Research on Employment and Entrepreneurship Potential Mining and Cultivation Mechanism of College Students Based on Decision Tree Modeling

1)

Data collection. This study mainly collects data from three aspects: basic information of college graduates, graduate performance information, and graduate employment information, and organizes them. Establish the data information form.

Correlation analysis of attributes affecting college graduates’ successful employment. In order to effectively construct the decision tree prediction model of whether graduates are successfully employed, it is necessary to analyze the correlation of the attributes affecting the employment of college graduates, obtain the test attributes, and ensure the accuracy of the decision tree prediction model. SPSS software is mainly used to conduct correlation analysis.

Table 2 shows the correlation analysis of the attributes affecting the employment of college graduates, and the four attributes of graduates’ comprehensive achievement, whether they are student cadres, birth source information and political identity are the test attributes affecting the successful employment of college graduates.

The C4.5 [24] algorithm is an improvement of the ID3 algorithm. Unlike the ID3 algorithm [25], the C4.5 algorithm selects attributes for each node of the tree based on the information gain rate. The algorithm selects the attribute with the highest gain rate as the test attribute for the current node. This attribute minimizes the amount of information needed to classify the samples in the resulting partition and reflects the minimal randomness or “impurity” of the partition. This theoretical approach minimizes the expected number of tests required to classify an object and ensures that a simple tree is found. For the sake of convenience, the concepts are explained below.

Definition 1: Let dataset s be a set containing s data samples and the category attribute can take m different values corresponding to m different categories C_i(i = 1, 2, ⋯, m). Assume s_i is the number of samples in category C_i; the amount of information required to categorize a given data object is called the entropy before s division, i.e.: 1I(s1,s2,⋯,sm)=−∑i=1mpilog2(pi)$$I({s_1},{s_2}, \cdots ,{s_m}) = - \sum\limits_{i = 1}^m {{p_i}} {\log_2}({p_i})$$

where p_i is the probability that any data object belongs to category C_i: p_i = s_i/s.

The entropy reflects the average uncertainty and purity of the sample set s. In general, the higher the entropy value, the higher the average uncertainty and the lower the purity.

Definition 2: Let an attribute A take v different discrete attribute values {a₁, a₂, ⋯a_v}. using attribute A the set S can be partitioned into v subsets {S₁, S₂, ⋯S_v}, which S_j contain data samples from the S set where the attribute A takes a_j values. If attribute A is selected as a test attribute, i.e., attribute A is used to partition the current sample set. Let s_ij be the number of samples belonging to C_i category in subset S_j. Then the information required to partition the current set of samples using the attribute A partitioned s the entropy after partitioning, i.e: 2E(A)=−∑j=1vs1j+⋯+smjsI(s1j,⋯,smj)$$E(A) = - \sum\limits_{j = 1}^v {\frac{{{s_{1j}} + \cdots + {s_{mj}}}}{s}} I({s_{1j}}, \cdots ,{s_{mj}})$$

where ∑i=1vs1j+⋯+smjs$$\sum\limits_{i = 1}^v {\frac{{{s_{1j}} + \cdots + {s_{mj}}}}{s}}$$ denotes the weight of subset S_j in dataset s.

Definition 3 Information gain is defined as: 3Gainl(Ar)=Il(s1,s2,⋯,smr)−El(Ar)$$Gainl(Ar) = Il({s_1},{s_2}, \cdots ,{s_m}r) - El(Ar)$$

For sample set S, the larger Gain(A) is, the higher the purity of subset partitioning.

Definition 4 Information gain rate is defined as: 4GainRatiol(Ar)=Gainl(Ar)SplitInfol(Ar)$$GainRatiol(Ar) = \frac{{Gainl(Ar)}}{{SplitInfol(Ar)}}$$

where SplitInfo(A)=−∑j=1vs1j+⋯+smjslog2s1j+⋯+smjs$$SplitInfo(A) = - \sum\limits_{j = 1}^v {\frac{{{s_{1j}} + \cdots + {s_{mj}}}}{s}{{\log }_2}\frac{{{s_{1j}} + \cdots + {s_{mj}}}}{s}}$$.

