human resources services, traditional HR service platforms rely primarily on simple algorithms and database technologies to achieve basic talent and position matching and recruitment functions. However, as companies' requirements for talent selection and development continue to increase, simple technical means are no longer able to meet the complex demands of building job competency models.

[0003] Current HR service platforms on the market primarily rely on simple technical means to achieve talent and position matching and recruitment. However, these platforms have significant shortcomings in building competency models and lack solutions integrated with professional HR talent identification tools and methods. Specifically, these systems fail to fully leverage AI (artificial intelligence) algorithms and large databases to accurately build and optimize competency models for specific positions, resulting in significant subjectivity and uncertainty in the talent assessment and selection process. Summary of the Invention

[0004] In order to solve or partially solve the problems existing in the relevant technologies, the present application provides a method and device for optimizing a job competency model based on an input correlation function, which can establish an input correlation function of a neural network based on a large language model, characterize the correlation of the input after processing by the neural network, and thus realize the evaluation and optimization of the neural network design for competency model construction in the human resources service platform.

[0005] In a first aspect, the present application provides a method for optimizing a job competency model based on an input-related function. The job competency model includes an optimization network, wherein the optimization network includes an I+1-layer neural network, and each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector. The method includes: Calculating the output data of each layer of the neural network based on preset input data, the network parameters and the inter-layer progressive function; the preset input data includes position data and talent data; Calculating the partial derivative of the output data with respect to the preset input data; calculating the parameter change rate of each layer of the neural network through a parameter change rate function based on the partial derivative and the weight matrix; Calculate the correlation component of the first I layers based on the partial derivatives of the first I layers and the weight matrix; Calculate the expected distribution of the network parameters of the I+1 layer through the output data of the I layer and the parameters of the I+1 layer; An input correlation function value of the I+1th layer is calculated based on the distribution expectation and the correlation component through an input correlation function; the input correlation function value and the parameter change rate are used to adjust the optimization network.

[0006] In some embodiments, the calculating the output data of each layer of the neural network based on the preset input data, the network parameters, and the inter-layer progressive function includes: When I=0, the preset input data is processed based on the inter-layer progressive function; The processing result of the inter-layer progressive function is used as the output data.

[0007] In some embodiments, the calculating the output data of each layer of the neural network based on the preset input data, the network parameters, and the inter-layer progressive function further includes: When I is greater than 0, the weighted bias result of the preset input data in the current layer of neural network is calculated based on the output data of the previous layer of neural network and the weight matrix and the bias vector; Based on the weighted bias result, the output data of the current layer neural network is calculated through the inter-layer progressive function.

[0008] In some embodiments, calculating the parameter change rate of each layer of the neural network based on the partial derivatives and the weight matrix using a parameter change rate function includes: The partial derivative and the weight matrix are multiplied by the parameter change rate function to calculate the parameter change rate of each layer of the neural network.

[0009] In some embodiments, calculating the input correlation function value of the (I+1)th layer based on the distribution expectation and the correlation component by using the input correlation function includes: The input correlation function value is calculated by adding the distribution expectation and the correlation component.

[0010] In some embodiments, after calculating the parameter change rate of each layer of the neural network, the method further includes: Constructing a parameter change rate model based on the network parameters and the parameter change rate, wherein the parameter change rate model is used to characterize the change characteristics of the weight matrix between adjacent layers and the bias vector with respect to the preset input data; The parameter change rate model is evaluated, and the parameter change rate model is optimized based on the evaluation result to optimize the optimization network.

[0011] In some embodiments, calculating the distribution expectation of the I+1 layer parameters using the output data of the I layer and the I+1 layer parameters includes: Counting the network parameters of the optimized network, generating a parameter distribution of the network parameters, and calculating characteristic parameters of the I+1 layer network parameter distribution; The distribution expectation is calculated based on the feature parameters and the output data of the Ith layer.

[0012] In some embodiments, after calculating the input correlation function value of the (I+1)th layer based on the distribution expectation and the correlation component by using the input correlation function, the method further includes: Obtain pre-processed job data and talent data to generate a training set; Inputting the training set into the optimization network, and adjusting the weight matrix and the bias vector by a back propagation algorithm to train the optimization network; Evaluate the trained optimization network using the input-related function value to generate an evaluation result; The network parameters and structure of the optimized network are optimized based on the evaluation results.

[0013] The second aspect of the present application provides a job competency model optimization device based on input correlation function, The job competency model includes an optimization network, the optimization network includes an I+1 layer neural network, each layer of the neural network includes initialization network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector; the device includes: An output data module, configured to calculate output data of each layer of the neural network based on preset input data, the network parameters, and the inter-layer progressive function; the preset input data includes position data and talent data; A parameter change rate module, configured to calculate the partial derivative of the output data with respect to the preset input data, and calculate the parameter change rate of each layer of the neural network through the partial derivative and the weight matrix using a parameter change rate function; a correlation component module, configured to calculate the correlation components of the first I layers based on the partial derivatives of the first I layers and the weight matrix; A distribution expectation module is used to calculate the distribution expectation of the network parameters of the I+1 layer based on the output data of the I layer and the parameters of the I+1 layer; An adjustment module is used to calculate an input correlation function value of the I+1th layer through an input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network.

[0014] A third aspect of the present application provides an electronic device, including: processor; and The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method described above.

