Analysis of Teachers’ Cognitive Ability and Teaching Motivation on the Academic Achievement of Students in Higher Education via Employment Data Guidance

Abstract

Teachers and students are the two basic elements in educational activities. Students are educated but are not exactly passive recipients of education. With subjective initiative, all educational impacts must be through the initiative of students to achieve the desired effect. Therefore, all activities of education must start from mobilizing students’ initiative and motivation so that they have sufficient motivation to learn actively and well. The effective analysis of employment data, at the statistical level of data analysis, is a favorable basis to support the influence of teachers on students. However, most of the previous methods are C4.5 algorithms, decision tree generation algorithms based on rough sets, etc., which are commonly used for employment data analysis. None of them can sufficiently deal with the problem of different decision accuracy requirements and noise adaptability. In this paper, we analyze the employment data of a university in 2012 as an example and compare the analysis results with those of the C4.5 algorithm and decision tree generation algorithm based on a rough set. The results show that the decision tree algorithm based on the multiscale rough set model generates a simple decision tree structure. In addition, our methods do not have indistinguishable datasets and are fast in terms of computing. This study provides an effective guide to the relevance of teachers’ cognitive abilities and teaching motivations for students’ employment.

1. Introduction

At present, improving students’ employability has become a problem to be solved in the development of higher education [1]. Higher education is for cultivating skilled and high-quality professionals in production, management, and service, and its philosophy is “service oriented and employment oriented”. Therefore, all educational and teaching activities in higher education institutions are “competence based” [2]. This teaching mode objectively requires teachers to have the basic abilities of educating and teaching, as well as the professional ability of vocational technology, i.e., “double teacher quality” [3]. Over the years, teaching reform and practice have shown that “dual-teacher” teachers have played a positive role in higher education [4]. The purpose of this paper was to explain the mechanism of “dual-teacher” teachers’ behaviors in improving higher education students’ employability through a structural equation modeling analysis to provide theoretical support for the improvement of higher education institutions’ talent cultivation capacity and decisions regarding faculty development [5]. Teachers with “dual-teacher” qualities can teach more effectively and practically according to the skills required by actual jobs, and they can arrange teaching contents scientifically from a practical point of view so that the theoretical requirements of “appropriate quantity and sufficient use” can be implemented, thus enabling students to master the practical skills required by enterprises. Students can grasp professional knowledge that meets the actual needs of enterprises, and they can reasonably use the knowledge learned to solve practical problems, which greatly improves students’ ability to apply professional knowledge [6]. In addition, lectures that are close to actual work will mobilize students’ interest in learning, make them understand the practicality of the knowledge they have learned, and help them perceive the higher quality of the lectures; of course, the perceived higher quality of the lectures will, in turn, attract students to actively participate in classroom learning, forming a virtuous teaching cycle.

At present, the huge amount of data accumulated inside a university’s student management information system, especially historical employment data, does not receive due attention, and the analysis of employment data is mainly aimed at discovering important patterns and knowledge hidden behind the data [7]. In recent years, scholars have proposed the application of data mining for this problem, mainly by applying decision trees, which is a widely used data mining classification method, to actual decision classification problems [8]. In [9], the C4.5 algorithm was used for the generation of a decision classification tree, which is an improved algorithm of H1, the earliest and most influential international ID3 decision tree generation algorithm, developed by Quinlan, but this method cannot well handle the problem of the ambiguity and uncertainty that exists in employment data. The fuzzy decision tree algorithm used in [10] is an expansion and improvement on the traditional decision tree, which extends the application of decision tree learning, making it able to handle the uncertainty in the data. In [11], a decision tree model based on the variable precision rough set algorithm was proposed, thus solving the problem of processing inconsistent information in employment data. Due to the diversity and complexity of actual employment situations, a school’s historical employment data are generally noisier datasets, and the requirements for decision accuracy vary among institutions at all levels; the above methods cannot solve problems well when dealing with the different requirements for decision accuracy and noise adaptability, etc. A decision tree algorithm based on the multiscale rough set model, which draws on the idea of variable precision rough sets and introduces the concept of multiscale into rough set theory, which can solve this problem very well [12,13]. In this paper, the decision tree algorithm based on the multiscale rough set model was proposed to be applied to the analysis of college employment data, and actual employment data were used as an example for the analysis. The results of the analysis were compared to those of the C4.5 algorithm and the decision tree generation algorithm based on the rough set.

