Abstract
In order to evaluate the competence of candidates in human resource management and select the most suitable talents, the behavior time interview method is used to build a personnel competency model, and various competency indicators are determined. According to the existing mature traditional analysis methods, we calculate the weight of each index and give the competency score. Based on the traditional competency model, the parameters of the training content of BP neural network are obtained. After the training results are tested, a new competency evaluation model based on artificial intelligence is proposed. The results show that the relative error between the model training results and the expected output is very small, the maximum value is −0.12%, and the maximum relative error between the output value obtained by BP neural network and the expected value is 3.8%. Therefore, the personnel competency model based on BP neural network constructed in this paper has accurate calculation results, and its application in the company’s human resource management is feasible and has strong applicability.
1. Introduction
With the continuous development of social economy, in the complex economic activities, human resources, as the first resource, play a core role in the efficiency of resource allocation [1, 2]. At the same time, with the continuous deepening of China’s market economic reform, managers are the key human resources in the organization, affecting the role of other human resources. People are increasingly aware of the importance of talent selection [3–6]. Therefore, using the correct competency model to study the selection of human resources is very important for the development of enterprises [7–9]. It has not only theoretical significance but also practical significance to improve the human resource management level of many enterprises.
The research of competency theory and the practical application of competency model originated in the United States, but in recent years, scholars at home and abroad have made many research achievements on the competency model [6, 10–12]. Miiller-Frommeyer et al. [13] believe that competency is an indispensable internal drive for individuals, which can promote individuals to acquire more professional abilities through learning and practice and form a competency model on this basis. Li et al. [14] mentioned that the success of a team depends on the theoretical knowledge and work attitude of employees under certain conditions. Scott et al. [15] reinterpreted competence in combination with practical research and believed that it is an important indicator to measure the contribution of an individual in the group. For individuals, competence plays a guiding role and contributes to the improvement of human resource management results. Culhane [16] analyzed enterprise managers and affirmed the importance of competency model in combination with enterprise crisis events. Leekitchwatana et al. [17] believe that specific competency factors include employees’ work attitude, employees’ theoretical knowledge reserve, and so on. Combined with the competency model, this model can improve the relationship between individuals and work performance and pave the way for the orderly promotion of various work and the smooth realization of organizational goals. Charles [18] explored the key factors of employee performance based on competency model, aiming to deeply tap the internal potential of employees.
From the above research results, it can be concluded that many foreign scholars studied the concept of competence based on different environments and perspectives [19–23]. However, many mathematicians believe that competence refers to some abilities that a certain type of staff has based on job requirements, job task requirements, industry norms, etc. In the research of competency model, foreign scholars have made great achievements, put forward the iceberg model, and tried to apply it to practical activities [24–27]. In terms of determining the dimensions of competency models, many scholars have created many new competency models based on existing theories. Based on the exploratory and confirmatory results, Jian [28] created a competency model for the Secretary General of the foundation with his own management dimension and employee management dimension as the entry point, including 20 directional evaluation indicators. Dingwei et al. [29] created a competency model for intelligence personnel, subdivided the model factors into 16, and completed the preparation of specific competency measurement tables in combination with his work profile. Patterson [30] and others analyzed the grass-roots directors of high-tech enterprises, started the matching test based on the relevant competency model, and built a feasible competency model for this position. Hazim [31] and others regarded the head of the college as the research object and built a competency model matching it according to their job requirements.
This paper presents a new competency evaluation model based on BP neural network on the basis of analyzing the concept, form, and composition principle of the competency model. The establishment method of BP neural network is emphatically studied. It uses traditional methods to build a multi-dimensional competency model and successfully applies it to the neural network analysis of MATLAB. The calculation results show that the calculation results of the modified model are correct, and the error range is very small. It has high applicability to the follow-up human resource management problems and can be directly used by enterprise managers.