[0015] The technical solution provided by this application may have the following beneficial effects: An embodiment of the present application provides a method for optimizing a job competency model based on an input correlation function, wherein the job competency model includes an optimization network, the optimization network includes an I+1-layer neural network, each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the parameters of the network parameters include at least a weight matrix and a bias vector; the method includes: calculating the output data of each layer of the neural network based on preset input data, network parameters and an inter-layer progressive function; the preset input data includes job data and talent data; calculating the partial derivatives of the output data with respect to the preset input data; calculating the parameter change rate of each layer of the neural network through a parameter change rate function based on the partial derivatives and the weight matrix; calculating the correlation components of the first I layers based on the partial derivatives and the weight matrix of the first I layers; calculating the distribution expectation of the network parameters of the I+1 layer through the output data of the I layer and the parameters of the I+1 layer; calculating the input correlation function value of the I+1 layer through the input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network. Through the above method, an input correlation function is introduced, which can characterize the correlation of input data after being processed by the neural network, and realize the evaluation and optimization of the optimized network design for the construction of job competency model in the human resources service platform, thereby constructing a more accurate job competency model; the parameter change rate is also used to characterize the neural network. As the input data evolves in the neural network, the parameter change characteristics are changed, and the dynamic optimization of the neural network is realized, thereby improving the accuracy and computational efficiency of the model.

[0016] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail the exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0018] Figure 1 1 is a flow chart of a method for optimizing a job competency model based on an input correlation function according to an embodiment of the present application; Figure 2 1 is another flow chart of a method for optimizing a job competency model based on an input correlation function according to an embodiment of the present application; Figure 3Schematic diagram of the structure of the device for optimizing the job competency model based on the input correlation function shown in an embodiment of the present application; Figure 4 It is a structural diagram of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although the accompanying drawings illustrate embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0020] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0021] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0022] Current HR service platforms primarily rely on simple technical means to match talent with positions and facilitate recruitment. However, these platforms lack significant expertise in building competency models and solutions that integrate with professional HR talent identification tools and methods. Specifically, these systems fail to fully leverage AI algorithms and large databases to accurately build and optimize competency models for specific positions, leading to significant subjectivity and uncertainty in the talent assessment and selection process.

[0023] The present application provides a job competency model optimization method based on input correlation function, which can establish an input correlation function of a neural network based on a large language model, characterize the correlation of the input after being processed by the neural network, and thus realize the evaluation and optimization of the neural network design for competency model construction in the human resources service platform.

[0024] The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 It is a flow chart of a method for optimizing a job competency model based on input-related functions shown in an embodiment of the present application.

[0026] See also Figure 1 The job competency model includes an optimization network, the optimization network includes an I+1 layer neural network, each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the parameters of the network parameters include at least a weight matrix and a bias vector; the method includes: Step 101, calculate the output data of each layer of the neural network based on preset input data, network parameters and inter-layer progressive function; the preset input data includes position data and talent data.

[0027] The job competency model is a large language model that can be used to match talent with positions. It primarily matches the job competency data for a position with the talent's attribute data. Job competency data includes multidimensional data, including job responsibilities, skill requirements, experience requirements, project experience, and educational background. For example, personal information includes age, gender, and educational background (education, major, and university of graduation). Work experience includes job history, length of work, project experience, performance evaluations, and promotion history. Skills and abilities include professional skills (such as programming, design, and management), language proficiency, and certifications or qualifications. Behavioral characteristics include personality tests, teamwork skills, and leadership assessments. Basic job information includes job title, department, rank, and salary range. Responsibilities and requirements include job description, core responsibilities, and key performance indicators (KPIs). Competency requirements include essential skills (such as "proficient in Python"), educational requirements, years of work experience, and certifications required for specific positions (such as an accounting qualification). Behavior and culture fit: Teamwork needs and compatibility with corporate culture (e.g., adaptability to a fast-paced work style or frequent travel). Talent attribute data includes multi-dimensional data, including educational background, work experience, skill certifications, project experience, and more.

[0028] In this embodiment, an optimization network is constructed within the job competency model. The number of neural network layers in the optimization network is set to I+1. The first layer is the input layer, responsible for receiving input data, which can receive an input vector consisting of job data and talent data. The middle layer is the hidden layer, responsible for nonlinear data transformation and feature extraction. The last layer is the output layer, responsible for outputting the processing results, such as the talent-job matching results or matching scores. I is a variable parameter that can be adjusted based on the complexity and needs of the actual application. For example, for a simpler job competency model, a smaller I value can be selected; for a more complex job competency model, the I value may need to be increased to improve the optimization network's fitting ability.

[0029] Each layer of a neural network contains several neurons, and the number of neurons in each layer can vary. The number of neurons in the i-th layer is set to n_i, where i∈[0,I]. For each neuron in the neural network, an activation function h is set to implement nonlinear data transformation. The number of neurons in the input layer, n_1, is determined based on the length of the input vector. The number of neurons in the input layer, n_1, can be the same as the length of the input vector. For each hidden layer, the number of neurons in that layer, n_i, is set based on the complexity of the network design and the characteristics of the data. The number of neurons in the output layer is set based on the specific task requirements. For example, if the output is a matching score, it may be one neuron; if the output is a list of recommended positions, the number of neurons may be related to the number of job categories. Each layer of the neural network has a corresponding output-related function.