In our opinion, the teaching mode of independent learning under the guidance of teachers should be applied flexibly according to the subject, textbook, and students’ characteristics, and it should not be uniform but must reflect the following three points: (1) to stimulate the internal drive of students’ active learning in teaching and to mobilize students’ enthusiasm for learning and initiative so that students’ learning has a lasting internal motivation; (2) to create conditions in teaching; (3) to guide and help students acquire skills to learn gradually.

2. Related Work

2.1. The Positive Role of Teachers for Students

Modern education theory also believes that school education should focus on teaching students to learn, cultivating students’ independent learning ability, and consolidating the foundation of students’ “lifelong learning”, which has become an educational slogan with the characteristics of the times [14,15]. From the aspect of learning, students’ independent learning means that students can be responsible for their own learning in the teaching process, and they can manipulate their own learning with the following characteristics. (1) Subjectivity: in other words, students are in the main position of teaching, they can participate in decision-making regarding teaching from the beginning to the end, they can consciously check and evaluate their own learning, they can implement adjustments in accordance with the possibility of self-development, and they become the master of learning. (2) Independence: in the learning process, students are able to choose their own unique learning methods according to their own learning characteristics and solve learning problems independently and strategically. (3) Forward-looking: students take the initiative to participate in teaching and learning, and they can find new learning modes for themselves to adapt to the learning activities for present and future social development so that they can meet the challenges when new problems arise from learning. Students’ independent learning is complex, which involves both a cognitive–psychological system, including sensory, perceptual, memory, thinking, and intellectual abilities, and an affective system, including motivation, attitude, interest, emotion, will, and personality. Its basic elements involve four main aspects. The psychological quality of independent learning includes attitudes and motivations toward independent learning. This is the inner motivation and premise of independent learning. The basic academic ability of independent learning is the corresponding knowledge and experience reserves that meet the requirements of independent learning. This is the foundation of independent learning. Intellectual qualities of independent learning involve observation, thinking, association, memory, etc. This is both an important part of the composition of an independent learning ability and a potential element for development. The methods and skills of independent learning involve mastering the characteristics and rules of subject learning, choosing the steps and methods of learning independently, having good study habits, and learning strategically, which are elements of the skill of independent learning. The above four elements constitute a four-dimensional structure of students’ independent learning, forming a four-in-one “alloy”. From the aspect of teaching, we emphasize that students are the main body, but we do not deny the role of teachers. We believe that in this model teachers not only impart knowledge to students but also play a leading role, i.e., guide interest (stimulating the engine and arousing interest), guide direction (revealing the goal), guide doubt (stimulating questions and triggering thinking), guide the method (teaching skills), guide feedback (timely correction), etc. In this way, the main role of students and the leading role of teachers can be optimally combined.

