Data and knowledge-driven postgraduate culture quality analysis method

Abstract

The invention provides a data and knowledge driven postgraduate student cultivation quality analysis method, and aims at solving the problems of large student cultivation effect and quality difference, evaluation index diversification and the like in the postgraduate student cultivation process to achieve the purpose of scientifically and objectively analyzing the postgraduate student cultivation quality. According to the method, the principal component analysis method is used for carrying out feature variable mining on the graduate student system data, the correlation analysis method is used for preprocessing the data, Min-Max standardization is used for normalizing the data, and the graduate student culture quality related feature indexes are mined, so that the problem of high data dimensionality of graduate student quality analysis is solved, and the quality of the graduate student is improved. The purpose of quality analysis data dimension reduction is achieved; a knowledge-driven graduate student cultivation quality evaluation model is designed, reasonable rule knowledge and available data are combined, weighted summation is conducted on relevant characteristic indexes of the graduate student cultivation quality, and a cultivation quality analysis model is obtained; a culture quality evaluation mechanism based on TOPSIS is established, the culture quality is sorted, and grading evaluation is achieved.

Description

Technical Field

[0001] Aiming at the problem that the quality of postgraduate training is difficult to evaluate, the present invention designs a data and knowledge-driven postgraduate training quality analysis method, which belongs to both the field of educational research and the field of data research. Background Art

[0002] New policies and new requirements for graduate education emerge in an endless stream, posing severe challenges to the information construction of colleges and universities. In response to the rapid growth of graduate student scale, multiple graduate student categories, increased management difficulty, and inadequate policy measures, the continuous upgrading and transformation of graduate education quality analysis methods and the updating of management system functions are effective ways to solve the problems. With the expansion of graduate student enrollment, the number of degrees awarded has also increased year by year. In order to further improve the quality of graduate education and improve the efficiency of degree management, the degree review system will be transformed from a single meeting process management to degree data monitoring and sharing, and digital methods will be used to empower graduate degree work and promote high-quality development of graduate education. To improve the quality of graduate education, it is necessary to strengthen the construction of a graduate education quality assurance system, and building a graduate education quality evaluation system is an important part of it.

[0003] The quality evaluation of graduate education is an important indicator to guide the high-quality development of training programs and education. However, the scale of graduate education is growing rapidly, management is becoming more and more difficult, new policies are not implemented in place, and supervision and management are superficial. At the same time, big data analysis and management are far from enough. In the face of the rapid development of graduate education, how to continuously innovate and improve management efficiency to meet the requirements of the country and society for the quality of graduate education has become an urgent problem to be solved. Therefore, designing a data and knowledge-driven graduate education quality analysis method can not only improve the efficiency of graduate education quality analysis, but also ensure the scientific objectivity, fairness and justice of graduate education quality evaluation.

[0004] By analyzing the characteristics of the graduate student training process, the present invention develops a characteristic variable method for graduate student training data, designs a knowledge-driven graduate student training quality assessment model, establishes a graduate student training quality evaluation mechanism based on TOPSIS, and realizes a scientific and objective analysis of graduate student training quality. Summary of the invention

[0005] The present invention proposes a data and knowledge-driven graduate education quality analysis method, which realizes scientific and objective analysis of graduate education quality. The method designs characteristic variables of graduate education process based on principal component analysis, designs a knowledge-driven graduate education quality assessment model, and establishes a graduate education quality evaluation mechanism based on TOPSIS.

[0006] The present invention adopts the following technical solutions and implementation steps:

[0007] 1. A data- and knowledge-driven method for analyzing the quality of graduate education, including:

[0008] Based on principal component analysis, the characteristic variables of graduate training process data are mined, and the graduate training process data needs to be preprocessed and normalized; a knowledge-driven graduate training quality assessment model is designed; and a graduate training quality evaluation mechanism based on TOPSIS is established.

[0009] (1) Mining characteristic variables of graduate training process data based on principal component analysis

[0010] The data of graduate student training process were preprocessed, and the 2019 graduate student training data of a university in Beijing was used as sample data for research, which included data from various training stages of graduate students from admission to graduation.

[0011] The data sample is represented as

[0012] Ω={X1,X2,...,X 11} (1)

[0013] Among them, Ω is the data set, X1 is the basic information of the students, X2 is the admission score, X3 is the English CET-4 and CET-6 scores, X4 is the course score, X5 is the student's project information, X6 is the student's paper blind review score, X7 is the paper publication score, X8 is the number of patents, X9 is the number of software copyrights, and X 10 The number of monographs written, X 11 The score for the award.