[0030] In each layer of the neural network, network parameters and inter-layer progressive functions can be initialized. Network parameters include at least the weight matrix W_i, the bias vector b_i, and possibly scaling factors. They may also include the number of layers and neurons in the optimized network. The weight matrix W_i of layer i is set to a matrix with n_i rows and n_(i+1) columns, representing the connection weights between the neurons in layer i and the neurons in layer i+1. The bias vector b_i of layer i is set to a vector with n_(i+1) columns, representing the bias values ​​for the neurons in layer i+1. Scaling factors include α_i for the weight matrix W_i and β_i for the bias vector b_i, which are used to adjust the influence of the weight matrix W_i and the bias vector b_i. The weight matrix W_i and the bias vector b_i can be randomly generated or initialized according to a predefined rule.

[0031] The preset input data is set to similar job data y and talent data x. For example, both job data y and talent data x are related to educational background, or both job data y and talent data x are related to work experience. The preset input data can be in any form, such as images, text, or numerical data. The preset input data is forward propagated through the optimization network. Based on the preset input data, weighted bias functions, and inter-layer progressive functions, the output data of each neural network layer is calculated. In the optimization network, the output data of each layer serves as the input data of the next layer, thus enabling the transfer of parameters between layers.

[0032] Step 102: Calculate the partial derivative of the output data with respect to the preset input data; calculate the parameter change rate of each layer of the neural network based on the partial derivative and the weight matrix through the parameter change rate function. In an embodiment of the present application, a parameter change rate function is also set in the optimization network. The parameter change rate function is used to quantify the change characteristics of the weight matrix W and the bias vector b between adjacent neural network layers in the optimization network as the preset input data is transmitted.

[0033] After obtaining the output data, we can calculate the partial derivatives of the output data with respect to the position data y and the talent data x. Based on these partial derivatives and the weight matrix, we can use the parameter change rate function to calculate the parameter change rate of each neural network layer. This parameter change rate function characterizes the degree of coupling between network parameters in the optimized network layers. When the number of neurons is large, network parameters (such as the weight matrix W and the bias vector b) follow a normal distribution. Based on this characteristic, in practical implementation, it is necessary to calculate the parameter change rate of each neural network layer. This can be achieved by recording the changes in each layer's network parameters (weight matrix W and bias vector b) between adjacent iterations or time steps during the optimization network training process.

[0034] Step 103: Calculate the correlation components of the first I layers based on the partial derivatives and weight matrix of the first I layers.

[0035] By calculating the partial derivatives of the output data of the first I layers and combining them with the weight matrix, we can obtain the correlation components generated by the first I layers. The specific calculation method is shown in formula (1): (1) Here, h[g_I(x)] and h[g_I(y)] represent the output data of the first I layers of the neural network after processing the preset input data x and y; W_I represents the weight matrix of layer I; and Σ represents the summation of the matrix elements. In practice, the automatic differentiation function in a deep learning framework can be used to calculate partial derivatives, and matrix operations can be performed using a matrix operation library (such as NumPy). The calculated partial derivatives and weight matrix are then substituted into the correlation component calculation formula to obtain the correlation component generated based on the first I layers.

[0036] Step 104: Calculate the expected distribution of the parameters of the I+1th layer using the output data of the Ith layer and the network parameters of the I+1th layer.

[0037] According to the output data h[g_I(x)] and h[g_I(y)] of the I layer and the network parameters of the I+1 layer, the distribution expectation of the parameters of the I+1 layer can be calculated. The distribution expectation can reflect the influence of the random distribution characteristics of the network parameter mapping between the I+1 layer neural network layers on the preset input data correlation. By calculating the distribution expectation of the random distribution characteristics of the network parameter mapping between the I+1 layer neural network layers, the contribution of this part to the preset input data correlation is obtained.

[0038] Step 105 : Calculate the input correlation function value of the I+1th layer through the input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network.

[0039] In the embodiment of the present application, an input correlation function Φ(x, y) is also set in the optimization network, which is a matrix of n_I×n_I, where n_I represents the number of neurons in the last layer of the optimization network, and the input correlation is used to express the correlation between the preset input data x and y after being processed by the optimization network.

[0040] Based on the calculated parameter change rate, the optimized network parameters can be adjusted and optimized. Specifically, the network parameters can be continuously adjusted through optimization algorithms such as gradient descent to make the output of the optimized network closer to the actual target value.

[0041] By analyzing the input correlation function Φ(x, y) of the optimization network, the optimization network design can be optimized. When the value of Φ(x, y) is large, it indicates that the preset input data x and y are highly correlated in the processing process of the optimization network. In this case, it is possible to consider reducing the number of layers in the optimization network or adjusting the connection method of neurons to reduce computational complexity. When the value of Φ(x, y) is small, it indicates that the preset input data x and y are less correlated in the processing process of the neural network. In this case, it is possible to consider increasing the number of layers in the optimization network or increasing the number of neurons to improve the accuracy of the model. Therefore, upper and lower limits can be set for the value of Φ(x, y), thereby determining when to reduce and increase the complexity of the optimization network. By inputting the talent's personal information, work experience, educational background, etc. into the optimization network, and using the input correlation function Φ(x, y) to calculate the correlation between the talent and the position, a more accurate job competency model can be constructed.