2.2. College Employment Data and Management

Based on the background of the era of big data, during this epidemic period, all college students in the country are taking online courses at home, which is a brand new approach. Therefore, what is the effect of these online courses? This involves data mining technology in the field of education, which is a major application of big data in this field. What is educational data mining? Educational data mining refers to the application of data mining technology to process massive amounts of educational data and extract meaningful and valuable information to better help students learn and educators work. To begin to conduct this, we first must find a research angle. Having just started online courses, it is not possible to measure a student’s learning from online courses from the dimension of academic performance. We measured this effect by the learning effect felt by the students themselves. Our perspective was to study the correlation between the direct or indirect factors of online course learning and the learning quality. The employment data warehouses of colleges and universities mainly provide various types of precipitated data used for analysis and decision-making and which can fully concentrate basic information on students, their job search, enterprises, and enterprise recruitment into a unified, standardized, and regulated data environment [16]. The source of employment information data is the basis of the employment data warehouse of colleges and universities, which includes basic personal information, majors, academic achievements, reward achievements during school, and the job search resumes of all students who have recently graduated, accumulated by colleges and universities over the years, as well as various data, such as the profiles of recruiters, job positions, recruitment requirements, and salary packages. Due to the different data collection methods, some of the obtained data are structured and stored in traditional relational databases; some of the data include documents, voice, images, and videos, which are unstructured; and some information sources are artificially generated, such as data generated by universities, job-seeking students, and recruiters on various mobile social media platforms, as well as clickstream data generated from page interactions. Data collection can be achieved by full or incremental collection as a way to build a bridge between the application system and the analysis system [17]. Numerous employment data sources of colleges and universities lead to different types of employment data information. For the different types of data, such as unstructured, semistructured, and structured data, HDFS components can be used for the bottom storage layer architecture, and Hive or Hbase and Solr storage components are used for the upper layer, so that the final data stored are also saved in HDFS [18]. The integrity, timeliness, standardization, accuracy, and authenticity of the data should be fully guaranteed in employment information data storage. Usually, the employment information stored in the database is not a direct copy of the original information, and the employment information that is not suitable for direct analysis is processed through a series of methods, such as data verification, extraction, cleaning, and conversion, to achieve error correction and deduplication, clarification of the logic, and conversion construction, etc., to ensure the legality, accuracy, and uniqueness of the various employment data entering the data warehouse so that it is compatible with the type of data warehouse, fields, etc. [19]. In addition, the employment data warehouses of colleges and universities store real information and data concerning job-seeking students and recruiting units, which contain a large amount of personal information, such as an individual’s name, ID card number, address, telephone number, job requirements, salary allocation, and other sensitive data; if these data and information are leaked, it would greatly impact students personally, as well as recruiting companies. Therefore, certain sensitive employment information must be handled by data desensitization, and data deformation is completed based on a set of rules in a desensitization strategy to reliably protect the sensitive private data of units and individuals [20]. After collecting, cleaning, and desensitizing the employment information data, data matching and integration can be carried out. Integrating the employment information data can solve the problem of data fragmentation and form a multidimensional model with multiple perspectives and levels for analysis. For example, for the same type of job positions, a multidimensional analysis can be conducted from recruitment companies, job locations, working hours, salary levels, work styles, job development, alumni feedback, etc., striving to be able to accurately correspond to student needs, improve the utilization and reusability of employment data resources, and be able to better provide a unified data interface, making the employment data warehouse more computable and scalable, facilitating the operation and maintenance planning of the whole data warehouse, and improving data control and security management capabilities.

2.3. Employment Data Analysis Algorithms

For the integrated and consolidated data in the employment data warehouse, the algorithm module of an artificial neural network can be used for routine statistics and analysis. At the same time, timely support should be provided according to the demands of the application scenario [21]. An artificial neural network (ANN) is created by abstraction on the basis of the human brain structure [22]. The human brain has billions of highly connected neuronal cells forming neural networks, which make human beings intelligent; an artificial neural network is a nonlinear and adaptive information processing system that imitates the neural network of the human brain, maps brain neurons into data processing units, and connects huge data processing units to each other in different ways by establishing a corresponding simple model. It has adaptive, parallelism, robustness, fault tolerance, associative comparison, and inferential deduction capabilities. For the data in the university employment data warehouse, the algorithms of the adaptive resonance theory (ART) network [23], learning vector quantization (LVQ) network [24], and Kohonen network [25] of an artificial neural network were applied to investigate the university’s basic situation; core indexes; professional development; professional indexes; professional resources; basic course resources; course indexes; course teaching quality; course settings; faculty scale; students’ careers, individual development plans, job searches, majors, locations, salaries, positions, recruiters, etc., and a clear and complete student employment data chain was formed on this basis of providing decision support for government authorities and universities to make structural adjustments in enrollment plans.