2. The Construction of Competency Model
2.1. Competency Overview
David McClelland, a professor at Harvard University, first proposed the concept of competency characteristics in 1973 [32]. He demonstrated the post-performance from two aspects: individual quality and work ability. Spencer, a scholar, pointed out in the concept of competency for the first time that measurable and determinable properties should be covered at the same time, and it is also the first time to summarize the main content of competency elements [33]. Therefore, experts in human resource management usually divide competency into six levels: knowledge, skills, social roles, self-concept, traits, and motivation. The competency evaluation model refers to the synthesis of competency characteristics that must be possessed by each work. Its essence is to shift the attention from hard indicators such as education and experience diploma to soft indicators and focus on observing the adaptability, adaptability, endurance, and creative thinking ability of candidates to environmental changes in human resource management.
The proposed competency model is the most widely used and classic one with the highest recognition at present. Therefore, this paper is a further study based on Spencer’s competency definition. Meanwhile, in the declining background of artificial intelligence, the application of post competency evaluation model to talent selection will make human resource management function play a new role, and human resource professionals can seek talent characteristics that will bring higher performance, ensuring the scientific and reasonable talent selection system in the application of the competency model.
2.2. Competency Model
The definition of competency, characteristic elements of competency, and grade of behavior indicators can build a model. So far, there are two internationally recognized competency general models: iceberg model and onion model. The details of these three general models are as follows.
Spencer proposed competency iceberg model. As shown in Figure 1, the model divides competency into upper and lower structures. On the top is the part exposed above the iceberg level, which includes two competency characteristic elements: knowledge and skills, both of which belong to explicit competency characteristic elements. The second part is below the hidden level of the iceberg which includes social roles, self-image, personal characteristics, motivation, and so on, which are internal and implicit competency characteristic elements. In short, the dominant elements above the level are obvious, prominent, and easy to measure. However, the hidden attribute below the level often determines the level of job performance, but it is often ignored in the selection of human resource management. The iceberg model contains a good dialectical thought and reflects the relationship between internal and external causes.
In general, the competency characteristic elements above the level are obvious, prominent, and easy to measure. However, the factors that really determine individual job performance are often hidden below the level, and they are easy to be ignored. It is also relatively difficult to measure. The iceberg model contains good dialectical ideas and reflects the relationship between internal and external factors.
Boyatzis proposed the competency onion model [34]. As shown in Figure 2, it is basically similar to the principle of the competency iceberg model. The competency onion model explains the characteristics that each constituent element of competency characteristics can be observed and measured gradually from the inside to the outside. The explicit competency characteristic elements are placed on the outermost layer, and the potential, hidden, and internal competency characteristic elements are placed on the inside, so as to build a layer by layer competency model from the inside to the outside. From the outside to the inside, the difficulty of observing, cultivating, and evaluating competency characteristic elements gradually increases. The outermost competency characteristic elements are relatively easy to be observed and evaluated and also the easiest to be cultivated. However, the more the inner layer is, the more it can reflect the level of future work performance. The motivation and personal characteristics at the core layer are the most reliable and stable competencies.
The International Human Resource Institute (IHRI) proposed a competency ladder model [35]. As shown in Figure 3, the competency ladder model is divided into six levels from top to bottom, including knowledge, skills, social roles, self-concept, personal characteristics, and motivation. Among them, self-concept refers to the individual’s perception and understanding of his own identity. Specifically, the top of the ladder is personal performance behavior, and the six levels at the bottom of the ladder affect the individual’s response to the job objectives to varying degrees, thus determining the personal performance behavior at the top of the ladder.
Facing different industry fields, different enterprises, and different jobs, scholars at home and abroad have conducted in-depth research and developed various competency models, but most competency models are improved based on the above three classic competency models. In other words, although the job competency model varies with the industry, enterprise, and job responsibilities, the basic principle remains the same. Similarly, this paper believes that the competency model is a combination of a series of competency elements that can be measured and developed and can decisively affect job performance under specific job situations, including knowledge, skills, self-concept, personal characteristics, motivation, and so on. Therefore, the competency model based on the management of P company in Laos to be developed in this paper is also based on the improvement of the above three classical models, and the principle is similar to the above three models.
2.3. Construction Principles of Competency Model
In order to let the enterprise human management leaders clearly know which aspects of the candidate are the selection indicators, this paper needs to establish a talent selection system based on the competency model, so as to make an objective, comprehensive, and accurate evaluation of the candidates’ working ability in their applied positions, give the value method of each competency index, and make an in-depth exploration and definition from the qualitative method to the quantitative analysis step.