[0042] In an embodiment of the present application, a job competency model can be constructed by the following method: obtaining original data, the original data including talent characteristics and job characteristics; calculating a first distance vector from talent characteristics to job characteristics, and a second distance vector from job characteristics to talent characteristics; calculating recovery variables based on talent characteristics, job characteristics, the first distance vector and the second distance vector; the recovery variables include recovery variables of talent characteristics and recovery variables of job characteristics; calculating mutual information based on talent characteristics, job characteristics, the first distance vector, the second distance vector, the recovery variables of talent characteristics and the recovery variables of job characteristics; calculating conditional entropy based on talent characteristics, job characteristics and the first distance vector; optimizing the first distance vector and the second distance vector based on the mutual information and conditional entropy; and constructing a job competency model based on the optimized first distance vector and the second distance vector.

[0043] An embodiment of the present application provides a method for optimizing a job competency model based on an input correlation function, wherein the job competency model includes an optimization network, the optimization network includes an I+1-layer neural network, and each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector; the method includes: calculating the output data of each layer of the neural network based on preset input data, network parameters and an inter-layer progressive function; the preset input data includes job data and talent data; calculating the partial derivatives of the output data with respect to the preset input data; calculating the parameter change rate of each layer of the neural network through a parameter change rate function based on the partial derivatives and the weight matrix; calculating the correlation components of the first I layers based on the partial derivatives and the weight matrix of the first I layers; calculating the distribution expectation of the network parameters of the I+1 layer through the output data of the I layer and the parameters of the I+1 layer; calculating the input correlation function value of the I+1 layer through the input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network. Through the above method, an input correlation function is introduced, which can characterize the correlation of input data after being processed by the neural network, and realize the evaluation and optimization of the optimized network design for the construction of job competency model in the human resources service platform, thereby constructing a more accurate job competency model; the parameter change rate is also used to characterize the neural network. As the input data evolves in the neural network, the parameter change characteristics are changed, and the dynamic optimization of the neural network is realized, thereby improving the accuracy and computational efficiency of the model.

[0044] Figure 2 It is a flow chart of a method for optimizing a job competency model based on input-related functions shown in an embodiment of the present application.

[0045] See also Figure 2The job competency model includes an optimization network, the optimization network includes an I+1 layer neural network, each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector; the method includes: Step 201, calculate the output data of each layer of the neural network based on preset input data, network parameters and inter-layer progressive function; the preset input data includes position data and talent data.

[0046] The job competency model is a large language model that can be used to match talents and jobs, mainly matching the job competency data of the job and the talent attribute data of the talent, wherein the job competency data includes multidimensional data, which may include job responsibilities, skill requirements, experience requirements, project experience, academic qualifications, etc. In the embodiment of the present application, an optimization network is constructed in the job competency model, and the number of neural network layers of the optimization network is set to I+1, wherein the first layer is the input layer, which is responsible for receiving input data and can receive input vectors composed of job data and talent data of the job; the middle layer is the hidden layer, which is responsible for nonlinear transformation and feature extraction of the data; the last layer is the output layer, which is responsible for outputting the processing results, such as the matching results or matching scores between talents and jobs. Among them, I is a variable parameter that can be adjusted according to the complexity and needs of the actual application. For example, for a simpler job competency model, a smaller I value can be selected; for a complex job competency model, it may be necessary to increase the I value to improve the fitting ability of the optimization network.

[0047] Each layer of a neural network contains several neurons, and the number of neurons in each layer can vary. The number of neurons in the i-th layer is set to n_i, where i∈[0,I]. For each neuron in the neural network, an activation function h is set to implement nonlinear transformations of the data. The number of neurons in the input layer, n_0, is determined based on the length of the input vector. The number of neurons in the input layer, n_0, can be the same as the length of the input vector. For each hidden layer, the number of neurons in that layer, n_i, is set based on the complexity of the network design and the characteristics of the data. The number of neurons in the output layer is set based on the specific task requirements. For example, if the output is a matching score, it may be one neuron; if the output is a list of recommended positions, the number of neurons may be related to the number of job categories. Each layer of the neural network has a corresponding output-related function.

[0048] In each layer of the neural network, network parameters and inter-layer progressive functions can be initialized. Network parameters include at least the weight matrix W_i, the bias vector b_i, and possibly scaling factors. They may also include the number of layers and neurons in the optimized network. The weight matrix W_i of layer i is set to a matrix with n_i rows and n_(i+1) columns, representing the connection weights between the neurons in layer i and the neurons in layer i+1. The bias vector b_i of layer i is set to a vector with n_(i+1) columns, representing the bias values ​​for the neurons in layer i+1. Scaling factors include α_i for the weight matrix W_i and β_i for the bias vector b_i, which are used to adjust the influence of the weight matrix W_i and the bias vector b_i. The weight matrix W_i and the bias vector b_i can be randomly generated or initialized according to a predefined rule.

[0049] The preset input data is set to similar job data y and talent data x. For example, both job data y and talent data x are related to educational background, or both job data y and talent data x are related to work experience. The preset input data can be in any form, such as images, text, or numerical data. The preset input data is forward propagated through the optimization network. Based on the preset input data, weighted bias functions, and inter-layer progressive functions, the output data of each neural network layer is calculated. In the optimization network, the output data of each layer serves as the input data of the next layer, thus enabling the transfer of parameters between layers.

[0050] In an optional embodiment, step 201 includes: When I=0, the preset input data is processed based on the inter-layer progressive function; The processing results of the inter-layer progressive function are used as output data.