3. Methodology

3.1. Decision Tree Generation Algorithm Based on the Multiscale Rough Set Model (MRSM)

The multiscale rough set model (MRSM)-based decision tree generation algorithm is based on variable precision rough set theory combining scale variables and scale functions to generate decision trees by using the feature that variables present different decision rules at different scales. To construct a multiscale rough set decision tree, the classification attribute at each node is selected first. If a classification attribute is selected to classify the sample data at the node, it can provide the most certain information for the decision rule. Then, it can be selected as the classification attribute. Due to the existence of the approximate boundary domain, the deterministic information will have some degree of an approximate inclusion problem, i.e., uncertain information may also provide useful decision rules. The MRSM algorithm, which defines the approximate classification accuracy to define the scope of this approximate inclusion, can provide more deterministic information and some uncertain information parts that may play a role in decision analysis; therefore, the one with the greatest approximate classification accuracy is selected. The attribute with the highest approximate classification accuracy is chosen as the extended attribute of the root node. The approximate classification accuracy is calculated as follows:

$$d_{ci}(D)=\frac{{\displaystyle \sum _{i=1}^n\left|\underset{\_}{ap{r}_{ci}^{f(s)}{(Y)}_i}\right|}}{{\displaystyle \sum _{i=1}^n\left|\overline{ap{r}_{ci}^{f(s)}{(Y)}_i}\right|}}$$

(1)

In the process of generating decision trees, the pruning of decision trees during the process of decision tree generation can reduce the steps of pruning after the generation of the decision trees and improve the speed of the decision tree generation by introducing suppression factors. At the same time, the generated decision tree is less complicated and easier for decision-makers to understand.

3.2. Application of the MRSM-Based Decision Tree Generation Algorithm in Employment Data Analysis

3.2.1. Data Collection and Processing

To mine and analyze employment data, the first aim is to have a clear data analysis object. The data in this paper were selected from the data of graduates of a college in 2012, and the employment-related attributes, such as gender, professional grade, foreign language grade, computer grade, skill grade, and employment unit, were extracted, as shown in

Table 1. Student employment data.

No.	$\mathbf{Professional} \mathbf{Grade} \left({\mathit{e}}_1\right)$	$\mathbf{Foreign} \mathbf{Language} \mathbf{Grade} \left({\mathit{e}}_2\right)$	$\mathbf{Computer} \mathbf{Grade} \left({\mathit{e}}_3\right)$	$\mathbf{Skill} \mathbf{Level} \left({\mathit{e}}_4\right)$	Employment Unit (d)
1	Medium (1)	Grade A	Level 1 (1)	Intermediate (2)	Dongguan Hu Cai Printing Co. ($B_2$)
2	Excellent (3)	Grade A	Level 2 (2)	Intermediate (2)	Huawei ($B_1$)
3	Excellent (3)	Grade 4	Level 2 (2)	Advanced (3)	Guangzhou Metro ($A_1$)
4	Good (2)	Grade A	Level 1 (1)	Intermediate (2)	Guangzhou Yinchuang Information Co., Ltd. ($B_2$)
5	Good (2)	Grade A	Level 2 (2)	Advanced (3)	Yangjiang Zhapo Grain Bureau ($A_2$)
6	Good (2)	Grade 4	Level 2 (2)	Advanced (3)	Foshan Haitian Seasoning Co., Ltd. ($A_2$)
7	Good (2)	Grade A	Level 1 (1)	Intermediate (2)	Guangzhou Chuangmeishi Beauty & Cosmetics Co. ($B_2$)
8	Good (2)	Grade A	Level 1 (1)	Intermediate (2)	Guangzhou Jiaxing Trading House ($B_2$)
9	Good (2)	Grade A	Level 1 (1)	Advanced (3)	Ping An Insurance Guangzhou Branch ($A_2$)
10	Medium (1)	Grade A	Level 1 (1)	Intermediate (2)	Chimei Electric ($B_2$)
11	Excellent (3)	Grade 4	Level 2 (2)	Advanced (3)	Guangzhou Metro ($A_1$)
12	Medium (1)	Grade 4	Level 2 (2)	Intermediate (2)	Casio ($C_2$)
13	Excellent (3)	Grade 6	Level 2 (2)	Advanced (3)	Samsung Electronics ($C_1$)
14	Excellent (3)	Grade A	Level 1 (1)	Advanced (3)	South China Electric Research Institute ($A_1$)
15	Medium (1)	Grade A	Level 2 (2)	Intermediate (2)	Chimei Electric ($B_2$)
16	Excellent (3)	Grade A	Level 1 (1)	Advanced (3)	Foshan Haitian Seasoning Co., Ltd. ($A_2$)
17	Excellent (3)	Grade A	Level 2 (2)	Intermediate (2)	Zhuhai Gree Electric ($B_1$)
18	Good (2)	Grade A	Level 1 (1)	Intermediate (2)	Chimei Electric ($B_2$)
19	Good (2)	Grade A	Level 2 (2)	Advanced (3)	Baosteel Zhanjiang Steel Branch ($A_2$)
20	Good (2)	Grade 4	Level 2 (2)	Beginner (1)	Deep Point Advertising ($B_2$)

