Artificial Intelligence-Based Prediction of Individual Differences in Psychological Occupational Therapy Intervention Guided by the Realization of Occupational Values.

As the competition between enterprises intensifies, employees have witnessed a decline of psychological health year by year and severe anxiety and depression. To ensure the normal work of employees, it is important to implement psychological occupational therapy intervention (POTI). The previous studies on corporate occupational values focus on the behavioral variables and attitude variables related to occupations. The research paradigm does not suit a specific group of employees. Therefore, this paper explores the individual differences in POTI guided by the realization of occupational values. For a specific group of employees, the correlations of the psychological problems with job attitude, job involvement, and psychological health level were reasoned logically, a theoretical model was established, and the full-model map was plotted to illustrate the influence of the realization of occupational values over POTI. Then, the evaluation data on the psychological intervention effect of 2,800 employees were explored based on multikernel learning, and an individual difference prediction model was constructed through multikernel learning. Finally, the predicted results of different groups of employees were compared, revealing the effectiveness of our algorithm.

1. Introduction

With the soaring development of social economy, the competition between enterprises is increasingly fierce, and employees face a growing stress, which raises a high concern [1-4]. There are multiple challenges against employees, including the heavy work load and job responsibilities. These challenges bring a high psychological risk to employees. The psychological risk is particularly high among female employees, who play multiple social roles, new employees, who lack social experience, and old employees, who find it hard to keep abreast with the times, adapt to new environment, and learn new things [5-10].

According to the 2019 China's Blue Book on Mental Health, employees have witnessed a decline of psychological health year by year and severe anxiety and depression. To ensure the normal work of employees, it is important to implement psychological occupational therapy intervention (POTI) [11-15]. POTI not only promotes employees' psychological elasticity but also enhances their subjective well-being. Occupational values, the embodiment of personal values in the professional field, make employees satisfied with their work and promote the POTI effect [16-20]. Therefore, it is of great significance to implement POTI among employees guided by the realization of occupational values. The smooth implementation of POTI can ease work pressure and improve the level of psychological health.

To better prepare students for construction jobs, it is a must to consider the complex variables and background factors that affect the job satisfaction of employees. Following Brown's value-based career theory, Bae et al. [21] surveyed 314 construction professionals in the United States and gained insights into their values when they decide to choose and remain in the construction industry. The conceptual model of occupational values was introduced to guide the practitioners and scholars on how to improve the existing construction professionals and prepare future construction professionals. Chang and Lin [22] discussed the role of career development motivation between occupational values and job satisfaction, analyzed 242 valid data by the structural equation model (SEM), and found that occupational values significantly affect career development motivation and job satisfaction. The post-90s are becoming the main force of the workplace in China. Their excellent learning ability, challenging spirit, and innovative mind inject vitality and fresh ideas to enterprises. However, there are some difficulties in the management of post-90s employees, such as quick resignation and high quit rate, because they are not very obedient and highly self-centric. Based on the concept of value realization and the theory of social interaction, Zhu et al. [23] discussed the influence of the realization degree of occupational values over the job involvement of post-90s employees and proposed the value management and incentive strategies for their job involvement.

Project employees, especially those in construction, engineering, and building industries, are suffering from increasing pressure from the uncertainty, complexity, and temporariness of projects. Facing the complex factors of demand and pressure, the variable-centered approach cannot detect the different pressure sources that lead to the same subjective well-being of employees. Wang et al. [24] combined the self-determination theory with the resource conservation theory to examine the structural influence of job pressure sources and psychological demand over the employee's subjective well-being in China's architecture, engineering, and construction (AEC) projects. A total of 256 questionnaires were collected from 27 AEC projects in China and subjected to fuzzy set quantitative comparison. Zahoor et al. [25] framed and tested the influence of technological innovation on employees' psychological health, as well as the degree of influence of employees' learning orientation and organizational support over their psychological health. The results show that both employees' learning orientation and organizational support enhance the inverted U-shaped relationship between technological innovation and employee's psychological subjective well-being. This finding is of theoretical and practical significance to innovation and organizational research.

With the improvement of cognition level, many enterprises start to positively intervene in employees' psychological health, aiming to promote their job involvement and thus drive corporate development. The previous studies on corporate occupational values focus on the behavioral variables and attitude variables related to occupations. But the research paradigm does not adapt to specific groups of people, such as female employees, new employees, and old employees. Therefore, this paper explores the individual differences in POTI guided by the realization of occupational values. Focusing on a specific group of employees, Section 2 deeply discusses the correlations of the psychological problems with job attitude, job involvement, and psychological health level, provides the details about the proposed theoretical model, and illustrates the influence of the realization of occupational values over POTI. Section 3 explores the evaluation data on the psychological intervention effect of enterprise employees through multikernel learning and predicts the individual differences. Finally, the predicted results of different groups of employees were compared, revealing the effectiveness of our algorithm.