[0051] When I = 0, and the number of neural network layers in the optimization network is 1, the optimization network has only input and output layers, with no intermediate layers. The pre-set input data x and y are processed by the inter-layer progressive function to produce f_0(x) = x and f_0(y) = y. The result of the inter-layer progressive function is then used as the output data for the next neural network layer.

[0052] In an optional embodiment, step 201 further includes: When I is greater than 0, the weighted bias result of the preset input data in the current layer of neural network is calculated based on the output data, weight matrix, and bias vector of the previous layer of neural network; Based on the weighted bias result, the output data of the current layer neural network is calculated through the inter-layer progressive function.

[0053] When I is greater than 0 and the number of neural network layers is greater than 1, the optimization network contains multiple intermediate layers, and the output data of each layer will be used as the input data of the next layer, thereby realizing the transfer of parameters between layers. In each layer, the preset input data is processed by the weight matrix W_i, the bias vector b_i, and the scaling factors α_i and β_i to obtain the output data of the layer. Specifically, for the preset input data x, the output data of the i-1 layer is f_(i-1)(x). According to the output result f_(i-1)(x) of the i-1 layer, the weight matrix, and the bias vector, the weighted bias result of the previous layer of the neural network (i-1 layer) is calculated. The weighted bias result can be calculated according to formula (2): (2) The output of the current neural network (layer i) can be calculated based on the weighted bias result g_(i-1)(x) of the previous neural network and the activation function h. The output data can be calculated according to formula (3): (3) For the preset input data y, the processing is the same.

[0054] Step 202, calculate the partial derivative of the output data with respect to the preset input data; multiply the partial derivative and the weight matrix by the parameter change rate function to calculate the parameter change rate of each layer of the neural network.

[0055] In order to explore the coupling relationship between neural network layers, the parameter change rate between neural network layers can be calculated to characterize the degree of coupling between layer parameters. That is, according to the defined parameter change rate function, the change characteristics of the network parameters between layers as the input data evolves in the neural network are described. After calculating the output data, the partial derivative of the output data with respect to the preset input data can be calculated. Then, the partial derivative and the weight matrix are multiplied by the parameter change rate function to calculate the parameter change rate Δ_Wb[g_I(x)] of each layer of the neural network. It can be calculated using formula (4): (4) Where Σ represents the sum of the multiplication of the elements in the matrix; h[g_I(x)] represents the function of nonlinear transformation of the weighted bias result g_I(x) of the optimized network; W_I represents the weight matrix of the Ith layer of the optimized network; Represents the partial derivative of the function h[g_I(x)] with respect to the input x.

[0056] This formula represents the parameter change rate function of a certain layer of the neural network, that is, as the input data x evolves, the change characteristics of the neural network parameters W and b can be obtained by calculating the sum of the partial derivative of the function h[g_I(x)] with respect to the input x and the product of the weight matrix W_I.

[0057] Based on the calculated parameter change rate, the optimized network parameters can be adjusted and optimized. Specifically, the network parameters can be continuously adjusted through optimization algorithms such as gradient descent to make the output of the optimized network closer to the actual target value.

[0058] In an optional embodiment, after step 202, the method further includes: A parameter change rate model is constructed based on network parameters and parameter change rates. The parameter change rate model is used to characterize the change characteristics of weight matrices and bias vectors between adjacent layers with respect to preset input data. The parameter change rate model is evaluated, and the parameter change rate model is optimized based on the evaluation result to optimize the optimization network.

[0059] Based on the definition of the parameter change rate function and the random distribution characteristics of network parameters between neural network layers, a parameter change rate function model is constructed. This model is used to describe the changing characteristics of the network parameter weight matrix W and bias vector b between adjacent layers during the evolution of the preset input data x.

[0060] Specifically, machine learning regression algorithms (such as linear regression and nonlinear regression) can be used to fit the parameter change rate function. During the fitting process, the preset input data x, the number of neural network layers I, and the parameter change rate between adjacent layers are used as input features, and the actual change in network parameters is used as the target variable for training. The parameter change rate model obtained through training can be used to predict the changing characteristics of network parameters under given input data and neural network structure.

[0061] After the parameter change rate function model is constructed, it needs to be optimized to improve its prediction accuracy. This can be achieved by adjusting the parameters of the parameter change rate function model (adjusting the coefficients of the regression algorithm, regularization terms, etc.). At the same time, methods such as cross-validation can be used to evaluate the performance of the parameter change rate function model, and the parameter change rate function model can be further adjusted and optimized based on the evaluation results.

[0062] Finally, the optimized parameter change rate function model is applied to actual network optimization tasks. By calculating the rate of change of each layer's network parameters and adjusting the network parameters accordingly, the optimized network can be optimized. To verify the optimization effect, the performance of the optimized network can be evaluated using metrics such as accuracy, recall, and F1 score, and compared with the optimized network optimized without the parameter change rate function.

[0063] Step 203 : Calculate the correlation components of the first I layers based on the partial derivatives and weight matrix of the first I layers.

[0064] By calculating the partial derivatives of the output data of the first I layers and combining them with the weight matrix, we can obtain the correlation components generated by the first I layers. For the specific calculation method, refer to formula (1).

[0065] Step 204 , statistically optimize the network parameters of the network, generate parameter distribution of the network parameters, and calculate characteristic parameters of the parameter distribution of the I+1th layer.