Therefore, when establishing the talent selection system of the multi-dimensional competency model, in addition to the construction principles of the general index system such as comprehensiveness, operability, accessibility, differentiation, and comparability, the following two special construction principles should be included.
The competency evaluation model established must correspond to the high level of work performance, that is, the talents selected according to the evaluation model must be employees with excellent work performance and should not just stay at the qualified level. This requires the development of student level features that can bring high performance. At the same time, it should be noted that the overall evaluation model must include the most basic competency indicators that meet the needs of each position according to different job analyses.
Enterprise culture is also an important factor that affects the ability evaluation model. Every enterprise has its own cultural charm. The unique cultural charm of the enterprise is the traditional customs that have always been preserved, such as the management style of the enterprise and the behavior style of employees. Then, in the competency evaluation model, the behavior and values represented by the description of self-concept, personal characteristics, and motivation not only are related to each position but also take into account the consistency of the overall cultural atmosphere and value system of the enterprise, so as to ensure that the recruited talents not only match the knowledge and skills with the post needs but also integrate into the corporate culture in the later work.
3. Application and Prospect of Machine Learning Methods
Machine learning method is the core of artificial intelligence, which can analyze data reasonably and correctly. Based on the existing cognitive experience, semi-automatic modeling or automatic modeling after the establishment of automatic learning machine has achieved the goal of reducing human intervention, and the machine operates, analyzes, learns, and solves problems by itself. With the diversification and complexity of data to be processed, machine learning methods still have many problems to be solved when dealing with complex problems, such as new challenges in learning objectives and classification efficiency.
3.1. Application of BP Neural Network
Because the most difficult step of machine learning is to refine the problems in actual production and life into machine learning problems, this requires researchers to deeply understand the actual problem itself. In addition, no matter how accurate the prediction of the machine is, if it cannot meet human needs, the result will be worthless. Therefore, applying BP neural network to the competency evaluation model of human resource management system and using machine learning method to solve the problems existing in talent management and selection is a new research method, which is of great value to human resource management. However, when using BP neural network for analysis, the most important thing is to establish a machine learning problem, that is, the problem in actual human resource management is abstracted into a machine learning problem for subsequent neural network training and parameter optimization. Use machine learning method to calculate human thinking problems, as shown in Figure 4.
3.2. Principle of BP Neural Network and Error Backpropagation
The training process of BP neural network includes data forward propagation and error backpropagation. The main workflow is as follows: the data are first transmitted from the input layer to the hidden layer and then transmitted to the output layer after the relevant data processing algorithm. At this time, if the error between the calculation result and the expected value is less than the specified range, the training will stop if it is qualified. If the error is too large, the error will be transmitted from the output layer to the hidden layer and finally return to the input layer. In this process, the network will correct and adjust the weights of nodes at all levels until the training results are perfect (see Figure 5 for the specific description of the algorithm).
BP neural network algorithm is a “backstepping” learning algorithm of multi-layer network. BP neural network algorithm structure is usually composed of three or more neural network layers connected in turn. These network layers are an input layer, an output layer, and multiple hidden layers (middle layer). Each layer of nerve contains multiple neurons, which are responsible for information processing and information transmission. The neurons in the same layer are independent of each other. The neurons in the adjacent two layers are fully connected by weight, and the neurons in each layer share parameters. Each nerve only receives the information input of the previous layer of nerve and is responsible for converting the information and outputs the converted information to the next layer of neuron, without getting feedback. The location of the hidden layer is between the input layer and the output layer. Usually, there is at least one hidden layer, and the number of hidden layers is not fixed, which is set by the designer. As an intermediate layer, the function of the hidden layer is to extract the feature of the information transmitted by the neurons in the input layer and transfer it to the output layer. The information extraction between each hidden layer is different. Therefore, the hidden layer can also be called the feature extraction layer. In the process of extracting features from the hidden layer, the hidden layer will “self-organize” the weights between the input layer and the hidden layer. In other words, the weights between the input layer, the hidden layer, and the output layer are gradually “automatically” adjusted from the initial random value during the training process of the BP neural network, so that the network can finally acquire the characteristics of the input mode and transmit it to the output layer. Figure 6 shows the multi-layer BP neural network.