2. POTI Guided by Realization of Occupational Values

Drawing on the relevant literature, this paper discusses the POTI action mechanism guided by the realization of occupational values, based on the self-determination theory and social exchange theory, and discusses the regulatory effect on job involvement and psychological health level. Referring to the existing theories and studies, correlations of the psychological problems with job attitude, job involvement, and psychological health level were reasoned logically, and a theoretical model (Figure 1) was established, aiming to explain the different POTI effects among employees and to provide countermeasures that benefit employee development and corporate management.

Figure 1

Conceptual model.

Based on professional selection theory, search theory, and social capital theory, this paper is aimed at exploring the action mechanism of occupational values on employment quality and testing the intermediating effect of job search strategy. To verify the impact of social capital on employment quality, a full conceptual model was established to measure the influence of realizing the employees' occupational values over POTI (Figure 2). The model can handle multiple explanatory variables simultaneously, allow explanatory and independent variables to contain errors, and estimate the structure and relationship of factors at the same time. In this way, it is possible to evaluate the goodness-of-fit of the entire model. The conceptual model investigates the influence of occupational values PV over the POTI effect on employees in seven dimensions, namely, sense of accomplishment, interests and special skills, rights and wealth, capabilities, self-growth, self-independence, and sense of security.

Figure 2

Full model map.

The evaluated POTI effect PL covers ten aspects: intelligence, emotional stability, emotional pleasure, self-discipline and self-control, suitable and coordinated behavioral response, matching between psychological activities and age, harmonious interpersonal relationship, social adaptability, self-cognition, and sound personality. The relationship between the variables of the full model reflects the positive effects of POTI on employees' job involvement and psychological health level, guided by the realization of occupational values.

3. Prediction of Individual Differences in POTI Effect

This paper is aimed at exploring the POTI effect and its individual differences guided by the realization of occupational values. By multikernel learning, the evaluation data on the psychological intervention effect of 2,800 employees were studied. Focusing on the psychological intervention effect in a fixed period, the features of the full-time series were learned with multiple kernels, and a prediction model was established for individual differences. The predicted results were applied to compare different groups of employees.

To prevent the high spatiotemporal complexity from affecting the actual application, this paper chooses to linearly combine the multiple kernel function in multikernel learning model through weighted summation, forming the popular weighted summation kernel. Let Ψ(a, a) be the combined kernel function; e be the weight coefficient of kernel Ψ; N be the number of kernels being combined. Then, we have the following:

The highly efficient simple multikernel learning model was adopted to achieve a high expansibility and universality in the prediction of individual differences in psychological intervention. The original optimization problem can be expressed as follows:

Based on Lagrange multiplier method, the following dual problem can be formed:

The simple multikernel learning model iteratively solves the prediction problem of individual differences in psychological intervention and obtains the weight coefficient of the combined kernel. Firstly, the weight coefficient e of the kernel matrix is solved to convert the original problem into a standard support vector machine (SVM) optimization problem and derive the optimal solution β∗. Then, e is updated by the variance reduced gradient descent. The two steps are executed iteratively until reaching the maximum number of iterations, the termination condition of the algorithm. The decision function can be expressed as follows: where

According to the needs of analyzing the individual differences in psychological intervention effect, this paper modifies the SVM into the support vector regression machine (SVRM), which fits in with the prediction task. The original optimization problem of the regressed simple multikernel learning model can be expressed as follows:

The corresponding dual problem can be expressed as follows:

The algorithm terminates upon reaching the maximum number of iterations. The decision function can be expressed as follows: where

The deep multikernel learning model is formed by superimposing the neural network and the above multiple kernel functions. On each layer, every node is a kernel function. Figure 3 illustrates the deep multikernel learning model. Let Ψ( be the kernel function of the t-th node in the C-th layer; θ( be the connection weight between the kernel function of the w-th node on the k − 1-th layer and that Ψ( of the t-th node on the k-th layer. Then, the set of kernels on each layer can be defined as follows:

Figure 3

Deep multikernel learning model.

The weight of each kernel function in deep multikernel learning is optimized through gradient descent. The kernel functions can be obtained through model training. Then, the individual differences are predicted based on the SVRM.