[0066] When the number of neurons is large, optimizing network parameters, such as the mapping between the weight matrix and the bias vector, follows a certain random distribution. Based on this characteristic, the network parameters can be statistically optimized to generate a parameter distribution for the network parameters. This parameter distribution describes the statistical regularity of the inter-layer network parameters. To simplify calculations, it is possible to assume that the inter-layer network parameters follow a common distribution type, such as a normal distribution or a uniform distribution. Based on this parameter distribution, parameter estimation methods in statistics (such as moment estimation, maximum likelihood estimation, and Bayesian estimation) can be used to estimate the characteristic parameters (such as the mean and variance) of the random distribution characteristics of the inter-layer parameter mapping.

[0067] Step 205: Calculate the distribution expectation based on the feature parameters and the output data of the Ith layer.

[0068] Substituting the characteristic parameters into the calculation formula of distribution expectation, we can obtain the distribution expectation E of the random distribution characteristics of the inter-layer parameter mapping of the neural network in the I+1 layer. The distribution expectation E can be calculated by formula (5): (5) Among them, α_I and β_I represent the characteristic parameters of the random distribution characteristics of the parameter mapping between the I+1th layer of the neural network respectively; [h[g_I(x)]]T represents the transposed matrix of h[g_I(x)] and h[g_I(y)] respectively; E{} represents the expectation operation.

[0069] Step 206: The input correlation function is used to add the distribution expectation and the correlation component to calculate the input correlation function value of the I+1th layer; the input correlation function value and the parameter change rate are used to adjust the optimization network.

[0070] For optimizing a multi-layer neural network (I>0), it is necessary to calculate the output data of each layer through the forward propagation algorithm and record the relevant weight matrix and bias vector of each layer. Using the forward propagation results and weight matrix of the neural network, the input correlation function Φ_ i (x,y) i∈[0,I] of each layer can be recursively calculated. The recursive process may involve complex matrix operations and the application of activation functions. After obtaining the input correlation functions of all intermediate layers, the final optimized network input correlation function Φ(x,y) can be determined based on the output data of the last layer and the correlation calculation formula.

[0071] Specifically, the input correlation function is used to add the expected random distribution characteristics of the correlation components generated by the first I layers and the inter-layer parameter mapping of the neural network in the I+1 layer to obtain the input correlation function value Φ_(I+1) of the I+1 layer. The input correlation function value Φ_(I+1) can be calculated by formula (6): (6) When I = 0 and the number of neural network layers in the optimization network is 1, the optimization network has only input and output, and no intermediate layers. The results of the pre-set input data x and y processed by the inter-layer progressive function are f_0(x) = x and f_0(y) = y. The input correlation function value is set to be equal to the output correlation function value Ψ_1(x, y) based on the same input. The calculation formula is shown in the following formula (7): (7) Among them, α_0 and β_0 are the parameter scaling factors related to the weight matrix and bias vector of the first layer of the neural network, and x^T represents the transpose of x.

[0072] By analyzing the input correlation function Φ(x,y) of the optimization network, the design of the optimization network can be optimized. When Φ(x,y) is large, it indicates that the processing of the preset input data x and y in the optimization network is highly correlated. In this case, reducing the number of layers in the optimization network or adjusting the neuron connections can be considered to reduce computational complexity. When Φ(x,y) is small, it indicates that the processing of the preset input data x and y in the neural network is less correlated. In this case, increasing the number of layers in the optimization network or increasing the number of neurons can be considered to improve model accuracy. Therefore, upper and lower limits can be set for Φ(x,y) to determine when to reduce or increase the complexity of the optimization network. By inputting personal information, work experience, and educational background into the optimization network, the input correlation function Φ(x,y) is used to calculate the correlation between the talent and the position, thereby constructing a more accurate job competency model.

[0073] This example constructs a neural network input correlation function based on "coupling + distribution." This method constructs a model for the transfer and distribution of network parameters between neural network layers, combined with an inter-layer rate of change model, to further explore the coupling relationship between layers. Ultimately, it develops a progressive model of the input correlation equivalent function based on "coupling + distribution." This innovation not only enriches the theoretical foundation of neural network models but also provides more accurate and reliable model support for practical applications.

[0074] In an optional embodiment, after step 206, the method further includes: Obtain pre-processed job data and talent data to generate a training set; The training set is input into the optimized network, and the weight matrix and bias vector are adjusted through the back-propagation algorithm to train the optimized network; Evaluate the trained optimized network by inputting relevant function values ​​and generate evaluation results; The network parameters and structure of the optimization network are optimized based on the evaluation results.

[0075] In this embodiment, job data, including key information such as job responsibilities, skill requirements, and work experience, can be collected and organized into a structured data format. Talent data, including educational background, work experience, skill certificates, and personal characteristics, can be collected and organized into a structured data format that matches the job data. The collected data is pre-processed through cleaning, denoising, and standardization to ensure data quality and consistency. A training set is generated based on the pre-processed job and talent data.

[0076] The optimized network is trained using the training set data through the forward propagation algorithm. Specifically, the weight matrix and the bias vector are adjusted through the back propagation algorithm to train the optimized network.

[0077] During or after training, the input correlation function Φ(x, y) can be used to evaluate the correlation between input data and optimize the network based on the evaluation results. For example, the network structure and parameters can be adjusted, such as the number of layers, number of neurons, or activation function, to improve the neural network's ability to fit the competency model and its matching accuracy.