Calculate the information gain rate for each attribute by using the above formula. The attribute with the highest information gain rate is selected as the test attribute for the given set s, a node is created and labeled as that attribute, and branches are created for each value of the attribute to divide the sample.

1)

Algorithm description

Assuming T is the training set, when constructing the decision tree for T, the attribute with the largest information gain rate is selected as the split node, and T is divided into n subsets according to this criterion. If the i th subset T_i contains tuples of the same category, the node becomes a leaf node of the decision tree and stops splitting. And for the other subsets of T that do not satisfy this criterion, recursively generate the tree as described above until all subsets contain tuples belonging to one category.

Figure 1.

Flowchart of the C4.5 algorithm

Pruning a decision tree is to cut out the replaceable subtrees and replace them with leaves to simplify the decision tree. The algorithm is also used to reduce the prediction error to improve the quality of the classification model. If the expected misclassification rate of the subtree is greater than the error rate predicted by a single leaf, replacement needs to be performed.

The C4.5 algorithm utilizes a post pruning approach, which uses pessimistic pruning when evaluating the prediction error. The method uses the training sample set itself to estimate the error before and after pruning to decide whether to actually prune or not. The formulas used in the method are as follows: 5Pr[f−qq(1−q)/N>z]=c$$Pr\left[ {\frac{{f - q}}{{\sqrt {q(1 - q)/N} }} > z} \right] = c$$

Where N is the number of instances, f = E/N is the observed error rate (where E is the number of classification errors in N instances), q is the true error rate, c is the confidence level (an input parameter of the C4.5 algorithm, with a default value of 0.25), and z is the standard deviation corresponding to c, the value of which can be obtained by checking the table of normal distribution according to the set value of c. The formula calculates an upper confidence limit for the true error rate q, which is used to make a pessimistic estimate for the error rate e of the node: 6e=f+z22N+ZfN−f2N+z24N21+z2N$$e = \frac{{f + \frac{{{z^2}}}{{2N}} + Z\sqrt {\frac{f}{N} - \frac{{{f^2}}}{N} + \frac{{{z^2}}}{{4{N^2}}}} }}{{1 + \frac{{{z^2}}}{N}}}$$

Determine the size of e before and after the pruning, and if e becomes smaller, prune.

Factor analysis [26] is a technique for data simplification, which mainly reflects the idea of data dimensionality reduction. By studying the internal interdependence between variables, a few abstract variables are identified that can synthesize the main information of all variables, which cannot be measured directly, and the abstract variables are usually called factors.

Factor analysis is similar to cluster analysis in that both types are classified as R and Q. R-type factor analysis analyzes variables and Q-type factor analysis [27] analyzes samples. In this paper, R-type factor analysis is chosen to analyze the variables, and the characteristics of R-type factor analysis are that the common factors in R-type cannot be observed intuitively, but they are common factors that exist objectively. Given a sample of n, p indicators, X = (X₁, X₂, …, X_p)^T is a random vector, and the common factor to be sought is F = (F₁, F₂, …F_m)^T, there are: 7Xi=ai1F1+ai2F2+⋯aimFm+εi,(i=1,2,...p)$${X_i} = {a_{{\text{i}}1}}{F_1} + {a_{i2}}{F_2} + \cdots {a_{im}}{F_m} + {\varepsilon_i},(i = 1,2,...p)$$

In the above equation F₁, F₂, …F_m is referred to as the common factor, ε_i is referred to as the special factor of X_i, which represents the variance of the variable due to influences other than the common factor, X_i is the measurable variable, and the following equation shows the matrix form of the model: 8X=AF+ε$$X = AF + \varepsilon$$