(a)
(b)
Input P pairs of learning samples to BP network for training, and the sample input value is XP = (X1, X2,…XP). The network output value of the th learning sample is , k = 1,2,3,…m, and the corresponding expected output value is , k = 1,2,3,…m, and the overall error of this sample iswhere , and is the error of the th training sample. The change of output layer weight iswhere is the learning rate. Define the error signal as
Equation (5) is the partial differential of the activation function of the output layer. Combining the above formula, we can get the adjustment formula of the connection weight of each neuron in the output layer and the hidden layer.
It is easy to see that in BP learning algorithm, the weight adjustment formula of each layer is the same in form, which is determined by three factors, namely, the learning rate , the proportion of error signals output by this layer , and the proportion of input signals Y or X. The error signal of the output layer is related to the difference between the expected output and the actual output of the network, and the hidden error signal of each layer is related to the error signal of the previous layer, which is transmitted layer by layer from the output layer. When the error is small enough, the sample training meets the accuracy requirements.
4. Construction of Personnel Competency Model Based on Machine Learning Method
4.1. Index System
In order to establish a suitable competency model, the model development and evaluation team is composed of human resource managers, human resource management professionals, and university consulting experts. This paper first carries out job analysis on the post, which adopts the method of combining questionnaire survey and key behavior event interview to clarify its job nature, job tasks, job responsibilities, job requirements, and performance standards. On this basis, the preliminary model of personnel competency is constructed. After that, the verification work was supplemented, and finally a competency model was established, including five first-class competency indicators and 13 second-class subcompetency indicators, including personality traits, interpersonal and cooperation, knowledge, technical ability, and learning ability. According to the idea of layering, the reduced model is shown in Figure 7. According to the model established by the survey, this paper uses the BP neural network method to evaluate the personnel competency of a company in Jiangsu, China, in the follow-up work.
4.2. BP Neural Network to Evaluate Personnel Competency Evaluation Model
Take the matrix group determined by the competency index data as the input data of BP neural network, the vector representing the corresponding evaluation results as the output neural network result, and the matrix data of analytic hierarchy process as the training sample, so that the weights and thresholds held by the neural network are the correct internal results of the network after adaptive learning, and the trained BP neural network can be used as an efficient tool. The objects other than the sample mode are evaluated accordingly, and the evaluation of BP network is carried out using MATLAB software.
4.2.1. BP Network Structure Design
The vector model of BP network in the model established in this paper is shown in Figure 8.
is the connection weight vector between the hidden layer neuron and the input vector, and the size is 13 × 14:
The threshold value of the output layer neuron, n2, is the intermediate operation result of the output layer neuron, and the size is 1 × 1.
a 2 is the output result of the output layer, and the size is 1 × 1:
4.2.2. BP Network Learning Algorithm and Selection of Relevant Parameters
For the BP neural network diagram shown in Figure 7, Newton algorithm is used to correct the weight and threshold of each layer. Set K as the number of iterations, and then correct it according to the following formula:
k is the connection weight vector or threshold vector between the layers of iteration; is the gradient vector of the neural network output error to each weight or threshold value of the kth iteration, and the negative sign indicates the opposite direction of the gradient, that is, the fastest descending direction of the gradient; EK is the neural network of the kth iteration. The design here is MSE (mean square error); AK is the current weight and threshold of the error performance function value of Hessian matrix (second derivative).
Although the Newton iterative method has fast convergence speed, it needs to solve Hessian matrix in every operation. In this way, the amount of calculation will be very large. Quasi Newton algorithm introduces a group of matrices to replace Hessian matrix. It does not need to calculate the second derivative, and can be approximated well, which not only keeps the speed of Newton's algorithm, but also avoids the tediousness of Newton's algorithm. This paper adopts BFGS algorithm, whose function is named in MATLAB neural network toolbox.