[0078] Apply the optimized network to a real-world job competency model matching task and verify its performance. Use metrics such as precision, recall, and F1 score to evaluate the performance of the optimized network and compare it with traditional matching methods. Based on the verification results and actual application requirements, further adjust and optimize the optimized network to improve its effectiveness in real-world applications.

[0079] The embodiment of the present application provides a job competency model optimization method based on an input correlation function. The input correlation function is introduced to characterize the correlation of input data after being processed by a neural network, thereby realizing the evaluation and optimization of the optimized network design for the construction of a job competency model in a human resources service platform, thereby constructing a more accurate job competency model. The method also characterizes the neural network through the parameter change rate, and realizes the change characteristics of the network parameters as the input data evolves in the neural network, thereby realizing dynamic optimization of the neural network and improving the accuracy and computational efficiency of the model.

[0080] Figure 3It is a structural diagram of a job competency model optimization device based on input correlation functions shown in an embodiment of the present application.

[0081] See also Figure 3 The job competency model includes an optimization network, which includes an I+1 layer neural network. Each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector. The device includes: Output data module 301, used to calculate the output data of each layer of the neural network based on preset input data, network parameters and inter-layer progressive function; the preset input data includes position data and talent data; The parameter change rate module 302 is used to calculate the partial derivative of the output data with respect to the preset input data, and calculate the parameter change rate of each layer of the neural network through the partial derivative and the weight matrix using the parameter change rate function; A correlation component module 303 is configured to calculate the correlation components of the first I layers based on the partial derivatives of the first I layers and the weight matrix; A distribution expectation module 304 is configured to calculate the distribution expectation of the network parameters of the I+1 layer using the output data of the I layer and the parameters of the I+1 layer; The adjustment module 305 is used to calculate the input correlation function value of the I+1th layer through the input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network.

[0082] In an optional embodiment, the output data module 301 includes: The processing submodule is used to process the preset input data based on the inter-layer progressive function when I=0; The first output data submodule is used to use the processing result of the inter-layer progressive function as output data.

[0083] The output data module 301 also includes: The weighted bias submodule is used to calculate the weighted bias result of the preset input data in the current layer of the neural network based on the output data, weight matrix, and bias vector of the previous layer of the neural network when I is greater than 0; The second output data submodule is used to calculate the output data of the current layer neural network through the inter-layer progressive function based on the weighted bias result.

[0084] The parameter change rate module 302 includes: The parameter change rate submodule is used to multiply the partial derivative and the weight matrix through the parameter change rate function to calculate the parameter change rate of each layer of the neural network.

[0085] The adjustment module 305 includes: The input correlation function submodule is used to calculate the input correlation function value by adding the distribution expectation and correlation components.

[0086] In an optional embodiment, the device further comprises: A construction module is used to construct a parameter change rate model based on network parameters and parameter change rates. The parameter change rate model is used to characterize the change characteristics of the weight matrix and bias vector between adjacent layers with respect to preset input data; The first evaluation module is used to evaluate the parameter change rate model and optimize the parameter change rate model based on the evaluation result to optimize the optimization network.

[0087] The distribution expectation module 304 includes: The statistical submodule is used to statistically optimize the network parameters of the network, generate the parameter distribution of the network parameters, and calculate the characteristic parameters of the I+1 layer network parameter distribution; The distribution expectation submodule is used to calculate the distribution expectation based on the feature parameters and the output data of the I layer.

[0088] The device also includes: The training set module is used to obtain pre-processed job data and talent data to generate a training set; The training module is used to input the training set into the optimization network and adjust the weight matrix and bias vector through the back-propagation algorithm to train the optimization network; The second evaluation module is used to evaluate the trained optimization network by inputting relevant function values ​​and generate evaluation results; The optimization module is used to optimize the parameters and structure of the optimization network based on the evaluation results.

[0089] An embodiment of the present application provides a job competency model optimization device based on an input correlation function. The input correlation function is introduced to characterize the correlation of input data after being processed by a neural network, thereby realizing the evaluation and optimization of the optimized network design for job competency model construction in the human resources service platform, thereby constructing a more accurate job competency model; the parameter change rate is also used to characterize the neural network, and as the input data evolves in the neural network, the changing characteristics of the network parameters are changed, thereby realizing dynamic optimization of the neural network and improving the accuracy and computational efficiency of the model.

[0090] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0091] According to an embodiment of the present application, any multiple modules among the output data module 301, the parameter change rate module 302, the correlation component module 303, the distribution expectation module 304, and the adjustment module 305 can be combined into a single module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to an embodiment of the present disclosure, at least one of the output data module 301, the parameter change rate module 302, the correlation component module 303, the distribution expectation module 304, and the adjustment module 305 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware by any other reasonable means of integrating or packaging circuits, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the output data module 301 , the parameter change rate module 302 , the correlation component module 303 , the distribution expectation module 304 and the adjustment module 305 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0092] Figure 4 It is a structural diagram of an electronic device shown in an embodiment of the present application.

[0093] See also Figure 4 , the electronic device 400 includes a memory 410 and a processor 420.