[0014] Analyze the correlation of data and remove redundant data. The correlation coefficient between data is:

[0015]

[0016] Among them, X i is the i-th data, X j is the jth data, Cov(X i ,X j ) is X i and X j The covariance of DX i For X i The standard deviation of DX j For X j The standard deviation of η = 0.8 is the threshold of the correlation coefficient. 11 The correlation coefficient is greater than or equal to η, which is a strong correlation, that is, a redundant feature. The one with a low correlation coefficient with other data items is eliminated, that is, X 11 The correlation coefficients among the remaining data are less than η, which is a weak correlation, so the data are retained.

[0017] Normalize the data of graduate training process. Use Min-Max standardization to normalize the original data and map its eigenvalues ​​to the interval [0,1]. The calculation formula is:

[0018]

[0019] Among them, u i,n is the nth sample of the i-th data, u i,max is the maximum value of the data, u i,min is the minimum value of the data. After normalization, all samples are between [0,1], and the data samples are expressed in the form of a matrix:

[0020]

[0021] Among them, x i,n is the nth sample of the i-th item of data after normalization, I=10 is the number of sample data items, and N is the number of samples.

[0022] The principal component analysis method is used to mine the characteristic variables of the graduate training process data. The covariance matrix of the calculated data sample X is

[0023]

[0024] Among them, C is the covariance matrix of data sample X, r n,n is the correlation coefficient.

[0025] Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix C, expressed as

[0026] C=VΛV T (6)

[0027] Among them, V is the eigenvector of the covariance matrix, and Λ is the diagonal matrix composed of the eigenvalues ​​of the matrix eigenvector, which is expressed as

[0028]

[0029] Among them, λ k is the kth eigenvalue, K is the order of Λ. k is k The feature vector of .

[0030] Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate of the first M eigenvalues.

[0031]

[0032]

[0033] Among them, λ mis the mth eigenvalue, φ m represents the contribution rate of the mth eigenvalue, κ(M) represents the cumulative contribution rate of M eigenvalues, and the value of M is determined by the cumulative contribution rate. When the cumulative contribution rate κ(M) is greater than or equal to 85%, the corresponding eigenvector is selected to form the transformation matrix Q T , according to Y = Q T X calculates the first M principal components to extract key features, and obtains M=4.

[0034] (2) Knowledge-driven postgraduate training quality assessment model

[0035] The knowledge of the knowledge-driven postgraduate training quality assessment model is rule knowledge. Using rule knowledge to constrain postgraduate training quality assessment:

[0036] Students whose course score X4 is less than 60 points will not be allowed to graduate; students whose project proposal time and graduation application time in X5 are less than the school's stipulated time will not be allowed to graduate; students whose paper blind review score X6 is less than 60 points will not be allowed to graduate; students whose published paper score X7 does not meet the school's requirements will not be allowed to graduate; the importance of students' academic achievements is ranked as follows: published papers>patents>software copyrights>monographs; students who meet the graduation requirements will be graded.

[0037] The knowledge-driven approach is used to establish a model for evaluating the quality of postgraduate education. The principal component loadings are calculated as

[0038]

[0039] Among them, L kp ∈L k For L k By analyzing the principal component loading values, the student paper blind review score X6 is closely related to principal component 1, the published paper score X7 is closely related to principal component 2, the number of patents X8 is closely related to principal component 3, and the number of software copyrights X9 is closely related to principal component 3. Therefore, X6, X7, X8 and X9 are selected as the component indicator vectors for the next step of analysis.

[0040] (3) Establish a TOPSIS-based evaluation mechanism for postgraduate training quality

[0041] TOPSIS is used to establish a mechanism for evaluating the quality of graduate education. Find the optimal index vector A * and the worst indicator vector A -

[0042]

[0043] Among them, x6 * The highest student paper blind review score, x7 * For the highest paper published, x8* The highest number of patents, x9 * The highest number of patents, x6 - The minimum blind review score for student papers, x7 - The minimum paper publication score, x8 - is the minimum number of patents, x9 - The minimum number of patents.