Among them: 9A=[ a11 a12 ... a1m a21 a22 ... a2m ... ... ... ... ap1 ap2 ... apm]=(A1,A2,...Am)$$A = \left[ {\begin{array}{*{20}{c}} {{a_{11}}}&{ {a_{12}}}&{ ...}&{ {a_{1m}}} \\ {{a_{21}}}&{ {a_{22}}}&{ ...}&{ {a_{2m}}} \\ {...}&{ ...}&{ ...}&{ ...} \\ {{a_{p1}}}&{ {a_{p2}}}&{ ...}&{ {a_{pm}}} \end{array}} \right] = ({A_1},{A_2},...{A_m})$$

is called the factor loading matrix, and a_ij is the factor loading whose actual meaning is the correlation coefficient between F_i and X_j. 10X=[ X1 X2 ⋮ Xp],F=[ F1 F2 ⋮ Fp],ε=[ ε1 ε2 ⋮ εp]$$X = \left[ {\begin{array}{*{20}{c}} {{X_1}} \\ {{X_2}} \\ \vdots \\ {{X_p}} \end{array}} \right],F = \left[ {\begin{array}{*{20}{c}} {{F_1}} \\ {{F_2}} \\ \vdots \\ {{F_p}} \end{array}} \right],\varepsilon \: = \left[ {\begin{array}{*{20}{c}} {{\varepsilon_1}} \\ {{\varepsilon_2}} \\ \vdots \\ {{\varepsilon_p}} \end{array}} \right]$$

This mathematical model needs to satisfy the following four aspects: 1)

m < p, i.e. the number of extracted common factors is less than the number of original variables;

The decomposition of the covariance matrix of the original variable X is given by X = AF + ε, yielding Cov(X) = ACov(F)A^T + Cov(ε), i.e.: 11Cov(X)=AAT+diag(σ12,σ22,...,σm2)$$Cov(X) = A{A^T} + diag(\sigma_1^2,\sigma_2^2,...,\sigma_m^2)$$

The smaller the value of σ12,σ22,...,σm2$$\sigma_1^2,\sigma_2^2,...,\sigma_m^2$$, the more components the common factor shares.

The loading matrix is not unique, let T be a m × m matrix, and let A^* = AT, F^* = T′F, then the model can be expressed as follows X = A^*F^* + ε. 1)

Statistical significance of factor loadings a_ij.

For the factor model: 12Xi=ai1F1+ai2F2+⋯aimFm+εi,(i=1,2,...p)$${X_i} = {a_{{\text{i}}1}}{F_1} + {a_{i2}}{F_2} + \cdots {a_{im}}{F_m} + {\varepsilon_i},(i = 1,2,...p)$$

The covariance between X_i and F_j can be obtained as: 13Cov(Xi,Fj) = Cov(∑k=1maikFk+ε,Fj)& = Cov(∑k=1maikFk,Fj)+Cov(εi,Fj) = aij$$\begin{array}{rcl} Cov({X_i},{F_j}) &=& Cov\left( {\sum\limits_{k = 1}^m {{a_{ik}}} {F_k} + \varepsilon ,{F_j}} \right)\& \\ &=& Cov\left( {\sum\limits_{k = 1}^m {{a_{ik}}} {F_k},{F_j}} \right) + Cov({\varepsilon_i},{F_j}) \\ &=& {a_{ij}} \\ \end{array}$$

If X_i is normalized, X_i has a standard deviation of 1, and F_j has a standard deviation of 1, there: 14γXi,Fj=cov(Xi,Fj)D(Xi)D(Fj)=Cov(Xi,Fj)=aij$${\gamma_{{X_{i,{F_j}}}}} = \frac{{cov({X_i},{F_j})}}{{\sqrt {D({X_i})\sqrt {D({F_j})} } }} = {\text{Cov}}({X_i},{F_j}) = {a_{ij}}$$

Then, for the standardized X_i, a_ij is the correlation coefficient between X_i and F_j, indicating that X_i depends on the weight, or power, of F_j. Psychologists call it the loadings, indicating the loadings of the i th variable on the j th common factor, reflecting the relative importance of the i th variable on the j th common factor. 2)

Statistical significance of variable commonality.