[0094] The processor 420 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0095] Memory 410 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage. ROM may store static data or instructions required by processor 420 or other computer modules. Permanent storage may be a readable and writable storage device. Permanent storage may be a non-volatile storage device that retains stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device utilizes a mass storage device (e.g., a magnetic or optical disk, flash memory). In other embodiments, the permanent storage device may be a removable storage device (e.g., a floppy disk, optical drive). System memory may be a readable and writable storage device or a volatile readable and writable storage device, such as dynamic random access memory (DRAM). System memory may store some or all instructions and data required by the processor during operation. Furthermore, memory 410 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (e.g., DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), as well as magnetic disks and / or optical disks. In some embodiments, the memory 410 may include a readable and / or writable removable storage device, such as a compact disc (CD), a read-only digital versatile disc (e.g., DVD-ROM, double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves and transient electronic signals transmitted wirelessly or wired.

[0096] The memory 410 stores executable codes. When the executable codes are processed by the processor 420 , the processor 420 may execute part or all of the above-mentioned methods.

[0097] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0098] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0099] The embodiments of the present application have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to the technology in the market, or to enable other persons skilled in the art to understand the embodiments disclosed herein.

Claims

1. A job competency model optimization method based on input correlation function, characterized in that: The job competency model includes an optimization network, the optimization network includes an I+1 layer neural network, each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector; the method includes: Calculating the output data of each layer of the neural network based on preset input data, the network parameters and the inter-layer progressive function; the preset input data includes position data and talent data; Calculating the partial derivative of the output data with respect to the preset input data; calculating the parameter change rate of each layer of the neural network through a parameter change rate function based on the partial derivative and the weight matrix; Calculate the correlation component of the first I layers based on the partial derivatives of the first I layers and the weight matrix; Calculate the expected distribution of the network parameters of the I+1 layer through the output data of the I layer and the parameters of the I+1 layer; An input correlation function value of the I+1th layer is calculated based on the distribution expectation and the correlation component through an input correlation function; the input correlation function value and the parameter change rate are used to adjust the optimization network.

2. The method according to claim 1, characterized in that The calculating the output data of each layer of the neural network based on the preset input data, the network parameters and the inter-layer progressive function includes: When I=0, the preset input data is processed based on the inter-layer progressive function; The processing result of the inter-layer progressive function is used as the output data.

3. The method according to claim 2, characterized in that The calculating of the output data of each layer of the neural network based on the preset input data, the network parameters and the inter-layer progressive function also includes: When I is greater than 0, the weighted bias result of the preset input data in the current layer of neural network is calculated based on the output data of the previous layer of neural network and the weight matrix and the bias vector; Based on the weighted bias result, the output data of the current layer neural network is calculated through the inter-layer progressive function.

4. The method according to claim 1, wherein The calculating the parameter change rate of each layer of the neural network based on the partial derivative and the weight matrix through a parameter change rate function includes: The partial derivative and the weight matrix are multiplied by the parameter change rate function to calculate the parameter change rate of each layer of the neural network.

5. The method according to claim 1, wherein The calculating the input correlation function value of the (I+1)th layer by inputting the correlation function based on the distribution expectation and the correlation component includes: The input correlation function value is calculated by adding the distribution expectation and the correlation component.

6. The method according to claim 4, characterized in that After calculating the parameter change rate of each layer of the neural network, the method further includes: Constructing a parameter change rate model based on the network parameters and the parameter change rate, wherein the parameter change rate model is used to characterize the change characteristics of the weight matrix between adjacent layers and the bias vector with respect to the preset input data; The parameter change rate model is evaluated, and the parameter change rate model is optimized based on the evaluation result to optimize the optimization network.

7. The method according to claim 1, characterized in that Calculating the distribution expectation of the I+1th layer parameters using the output data of the Ith layer and the I+1th layer parameters includes: Counting the network parameters of the optimized network, generating a parameter distribution of the network parameters, and calculating characteristic parameters of the I+1 layer network parameter distribution; The distribution expectation is calculated based on the feature parameters and the output data of the Ith layer.

8. The method according to claim 1, characterized in that After calculating the input correlation function value of the (I+1)th layer based on the distribution expectation and the correlation component by inputting the correlation function, the method further includes: Obtain pre-processed job data and talent data to generate a training set; Inputting the training set into the optimization network, and adjusting the weight matrix and the bias vector by a back propagation algorithm to train the optimization network; Evaluate the trained optimization network using the input-related function value to generate an evaluation result; The network parameters and structure of the optimized network are optimized based on the evaluation results.

9. A device for optimizing a job competency model based on an input correlation function, characterized in that: The job competency model includes an optimization network, the optimization network includes an I+1 layer neural network, each layer of the neural network includes initialized network parameters and an inter-layer progressive function, wherein the network parameters include at least a weight matrix and a bias vector; the device includes: An output data module, configured to calculate output data of each layer of the neural network based on preset input data, the network parameters, and the inter-layer progressive function; the preset input data includes position data and talent data; A parameter change rate module, configured to calculate the partial derivative of the output data with respect to the preset input data, and calculate the parameter change rate of each layer of the neural network through the partial derivative and the weight matrix using a parameter change rate function; a correlation component module, configured to calculate the correlation components of the first I layers based on the partial derivatives of the first I layers and the weight matrix; A distribution expectation module is used to calculate the distribution expectation of the network parameters of the I+1 layer based on the output data of the I layer and the parameters of the I+1 layer; An adjustment module is used to calculate an input correlation function value of the I+1th layer through an input correlation function based on the distribution expectation and the correlation component; the input correlation function value and the parameter change rate are used to adjust the optimization network.

10. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to perform the method according to any one of claims 1 to 8.

Images