[0044] Calculate sample A to the optimal index vector A * The Euclidean distance is

[0045]

[0046] Among them, A = [x6, x7, x8, x9] is the sample value, x i ∈A is the sample data value, i = 6, 7, 8, 9, x6 is the blind review score of the student paper, x7 is the published paper score, x8 is the number of patents, x9 is the number of patents, is the contribution rate of the mth eigenvalue after processing

[0047]

[0048] Calculate A to the worst index vector A - The Euclidean distance is

[0049]

[0050] Calculate the relative proximity. The formula for calculating the relative proximity of sample A to the optimal indicator vector is as follows

[0051]

[0052] Among them, g is the relative proximity value of sample A. The g value is proportional to the relative proximity of sample A to the optimal indicator vector, that is, the larger g is, the closer sample A is to the optimal indicator vector.

[0053] The calculated value of g is the comprehensive score of student training quality. The larger the g, the higher the ranking. The top 10% of graduate students are rated as "excellent", the top 30% are rated as "good", the top 70% are rated as "medium", the top 100% are rated as "passing", and the quality of graduate students who are "not allowed to graduate" is "failed". BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the postgraduate training quality analysis of the present invention;

[0055] Figure 2 It is the quality indicator of postgraduate training of the present invention. DETAILED DESCRIPTION

[0056] 1. A data and knowledge-driven postgraduate training quality analysis method, comprising the following steps:

[0057] (1) Mining characteristic variables of graduate training process data based on principal component analysis

[0058] The data of graduate student training process were preprocessed, and the 2019 graduate student training data of a university in Beijing was used as sample data for research, which included data from various training stages of graduate students from admission to graduation.

[0059] The data sample is represented as

[0060] Ω={X1,X2,...,X 11} (1)

[0061] Among them, Ω is the data set, X1 is the basic information of the students, X2 is the admission score, X3 is the English CET-4 and CET-6 scores, X4 is the course score, X5 is the student's project information, X6 is the student's paper blind review score, X7 is the paper publication score, X8 is the number of patents, X9 is the number of software copyrights, and X 10 The number of monographs written, X 11 The score for the award.

[0062] Analyze the correlation of data and remove redundant data. The correlation coefficient between data is:

[0063]

[0064] Among them, X i is the i-th data, X j is the jth data, Cov(X i ,X j ) is X i and X j The covariance of DX i For X i The standard deviation of DX j For X j The standard deviation of η = 0.8 is the threshold of the correlation coefficient. 11 The correlation coefficient is greater than or equal to η, which is a strong correlation, that is, a redundant feature. The one with a low correlation coefficient with other data items is eliminated, that is, X 11 The correlation coefficients among the remaining data are less than η, which is a weak correlation, so the data are retained.

[0065] Normalize the data of graduate training process. Use Min-Max standardization to normalize the original data and map its eigenvalues ​​to the interval [0,1]. The calculation formula is:

[0066]

[0067] Among them, u i,n is the nth sample of the i-th data, u i,max is the maximum value of the data, u i,min is the minimum value of the data. After normalization, all samples are between [0,1], and the data samples are expressed in the form of a matrix:

[0068]

[0069] Among them, x i,n is the nth sample of the i-th item of data after normalization, I=10 is the number of sample data items, and N is the number of samples.

[0070] The principal component analysis method is used to mine the characteristic variables of the graduate training process data. The covariance matrix of the calculated data sample X is

[0071]

[0072] Among them, C is the covariance matrix of data sample X, r n,n is the correlation coefficient.

[0073] Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix C, expressed as

[0074] C=VΛV T (6)

[0075] Among them, V is the eigenvector of the covariance matrix, and Λ is the diagonal matrix composed of the eigenvalues ​​related to the matrix eigenvector, which is expressed as

[0076]

[0077] Among them, λ k is the kth eigenvalue, K is the order of Λ. k is k The feature vector of .

[0078] Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate of the first M eigenvalues.

[0079]

[0080]

[0081] Among them, λ m is the mth eigenvalue, φ mrepresents the contribution rate of the mth eigenvalue, κ(M) represents the cumulative contribution rate of M eigenvalues, and the value of M is determined by the cumulative contribution rate. When the cumulative contribution rate κ(M) is greater than or equal to 85%, the corresponding eigenvector is selected to form the transformation matrix Q T , according to Y = Q T X calculates the first M principal components to extract key features, and obtains M=4.