There are factor models known: 15D(Xi) = ai12D(F1)+ai22D(F2)+⋯+aim2D(Fm)+D(εi) = ai12+ai22+⋯+aim2+D(εi) = hi2+σi2$$\begin{array}{rcl} D({X_i}) &=& a_{i1}^2D({F_1}) + a_{i2}^2D({F_2}) + \cdots + a_{im}^2D({F_m}) + D({\varepsilon_i}) \\ &=& a_{i1}^2 + a_{i2}^2 + \cdots + a_{im}^2 + D({\varepsilon_i}) \\ &=& h_i^2 + \sigma_i^2 \\ \end{array}$$

The common degree of quantity X_i is: 16hi2=∑j=1maij2i=1,2,...,p$$h_i^2 = \sum\limits_{j = 1}^m {a_{ij}^2} \quad i = 1,2,...,p$$

If X_i is normalized, there: 171=hi2+σi2$$1 = h_i^2 + \sigma_i^2$$

3)

Statistical significance of the variance contribution gj2$$g_j^2$$ of the common factor F_i Let the factor loading matrix be A, and call the sum of squares of the elements of column j, i.e: 18gj2=∑i=1paij2j=1,2,...,m$$g_j^2 = \sum\limits_{i = 1}^p {a_{ij}^2} \quad j = 1,2,...,m$$

is the contribution of public factor X_i to F_j, which is gj2$$g_j^2$$, indicating the extent to which the information of variable X_i can be described by the extracted k public factors, with a value interval of (0, 1), and the larger the value of gj2$$g_j^2$$, the higher the ratio of information that can be interpreted by the public factors for that variable. The relative importance of the common factor F_j can be measured by gi2/h$$g_i^2/h$$, which is called the contribution of the common factor F_i to X. The purpose of factor analysis is to derive the solution of the factor analysis model from the covariance array Σ or correlation array R of the original random vectors, i.e., to derive the loadings array A and the characteristic variance array D_ε, and to make the relevant explanatory statements.

First, the original data with high reliability and authenticity are selected, and such data are usually obtained based on the research of actual problems.

Second, standardize all the original variables to eliminate the influence of variables in the order of magnitude, and then obtain the correlation matrix based on the standardized data and convert it into the correlation between variables.

Third, the principal component analysis [28] is used to solve the common factor and the factor loading matrix is derived.

Fourth, in order to make the coefficients in the factor loading matrix more significant, the factor loading matrix can be rotated, and in this paper, the maximum variance orthogonal rotation method is utilized to maximize the relative sum of squares of the loadings, and the factors are interpreted by naming.

Fifth, calculate the component matrix scores of the factors.

Sixth, analyze the results and draw conclusions

The logic diagram of factor analysis is shown in Figure 5:

Figure 5.

Logic diagram of factor analysis

In order to initially choose the measurement indicators to cover as many factors affecting the employment situation of graduates as possible, this paper establishes a relatively reasonable indicator system for the description of graduates.

Starting from the needs of enterprises, the factors affecting the employment quality of college students are shown in Figure 6. On the basis of a comprehensive understanding of the employment situation of fresh graduates of the information class, 15 measurement indicators are proposed following the five principles of indicator selection mentioned above.

Figure 6.

Factors affecting the employment quality of college students

A sample survey was conducted on the graduates of the last three years from three universities, namely, University X, University Y and University Z, in the area of City B. A total of 620 questionnaires were distributed and 600 questionnaires were recovered, among which 600 questionnaires were valid, and the validity rate of the questionnaires was 100%.

This paper divides the salary grade according to the proportion difference between the individual sample salary and the average salary of the sample set as a criterion, and the division is shown in Table 6, in which St indicates the salary of a single sample and S_d indicates the average salary of all sample data.

In this paper, 15 observable indicators are used as influencing factors for determining the salary level of college students’ employment, and these indicators are expressed in the form of variables: where X1 indicates GPA scores, X2 indicates English scores, X3 indicates scholarships won, X4 indicates course design scores, X5 indicates project experience, X6 indicates award-winning experience, X7 indicates award level, X8 indicates student cadre experience, X9 indicates cadre level, X10 denotes campus honors awarded, X11 denotes social situation, X12 denotes political appearance, X13 denotes part-time job experience, X14 denotes internship experience, and X15 denotes interview experience.