As a statistical analysis, the canonical correlation analysis intends to find the linear combination of different random variables, such that the correlation between two random variables maximizes after linear combination. To explore the correlation between the realization of occupational values and POTI effect, it is assumed that there are m samples for the evaluation of occupational values and m samples for the evaluation of POTI effect. The feature dimensions of the two sample sets are denoted as t and w, respectively. Then, the two sample sets can be expressed as follows:

After zero-mean processing, the samples can be expressed as follows:

The typical correlation analysis is aimed at finding two projection vectors θ ∈ R and θ ∈ R for the two sample sets A and B that maximize the correlation coefficient between linear combinations h = θa and u = θb. Let D = AA ∈ R × R and D = BB ∈ R × R be the within-set covariance matrices; D = XY ∈ R × R be the between-set covariance matrix satisfying D = D. Then, we have the following:

The corresponding constraint can be expressed as follows:

Solving formula (20) with the Lagrange multiplier, it is possible to obtain the h and u corresponding to θ and θ, under the constraint of formula (19). The transformed eigenvectors are cascaded or summed up by formulas (21) and (22) to realize the feature fusion for building the subsequent prediction model:

In the actual situation, the two sets of evaluation results on the specific group of employees are usually nonlinearly correlated. In this case, the features extracted by canonical correlation analysis are not necessarily valuable. To predict the differences between employees in a specific group, this paper introduces the kernel canonical correlation analysis (Figure 4). Firstly, the two sets of evaluation samples A and B are mapped to the higher-dimensional space:

Figure 4

Kernel canonical correlation analysis.

In the transformed feature spaces G and G, canonical correlation analysis is carried out to find the two projection vectors θ ∈ G and θ ∈ G for the two sample sets A and B that maximize the correlation coefficient between linear combinations v = θζ(A) and u = θζ(A). Following the previous flow, the Lagrange multiplier is introduced to solve the parameters, the features are fused, and a prediction model is constructed to predict individual differences.

4. Experiments and Results Analysis

To verify the effectiveness of the estimation model and estimate the key model parameters, the index data were fitted by the proposed individual-based solvable network model. The unknown parameters of the model were estimated through MATLAB programming. Based on the evaluation data on the psychological intervention effect on 2,800 employees, this paper extracts the features of the full-time series in a fixed period, establishes a prediction model for POTI effect based on the regressed deep multikernel learning model, and evaluates the model performance with the correlation coefficient and mean squared error (MSE) between the predicted and actual differences among the employees in the specific group. As shown in Table 1, the individual differences in the POTI effect on employees gradually decreased as the period extended from 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 7 months, to 8 months. This confirms that POTI has a positive effect guided by the realization of occupational values. Meanwhile, without changing the other experimental conditions, the individual differences in the POTI effect on employees slightly increased with the growing number of female, new, and old employees. Thus, the interpersonal environment of employees negatively affects POTI.

Table 1

Predicted individual differences.

Group size		200	700	1200	1700	2200
Periods	T 1	0.15	0.29	0.36	0.33	0.39
	T 2	0.18	0.32	0.45	0.48	0.35
	T 3	0.28	0.21	0.39	0.35	0.37
	T 4	0.17	0.23	0.34	0.38	0.31
	T 5	0.12	0.27	0.25	0.22	0.34
	T 6	0.11	0.24	0.22	0.27	0.29
	T 7	0.09	0.19	0.20	0.24	0.27
	T 8	0.05	0.18	0.19	0.22	0.24

Figure 5 shows the relationship between the predicted and actual differences among the employees in the specific group. The small correlation coefficient and MSE demonstrate the high prediction accuracy of our model.

Figure 5

Relationship between predicted and actual differences.

To further verify the effectiveness of our model in predicting the individual differences of psychological intervention effect, this paper summarizes the mean prediction errors of five models, namely, nonparametric kernel density estimation, Wasserstein distance, cosine similarity, KL divergence, and our model. To ensure the accuracy of the comparison results, the sample size and the dimensionality of the collected data were kept consistent. The results in Table 2 show that all five models could measure the said differences on the current sample set, but with different mean prediction errors. Our method achieved the smallest mean prediction error. Although our model consumed a slightly longer time than nonparametric kernel density estimation and KL divergence, the results of our method are very satisfactory.

Table 2

Prediction results of different methods.

Method	Mean prediction error	Time cost (s)
Nonparametric kernel density estimation	2.51	3.36
Wasserstein distance	3.26	485.81
Cosine similarity	4.18	132.63
Kullback–Leibler (KL) divergence	2.62	3.48
Our model	0.43	4.62

Then, a test group and a control group were formulated. The scores of the psychological intervention effect on the two groups of employees were subjected to normality test and independent samples t-test. The results show that the scores both obeyed the normal distribution and had no statistical difference (Table 3).

Table 3

Scores of test group and control group.