[0082] (2) Knowledge-driven postgraduate training quality assessment model

[0083] The knowledge of the knowledge-driven postgraduate training quality assessment model is rule knowledge. Using rule knowledge to constrain postgraduate training quality assessment:

[0084] Students whose course score X4 is less than 60 points will not be allowed to graduate; students whose project proposal time and graduation application time in X5 are less than the school's stipulated time will not be allowed to graduate; students whose paper blind review score X6 is less than 60 points will not be allowed to graduate; students whose published paper score X7 does not meet the school's requirements will not be allowed to graduate; the importance of students' academic achievements is ranked as follows: published papers>patents>software copyrights>monographs; students who meet the graduation requirements will be graded.

[0085] The knowledge-driven approach is used to establish a model for evaluating the quality of postgraduate education. The principal component loadings are calculated as

[0086]

[0087] Among them, L kp ∈L k For L k By analyzing the principal component loading values, the student paper blind review score X6 is closely related to principal component 1, the published paper score X7 is closely related to principal component 2, the number of patents X8 is closely related to principal component 3, and the number of software copyrights X9 is closely related to principal component 3. Therefore, X6, X7, X8 and X9 are selected as the component indicator vectors for the next step of analysis.

[0088] (3) Establish a TOPSIS-based evaluation mechanism for postgraduate training quality

[0089] TOPSIS is used to establish a mechanism for evaluating the quality of graduate education. Find the optimal index vector A * and the worst indicator vector A -

[0090]

[0091] Among them, x6 * The highest student paper blind review score, x7 * For the highest paper published, x8 * The highest number of patents, x9* The highest number of patents, x6 - The minimum blind review score for student papers, x7 - The minimum paper publication score, x8 - is the minimum number of patents, x9 - The minimum number of patents.

[0092] Calculate sample A to the optimal index vector A * The Euclidean distance is

[0093]

[0094] Among them, A = [x6, x7, x8, x9] is the sample value, x i ∈A is the sample data value, i = 6, 7, 8, 9, x6 is the blind review score of the student paper, x7 is the published paper score, x8 is the number of patents, x9 is the number of patents, is the contribution rate of the mth eigenvalue after processing

[0095]

[0096] Calculate A to the worst index vector A - The Euclidean distance is

[0097]

[0098] Calculate the relative proximity. The formula for calculating the relative proximity of sample A to the optimal indicator vector is as follows

[0099]

[0100] Among them, g is the relative proximity value of sample A. The g value is proportional to the relative proximity of sample A to the optimal indicator vector, that is, the larger g is, the closer sample A is to the optimal indicator vector.

[0101] The calculated value of g is the comprehensive score of student training quality. The larger the g, the higher the ranking. The top 10% of graduate students are rated as "excellent", the top 30% are rated as "good", the top 70% are rated as "medium", the top 100% are rated as "passing", and the quality of graduate students who are "not allowed to graduate" is "failed".