, where the conditional attribute professional grade, $e_1$, was granted according to the weighted average score of the students’ professional grades and divided into three categories: medium (weighted average score < 70), good (70 ≤ weighted average score/85), and the values were taken separately (1 means medium; 2 means good; 3 means excellent); the foreign language level, $e_2$, was granted according to the level of the English certificate taken by the students (1 means grade A; 2 means grade 4; and 3 means grade 6); the computer level, $e_3$, was granted according to the level of the computer certificate taken by the students (1 means grade 1; 2 means grade 2); and the skill level, $e_4$, was granted according to the level of the skill certificate taken by the students (1 denotes beginner; 2 denotes intermediate; and 3 denotes advanced). For the decision attribute employment unit, d, first, according to the nature of the unit of employment of the students, the employment unit was divided into three categories: institutions (A), private enterprises (B), and foreign-funded enterprises (C). Institutions broadly included government units, state-owned enterprises, universities and colleges, etc.; private enterprises broadly included enterprises operated by private individuals or organizations; and foreign-funded enterprises broadly included foreign-owned or Sino–foreign joint ventures. In addition, an echelon was divided according to the treatment benefits provided by each type of enterprise, geographical location, etc. The quantitative values were taken as good institutions ($A_1$), general institutions ($A_2$), good private enterprises ($B_1$), general private enterprises ($B_2$), good foreign enterprises ($C_1$), and general foreign enterprises ($C_2$). The results of the data quantification are shown in Table 1.

3.2.2. Constructing the Decision Tree

According to the MRSM-based decision tree generation algorithm, provided above, we first set the scale function, $f(s)$ = 0.6, and the threshold, $f(s)$ = 0.8. The process of constructing a decision tree using this algorithm is as follows: (1) Calculate the approximate accuracy of each conditional attribute relative to the decision attribute with respect to the scale function, f(s), according to Equation (1) and obtain $d_{ci}(D)$ = 0.74, $d_{ci}(D)$ = 0.15, $d_{ci}(D)$ = 0.32, $d_{ci}(D)$ = 0.15, and $d_{ci}(D)$ = 0.32. (2) According to Step 2 of the algorithm, the attribute $e$ (i.e., professional achievement) is marked as the root node $e$. (3) As $d_{ci}(D)$ = 0. 74 > 0.6, proceed to Step 4 of the algorithm. (4) From the attribute e, there are 3 possible values (1, 2, and 3), and it is known that the formed tree has 3 different branches, where, in the case of $e_1$ = 1, the value of the suppression factor is obtained as 1 > A. Therefore, the attribute e is identified as a leaf. In the 2 cases of $e_2$ = 2 and $e_3$ = 3, the value of the suppression factor obtained does not satisfy $d_{ci}(D)$ ≥ A; then, the current subset is taken; return to Step 2 for the calculation. (5) Again, the approximate accuracy value is calculated according to Equation (1), and it is concluded that $d_{ci}(D)$ = 1 has the largest value; therefore, the node with the selected attribute (i.e., skill level) is the tree. Continue until a decision tree with a complexity of 8, a depth of 3, and 5 leaves is finally obtained (Figure 1).

The MRSM decision tree generation algorithm can obtain decision trees with different angles and scales based on different scale functions, $f(s)$; therefore, we took $f(s)$ = 0.8 again for the computational analysis. After completing the above operation process, a decision tree with a complexity of 10, depth of 3, and 6 leaves was obtained (Figure 2).