	Test group	Control group	t	P
PL1	2.05 ± 0.36	2.41 ± 0.39	0.295	0.748
PL2	2.15 ± 0.68	2.59 ± 0.32	0.213	0.819
PL3	2.47 ± 0.51	2.84 ± 0.94	0.051	0.925
PL4	2.84 ± 0.63	2.05 ± 0.58	0.842	0.436
PL5	5.91 ± 0.85	5.16 ± 0.86	-0.041	0.915
PL6	4.25 ± 1.32	3.59 ± 0.72	0.625	0.615
PL7	5.86 ± 0.84	5.28 ± 1.39	1.293	0.247
PL8	3.17 ± 1.46	3.28 ± 1.65	1.485	0.326
PL9	0.84 ± 0.38	1.27 ± 0.36	-0.748	0.528
PL10	0.86 ± 0.45	0.69 ± 0.24	-0.326	0.823

The POTI effects of the two groups of employees were measured in eight periods and examined through analysis of variance (ANOVA) for repeated measurements. Figure 6 shows the trend of the psychological intervention effect. It can be seen that, guided by the realization of occupational values, the POTI strategy can significantly enhance the job involvement and psychological health level of individual employees. In the eight psychological evaluations, the individual job involvement and psychological health level were all above the baseline. This means the POTI has a sustained and marked effect on employees guided by the realization of occupational values.

Figure 6

Trend of psychological intervention effect.

Comparatively, the POTI effect of the test group was on the rise, while that of the control group was failing. This phenomenon is probably because the performance appraisal increases the stress on employees. The employees in the test group can cope with the stress better than those in the control group, thanks to the POTI. This agrees with the result in the previous parts.

Table 4 shows the significance test results on the difference in the POTI effect scores of the test group before the intervention, right after the intervention, and 8 weeks after the intervention. The post hoc test shows that, for the test group, the evaluated occupational values and POTI effect measured right after the intervention did not significantly deviate from those measured 8 weeks after the intervention. Therefore, the POTI has a stable effect on employees under the guidance of the realization of occupational values (Table 5).

Table 4

Results of significance test.

	Preintervention		Postintervention		Postintervention with a lag		F
	M	SD	M	SD	M	SD
Occupational values	9.15	0.62	1.27	7.39	1.26	1.29	8.49∗∗∗
POTI effect	9.38	1.36	7.69	7.15	1.48	1.61	6.27∗∗
Total score	18.24	2.81	16.25	15.29	2.96	2.75	8.37∗∗∗

Note: M and SD are short for mean and standard deviation, respectively.

Table 5

Post hoc test results on group differences.

	Group	Mean difference	P	2 > 5∗∗∗
Occupational values POTI effect	1	2.15	0.02	3 > 7∗∗
	2	2.69	0.06	1 > 4∗∗∗
	3	0.35	0.01	1 > 6∗∗∗∗
Occupational values POTI effect	1	2.07	0.03	2 > 5∗∗
	2	2.95	0.07	1 > 3∗∗
	3	0.62	0.05	4 > 7∗∗∗
Occupational values	1	4.28	0.04	3 > 8∗∗
	2	5.24	0.02	2 > 7∗∗∗
	3	0.69	0.06	6 > 9∗∗

5. Conclusions

Occupational therapy treats or assists people with physiology, psychology, development disorders, or obstacles in social functions, so that they can achieve the highest independence in life. This paper explores the individual differences in POTI under the guidance of the realization of occupational values. Firstly, the correlations of the psychological problems with job attitude, job involvement, and psychological health level were reasoned logically for a specific group of employees, a theoretical model was set up, and the full-model map was prepared for the influence of the realization of occupational values over POTI. After that, multikernel learning was adopted to explore the evaluation data on the psychological intervention effect of 2,800 employees. In addition, an individual difference prediction model was constructed through multikernel learning.

In the experimental section, the different prediction results of a specific group of employees were obtained. It was observed that, with the extension of intervention period, the difference in POTI effect among employees gradually declined. This confirms that that POTI has a positive effect guided by the realization of occupational values. Then, the relationship between the predicted and actual differences among the employees in the specific group was plotted. The small correlation coefficient and MSE demonstrate the high prediction accuracy of our model. Furthermore, the mean prediction errors of five models, namely, nonparametric kernel density estimation, Wasserstein distance, cosine similarity, KL divergence, and our model were summarized. The results show that our model can effectively predict the POTI difference.

After that, a test group and a control group were formulated. The scores of the psychological intervention effect on the two groups of employees were subjected to normality test, and independent samples t-test. The results show that the scores both obeyed the normal distribution and had no statistical difference. It was also observed that, for the test group, the evaluated occupational values and POTI effect measured right after the intervention did not significantly deviate from those measured 8 weeks after the intervention. Therefore, the POTI has a stable effect on employees under the guidance of the realization of occupational values.