4.2.3. BP Network Training and Testing
The competency indicators C11, C12, C13, C14, C21, C22, C31, C32, C41, C42, C43, C44, C45, and B5 are used as the input vectors P1, P2,…P14 of the neural network. The competency indicators C11, C12, C13, C14, C21, C22, C31, C32, C41, C42, C43, C44, C45, and B5 are used as the input vectors P1, P2,…P14 of the neural network. First, normalize the value of each competency index, that is, divide all the data by 10 to unify it to the range of (0, 1). The processed data are divided into two parts. The first 18 groups of data are used as network learning samples for training neuron connection weights, and the last 6 groups of data are used for testing. When the training times are 130, the training reaches the required accuracy.
The first 18 groups of data are used as network learning samples for training neuron connection weights, and the last 6 groups of data are used for testing. When the training times are 130, the training reaches the required accuracy.
As can be seen from Table 1, the relative error between the training results and the expected output is very small, and the maximum relative error is −0.12%. After the training, input the six groups of verification data to the trained BP network and get the corresponding competency evaluation results of six personnel, as shown in Table 2. It can be seen that the maximum relative error between the output value (test result) obtained by using BP network and the expected value is 3.8%. In the competency evaluation, this error range is completely acceptable. Store the trained BP neural network in the file; in this way, when encountering similar evaluation problems, as long as the competency index data of the personnel to be evaluated are input, the evaluation results can be obtained immediately by starting the network, and the network output results should be multiplied by 10 to restore to the competency evaluation score.
5. Discussion
This paper is a small part of the multi-dimensional post competency evaluation model in human resource management under the background of artificial intelligence. It applies the existing machine learning methods to a reasonable talent selection system. Consider the comprehensive quality of candidates from a multi-dimensional perspective.
In addition, facing the increasingly competitive market environment, enterprises especially need talents who can bring high performance in their jobs to control the development and growth of enterprises in the market competition. Therefore, the talent selection of enterprises is also particularly important. However, because the traditional talent selection pays too much attention to the educational background and knowledge and skill level of candidates, it cannot meet the needs of matching people with one job and one organization in the new era of enterprises, so the people selected are often only qualified employees of the position rather than having excellent performance. Based on this, this paper explores an objective and effective way of talent selection around the competency model theory, which provides a basis for enterprises to select and employ people scientifically and effectively. The research results can be used not only for the evaluation of job competency of external candidates but also for the promotion and selection of various talents within the enterprise. The research results created in this paper can be directly used by human resource management departments, not only for the evaluation of candidates’ work ability, but also for the promotion and selection of all kinds of talents within the enterprise. The assessment process for the post competence of the company's top management can be roughly divided into three steps: preparation for the assessment, implementation of the assessment, and completion of the assessment. The analysis of the evaluation results of managers’ post competence can achieve the purpose of distinguishing, evaluating, and developing the abilities of senior managers. Through the post competency assessment of senior managers, the gap between the tested and the standard post competency level requirements can be found, and then the weak elements or links of the tested can be found, and then targeted training can be carried out and targeted training plans can be formulated to improve the effect of training development.
6. Conclusion
Based on the analysis of the main principles and characteristics of competency model in current human resource management, this paper selects different level indicators of the model. Aiming at the human resource management goal of competency model, the modeling analysis is carried out based on the BP neural network method. The main conclusions are as follows:(1)The competency model and talent selection system are described in detail, which lay a theoretical foundation for human resource management based on competency model.(2)The model includes five competency indicators and 13 subcompetency indicators, including personality traits, interpersonal and cooperation, knowledge, technical ability, and learning ability.(3)The personnel competency evaluation method based on BP neural network obtains the knowledge and experience of evaluation experts by learning the existing sample mode. When it is necessary to evaluate the personnel competency in the future, as long as the trained BP network is input with the corresponding competency index data matrix, the BP network will reproduce the expert’s knowledge and experience and respond immediately without human intervention, so as to avoid human errors in the evaluation process. At the same time, the network system also has strong fault tolerance.(4)The number of hidden nodes is the focus of BP network design. If there are too many nodes, the machine learning time is longer. On the contrary, the calculation time is short, but the fault tolerance is poor, and the ability to recognize the data that have not been learned and processed is poor. The number of hidden nodes is determined according to the plurality of results. For small-scale training, the quasi-Newton method has the advantage of fast convergence compared with other methods.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there are no conflicts of interest.