3.2.3. Discussion of the Scales

Analyzing the experimental results of this paper, as the scale variable increased, $f(s)$ became larger, the expression of knowledge became more detailed, and the number of decision rules gradually became larger. However, the complexity of the generated decision tree structure also increased. This is because in the MRSM decision tree generation algorithm, as the scale variables increases, the range of the approximation boundaries corresponding to the decision attributes gradually becomes narrower, and the coverage of the decision rules increases. However, it should be noted that in the case of more noise, the increased coverage of the decision rules sometimes results in rules that are uncertain. Therefore, the requirements of the different users for decision accuracy should be fully considered, and the parameters of the scale function, $f(s)$, should be selected reasonably according to the dataset faced in the decision analysis and the user’s degree of accuracy for the research problem.

3.2.4. Rule Knowledge Description

According to the MSRM-based decision tree generation algorithm, a decision rule can be obtained from the root node to the leaf nodes. Combined with the results of the analysis conducted on the employment training set of this paper, the decision rules can be derived from Figure 1 when $f(s)$ = 0.6. Rule 1: if the professional grade = “medium”, then the employment, in general, is private enterprises. Rules 2: if the professional grade = “good” and the skill level = “intermediate”, then the employment, in general, is based on private enterprise units. Rules 3: if the professional grade = “good” and the skill level = “advanced”, then the employment, in general, is institutions. Rules 4: if the professional grade = “excellent” and the skill level = “medium”, then the employment is in good private enterprises. Rules 5: if the professional grade = “excellent” and the skill level = “advanced”, then the employment is in a good career. Figure 2 also shows the decision rules when the decision function is $f(s)$ = 0.8. The specific rules are not listed in detail here. From the analysis of the rules derived from the above two different values of the decision function, to improve the employment quality of graduates, we should increase the training of students’ professionalism in the personnel training program, and the setting of professional courses must be close to the actual work. In order to improve the employment quality of graduates, we should increase the cultivation of students’ professionalism in the talent training program, and the curriculum of professional courses must be close to the work reality.

4. Experiments

4.1. Implementation Details

To verify the effectiveness of the MRSM-based decision tree generation algorithm for employment data mining, the following experimental environment was used: hardware: Intel(R) Core(TM) 2 Duo CPU 2.93 GHz, 2 GB memory; software: Windows XP(SP3) and Matlab6.5. The experimental training set was the employment data in Table 1, and this algorithm was compared with the C4.5 and the decision tree generation algorithms based on the rough set, which was based on the decision tree generation algorithm. The results are shown in

Table 2. Comparison of the decision trees generated by the different algorithms.

Algorithm	Scale	Depth	Number of Rules	Nonseparable Dataset
$MRSM(f(s)=0.6)$	8	2	6	0
$MRSM(f(s)=0.8)$	10	2	5	0
C4.5	22	6	9	3
RS	16	5	9	2

. It can be seen that the decision tree generation algorithm based on the multiscale rough set model was used to analyze the employment data, and the scale and depth of the tree structure were not large. Thus, the number of rules generated was relatively concise, but there was no indistinguishable dataset. While the other two algorithms analyzed the employment data, the resulting decision trees were relatively complex, the number of rules generated was larger, and there existed an unpredictable dataset.

4.2. Performance Evaluation of the MRSM-Based Decision Tree Generation Algorithm

To evaluate the performance of each decision tree generation algorithm, the complexity of the decision tree and the classification accuracy are two more important factors. Complexity refers to the simplicity and complexity of the rule description of the problem according to the classification discovery model; the simpler the rule description, the easier it is to understand, such as the size and depth of the decision tree and the calculation time consumption index. Classification accuracy refers to the ability to accurately predict new or unknown data classes according to the resulting classification model; a high accuracy means that more accurate classification data can be obtained when dealing with huge amounts of data [26].