In order to make the sample model more objective in describing the characteristics of the sample, the data of all the observed variables are normalized to eliminate the gap of the data outline, and all the data are transformed to between [0, 1] for data transformation process. 19xi=xi−xminxmax−xmin$${x_i} = \frac{{{x_i} - {x_{\min }}}}{{{x_{\max }} - {x_{\min }}}}$$

The improvement of college students’ employment and entrepreneurship ability requires the continuous improvement of their comprehensive quality. In the process of their own ability to cultivate, actively absorb professional knowledge, enthusiastically participate in social practice, to adapt to the enterprises and institutions on the demand for talent, and constantly improve their own quality level, and ultimately to achieve the purpose of employment choice and entrepreneurship.

Career planning, also called career design, combines the individual and the organization, comprehensively considers and summarizes the subjective and objective conditions of the assessor’s career, weighs the assessor’s personal interests, abilities, external evaluations and social relations, combines with the needs of the DU and the assessor’s personal career inclination, determines the most reasonable career goal, and makes scientific and reasonable arrangements for the realization of this goal. Cultivating employment and entrepreneurship is a long and systematic project, and the prerequisite and foundation is scientific and reasonable career planning. The phenomenon of lack of career planning is especially prominent among college students in independent colleges, which largely affects the quality and level of employment and entrepreneurship of college students in independent colleges.

Only by highlighting the marketization and internationalization of the school, strengthening the construction of characteristics, and running the school independently according to the law for the society, can the school adapt to the trend of education development, survive and develop in the increasingly fierce competition, and it is also the fundamental way out and the inevitable choice to cope with the challenge of global economic integration. With the deepening of the new industrial revolution, the development of knowledge-based economy and economic globalization, the trend of internationalization of higher education has come, as a close link with the market economy of the school can not ignore the development of education to send a trend. In the face of intensifying competition, schools can only adapt to the new trend with distinctive schooling characteristics. Can only rise to the challenge, actively participate in the market, international competition, hard work, accurate positioning of their own, school mode should have characteristics, and actively and market supply and demand, in order to turn the crisis into an opportunity for students to create good conditions for employment and entrepreneurship, in order to win for the school’s survival and opportunities for development and growth.

Public service function is one of the important tasks of the modern government, as the provider and manager of this function, whether it is guiding the employment and entrepreneurship activities on the micro level or regulating the employment and entrepreneurship activities on the macro level, it plays a key leading role in solving the employment and entrepreneurship problems of college students.

The government should take the initiative, strengthen the sense of service, and continuously introduce policies and measures to promote the employment and entrepreneurship of college students according to the reality. First, to create a high-quality market economic environment, and work together to create a capital market for college students’ entrepreneurship. Financing difficulties is one of the biggest difficulties faced by college students in the process of entrepreneurship, to solve the financing difficulties of college students’ entrepreneurship, mobilize the enthusiasm of college students’ entrepreneurship, and solidly promote the progress of college students’ entrepreneurship. In addition, we can join hands with enterprises, banks and even government departments to build a capital market for college students’ entrepreneurship, so as to solve the difficulty of starting capital for college students’ employment and entrepreneurship. Second, the government should build a social support and guarantee system for student entrepreneurship. The government should coordinate the relationship between the relevant ministries and departments, increase the policy training, and promote the implementation of the policy among college students. The relevant ministries and departments can set up entrepreneurship guidance organizations to provide employment and entrepreneurship training directly to college students or the general public, and they can also w guide and supervise universities to offer relevant employment and entrepreneurship courses. The government should introduce more and more favorable policies to encourage and support the entrepreneurial activities of university students. For example, it should reduce or waive tuition fees, provide tax incentives, simplify registration procedures, and open entrepreneurial parks so that university students can carry out their entrepreneurial activities without too many worries.