Claims

1. A data- and knowledge-driven approach to analyzing the quality of graduate education. include: Mining characteristic variables of graduate student training process data based on principal component analysis, where the graduate student training process data needs to be preprocessed and normalized; Design a knowledge-driven graduate education quality assessment model; establish a graduate education quality evaluation mechanism based on TOPSIS; (1) Mining characteristic variables of graduate training process data based on principal component analysis Preprocess the data of graduate training process, including data of each training stage from admission to graduation; The data sample is represented by Ω = {X 1 ,X 2 ,...,X 11 } (1) Among them, Ω is the data set, X 1 Basic information for students, X 2 is the admission score, X 3 is the English level 4 and 6 scores, X 4 is the course grade, X 5 Provide information for students to start the project, X 6 Blind review grades for student papers, X 7 For the publication of papers, X 8 is the number of patents, X 9 is the number of soft-selling, X 10 The number of monographs written, X 11 The score for the award; Analyze the correlation of data and remove redundant data; the correlation coefficient between data is: Among them, X i is the i-th data, X j is the jth data, Cov(X i ,X j ) is X i and X j The covariance of DX i For X i The standard deviation of DX j For X j The standard deviation of; η = 0.8 is the threshold of the correlation coefficient; X 7 and X 11 The correlation coefficient is greater than or equal to η, which is a strong correlation, that is, a redundant feature. The one with a low correlation coefficient with other data items is eliminated, that is, X 11 ; The correlation coefficient between the remaining data is less than η, which is a weak correlation, so the data is retained; Normalize the data of the graduate training process; use Min-Max standardization to normalize the original data and map its eigenvalues ​​to the interval [0,1]. The calculation formula is: Among them, u i,n is the nth sample of the i-th data, u i,max is the maximum value of the data, u i,min is the minimum value of the data; after normalization, all samples are between [0,1], and the data samples are expressed in the form of a matrix: Among them, x i,n is the nth sample of the i-th data after normalization, I=10 is the number of sample data items, and N is the number of samples; The principal component analysis method is used to mine the characteristic variables of the graduate training process data; the covariance matrix of the calculated data sample X is Among them, C is the covariance matrix of data sample X, r n,n is the correlation coefficient; Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix C, expressed as C=VΛV T (6) Among them, V is the eigenvector of the covariance matrix, and Λ is the diagonal matrix composed of the eigenvalues ​​related to the matrix eigenvector, which is expressed as Among them, λ k is the kth eigenvalue, K is the order of Λ; L k is k The eigenvector of Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate of the first M eigenvalues. Among them, λ m is the mth eigenvalue, φ m represents the contribution rate of the mth eigenvalue, κ(M) represents the cumulative contribution rate of M eigenvalues, and the value of M is determined by the cumulative contribution rate. When the cumulative contribution rate κ(M) is greater than or equal to 85%, the corresponding eigenvector is selected to form the transformation matrix Q T , according to Y = Q T X calculates the first M principal components to extract key features; M = 4; (2) Knowledge-driven postgraduate training quality assessment model The knowledge of the knowledge-driven postgraduate training quality assessment model is rule knowledge; rule knowledge is used to constrain the postgraduate training quality assessment: Course Grade X 4 Those with less than 60 points will not be allowed to graduate; X 5 If the time between the thesis proposal and the graduation application is less than the time stipulated by the school, the student will not be allowed to graduate; the blind review score of the student's paper is X 6 Students with a score below 60 will not be admitted for graduation; those with a score of X 7 Students who do not meet the school's requirements will not be allowed to graduate; the importance of students' academic achievements is ranked as follows: published papers > patents > software works > monographs; students who meet the graduation requirements will be graded; The knowledge-driven method is used to establish a postgraduate training quality assessment model; the principal component loading is calculated as Among them, L kp ∈L k For L k By analyzing the principal component loading values, the principal component 1 is closely related to the student paper blind review score X 6 , which is closely related to principal component 2 is the published paper score X 7 , the number of patents X is closely related to principal component 3 8 , which is closely related to principal component 3 is the number of soft spots X 9 So choose X 6 , X 7 , X 8 and X 9 As the component indicator vector for the next step of analysis; (3) Establish a TOPSIS-based evaluation mechanism for postgraduate training quality TOPSIS is used to establish a quality evaluation mechanism for postgraduate training; find the optimal indicator vector A * and the worst indicator vector A - Among them, x 6 * is the highest student paper blind review score, x 7 * is the highest paper published, x 8 * is the highest number of patents, x 9 * is the highest number of patents, x 6 - is the minimum blind review score of student papers, x 7 - is the minimum paper publication score, x 8 - is the minimum number of patents, x 9 - is the minimum number of patents; Calculate sample A to the optimal index vector A * The Euclidean distance is Where A=[x 6 ,x 7 ,x 8 ,x 9 ] is the sample value, x i ∈A is the sample data value, i=6,7,8,9,x 6 is the blind review score of the student paper, x 7 For the paper publication results, x 8 is the number of patents, x 9 is the number of patents, is the contribution rate of the mth eigenvalue after processing Calculate A to the worst index vector A - The Euclidean distance is Calculate the relative proximity. The formula for calculating the relative proximity of sample A to the optimal indicator vector is as follows Among them, g is the relative proximity value of sample A; the g value is proportional to the relative proximity of sample A to the optimal indicator vector, that is, the larger the g, the closer sample A is to the optimal indicator vector; The value of g is calculated, which is the comprehensive score of student training quality. The larger the g, the higher the ranking. The training quality of the top 10% of graduate students is "excellent", the training quality of the top 30% of graduate students is "good", the training quality of the top 70% of graduate students is "average", the training quality of the top 100% of graduate students is "passing", and the training quality of graduate students who are "not allowed to graduate" is "failed".

Images