According to the two phases of the decision tree application, the learning phase and testing phase, 1000 employment data were selected as the test set to test the above decision tree model built using the training set, i.e., the generated decision model was used to classify the data in the input test set. The experiments were conducted at $f(s)$ = 0.6 and $f(s)$ = 0.8 for the different scale functions, and the algorithm was compared with the C4.5 and RS algorithms in terms of the classification accuracy and running time; the results are shown in

Table 3. Comparison of the decision tree performances of the different algorithms.

Algorithm	Classification Accuracy	Running Time (s)
$MRSM(f(s)=0.6)$	0.80	100.1
$MRSM(f(s)=0.8)$	0.82	107.4
C4.5	0.72	120.2
RS	0.84	108.5

. The results show that the multiscale rough set model-based decision tree generation algorithm outperformed the C4.5 algorithm in terms of the classification accuracy and running speed, although its classification accuracy was lower than or equal to that of the RS algorithm for the different values of the scale function, but it outperformed the other algorithm in terms of the running speed. It should be noted that the experimental sample set of data had an impact on the decision tree’s performance, and the value of the scale function in the decision tree generation algorithm based on the multiscale rough set model also had an important impact on the decision tree’s performance in conjunction with the above analysis. Therefore, the decision analysis should focus on the selection of parameters according to the dataset and the user’s accuracy of the research problem.

4.3. Extended Study—Experiment on the Graduation Destination Distribution

From the information on the graduates, the distribution of 14 graduation destinations (including employment in the form of signing an employment agreement, research assistant, pending employment, self-employment, and further education) of students from 120 institutions of higher education was calculated. Through the analysis, most of the institutions’ graduates were mainly employed in the form of signing labor contracts, signing employment agreements, and other employment forms. Through the algorithm proposed in this paper, the overall clustering results of the graduates’ destinations in each higher education institution were obtained, as shown in Figure 3.

In Figure 3, the left part of the generated radiation tree has the worst fitting distribution of the graduates’ graduation direction (i.e., the most scattered distribution of the graduation direction of this type for the colleges and universities); the bottom “bulb” part in Figure 3 indicates the group of colleges and universities, and the graduation direction distribution of this part of the colleges and universities is disordered; thus, it is difficult to distinguish the superiority and inferiority; however, there are two types of colleges and universities with a similar distribution of graduation destinations. In Figure 4, the top-right branch of colleges and universities has a better distribution of employment and graduation destination; the top-right part of the colleges and universities has the best distribution of employment provinces (i.e., the most concentrated distribution of employment types in this type of colleges and universities). From the information on the graduates, the distribution of the employment geographical types of graduates from 120 colleges and universities was counted. Through the analysis, the most employed geographic areas of the graduates from most institutions were provincial capitals, followed by prefecture-level cities. The overall clustering results of the employment geographic types of the graduates from each higher education institution are shown in Figure 4, indicating that these colleges and universities have similar employment geographic types, and the employment geographic distribution of these colleges and universities is poorly fitted (i.e., the employment types of these colleges and universities are scattered). The “bulb”-shaped part on the top left indicates the group of colleges and universities, and the distribution of the employment geographic types of these colleges and universities is disorganized; thus, it is difficult to distinguish the advantages and disadvantages. The geographical distribution of the employment of the graduates from higher education institutions in the top-right branch had the worst fit. In Figure 4, the geographical distribution of the employment in the lower part of the “light bulb” shape is better. In Figure 4, the geographical distribution of the employment in the bottom-right part of the colleges and universities had the best fit, which means that the distribution of the employment types in this type of colleges and universities is the most concentrated.

5. Conclusions

We propose the application of the decision tree algorithm based on a multiscale rough set model to mine and analyze the employment data of colleges and universities to uncover useful patterns and knowledge behind the massive amount of employment data. The multiscale rough set model-based decision tree algorithm introduces scale variables and scale functions to allow for the generated decision trees to meet different users’ needs for decision accuracy. The decision tree algorithm based on the multiscale rough set model was applied to the analysis of college employment data, and the rules mined could meet the precision requirements of different decision-makers in the schools, which can effectively help management at all levels of the school to provide more accurate and scientific decisions on various employment work and talent training programs of the school. The solution can be fed back to the academic affairs management system as one of the performance evaluation indicators for teachers.