Post feature evaluation model and system based on machine learning

Abstract

The invention relates to the technical field of computer software, and discloses a post characteristic evaluation model and system based on machine learning, and the model and system comprise post recruitment data collection, taking the collected data as analyzed sample data, carrying out the ETL processing of the sample data, and obtaining a post characteristic evaluation result; and performing feature extraction, principal component analysis, clustering and grading on the sample data after ETL processing, and classifying new sample data according to an analysis result after clustering. According to the invention, the user group is oriented to working personnel of higher vocational education schools, and can also be used for data analysis of vocational training institutions or industry analysis consultation units. The method can effectively reduce the influence of factors such as manual judgment level, provides scientific, objective and quantitative evaluation of post characteristics, improves the efficiency and accuracy of post investigation, and is especially suitable for the industry field in which post technical knowledge is rapidly updated.

Description

Technical Field

[0001] The present invention relates to the field of computer software technology, and in particular to a job characteristic evaluation model and system based on machine learning. Background Art

[0002] The Ministry of Education pointed out that the scale and increase of graduates from ordinary colleges and universities in 2022 hit a record high, and the employment situation was complex and severe, requiring "continuous deepening of higher education reform and promotion of school-enterprise supply and demand docking in terms of human resources improvement". In order to implement the school-enterprise supply and demand docking, it is crucial to understand the real job demand characteristics of enterprises. However, at present, the evaluation of job characteristics in the market research stage often relies on expert judgment, which makes the evaluation effect easily affected by factors such as regional scope and human cognition. Therefore, it is necessary to establish a scientific, objective job characteristic evaluation system that meets the real needs of enterprises.

[0003] The present invention provides an innovative job characteristic evaluation method. By using machine learning technology, the quantitative evaluation of key job characteristics and the quantification of job classification and grading can be achieved by simply using the enterprise recruitment data as the training set sample of the present invention. If the training set sample is synchronized with the enterprise recruitment data, it can also be used to compare the qualitative problem of whether the job characteristics of professional training in higher education are keeping pace with the times, so it is very suitable for industries with fast job demand updates. Summary of the invention

[0004] In view of this, the present invention adopts a machine learning algorithm to analyze positions in three aspects: classification, grading and key feature composition of positions, and establishes an evaluation model for position characteristics.

[0005] The system structure of the present invention includes a data acquisition and extraction-transformation-load (ETL) processing module (hereinafter referred to as data acquisition and ETL processing module), an evaluation model module, and a new data sample classification module.

[0006] The system includes 14 functions in total. They are divided into modules: Crawler function, data cleaning, data format processing, and data storage in the data collection and ETL processing module; Feature extraction, word frequency statistics, index design, principal component analysis, cluster analysis, main feature analysis of sub-category positions, and grading functions in the evaluation model module; Feature extraction, principal component analysis, and new data classification in the new data sample classification module.

[0007] The system's workflow goes through seven steps, including data collection, data ETL, feature extraction, principal component analysis, clustering, grading, and classification.

[0008] The data collection and ETL processing module uses web crawlers to obtain and analyze corporate job information, collects job recruitment data publicly released on the Internet by several recruitment websites and corporate official websites (hereinafter referred to as "samples"), and sets up automatic collection and customized collection functions for different recruitment websites according to the technical characteristics of the website; the collected data is cleaned to remove invalid or redundant information. Then the data items are extracted, and four data contents are selected, including job title, job release time, job requirements, and job description. The extracted data item content is then converted into Json format and loaded into a local file for storage.

[0009] The evaluation model module defines the sample data after ETL processing as the training set required for the evaluation model. ,in s Sample data for job recruitment information. n is the total number of samples. TS There are many similar or identical text data in the text, but not all text data are equally important. Therefore, we use word frequency-inverse document frequency ( Term Frequency-Inverse Document Frequency , referred to as TF-IDF) value to extract TS Feature keywords, define feature sets KW , ,in k is the characteristic keyword, w is the TF-IDF value of the feature keyword, , n for s The total number of characteristic keywords, m For collection TS middle s The total number of .

[0010] The more sample data there are, the more features there will be, which will lead to a larger dimension of the feature vector. However, there is a certain degree of correlation between these features, which means that their information will overlap to a certain extent. Therefore, before clustering analysis, KW The dimension of is reduced by principal component analysis (PCA) to reduce data redundancy. In order to complete the dimensionality reduction of the feature vector before clustering, the following processing steps are performed: right KW In kPerform word frequency statistics and define the statistical results as an ordered list , ,in n for KW In k The total number of k is the characteristic keyword. The characteristic matrix is ​​defined according to the word frequency statistics. W , ,in m For collection TS middle s The total number of n For List K middle k The total number of w for k TF-IDF value, column subscript j for k exist K If ,but After completion, you will get a A sparse matrix of

[0011] The matrix W As input to the PCA algorithm, and according to the number of categories required by the K-means clustering algorithm, the PCA algorithm is preset From each principal component, extract the factor with a positive coefficient k , and for the same k After removing duplicates, the factor coefficients are accumulated and defined as an ordered list , ,in n is the total number of factors with positive coefficients, NK Truncate and remove the ones that have little influence on the main components k , defined as an ordered list , .

[0012] Define the feature reduction matrix based on the truncation results C , ,in m For collection TS middle s The total number of t for MK The total number of members, c for MK middle k TF-IDF value, column subscript j for k exist MK If ,but After completion, you will get a The sparse matrix of .

[0013] In the clustering stage, the matrix C As the input of K-means algorithm, we get TS Clustering model of , whose category index list ,in n For collection TS middle s The total number of d is the category index calculated by the clustering algorithm. D Remove duplicate elements and obtain m index value. Therefore, the entire TS Can be divided into m A set of class samples, namely ,in cs It is a set divided by category, and the members in the set come from the matrix C , m is the total number of categories.

[0014] After completing the cluster analysis, cs The main features of cs As input data for the PCA algorithm, define the total contribution rate ,get n The factor loadings of the principal components can reflect the correlation between the original variables and the factors, and define an ordered list of the main characteristic keywords. ,in n is the total number of principal components, fk For feature keywords, this list has two constraints: ;and For the same fk After removing duplicates, the factor loadings are accumulated. The factor loading calculation formula is as follows: ,in Eigenvalue is the eigenvalue of the covariance matrix, Eigenvector For the corresponding Eigenvalue The unit eigenvector of sqrt This is the square root operation.

[0015] because FK The elements in come from the results of PCA, so the proportion of elements The contribution of the principal component.

[0016] Due to different regions, the monthly salary standards for positions of the same category and level are not uniform in different regions; and due to different companies, the educational requirements for positions of the same category and level in the same region are also different in different companies. Therefore, the monthly salary and educational requirements in the sample cannot be used as the criteria for classifying positions. n The contribution rate of the main feature keywords accounts for the highest proportion. Therefore, in the classification stage, the grade factors are defined based on the contribution rate of the main feature keywords to realize the grade classification of positions. The following processing steps are performed: extract cs exist FK Center front n indivual w , defined as and ,in r for fk The contribution rate of ; Defining rank factors , grading standards ,in .

[0017] The new data sample classification module uses the K-Nearest Neighbors (KNN) algorithm to classify new data and predict its category. The following processing steps are performed: Define new samples as sets ,in s is the sample text data, n is the total number of new samples. DS The collection with clustering stage TS consistent in structure; right DS Extract TF-IDF-based feature keywords and define them as a set of TF-IDF values ​​for feature keywords. KW ; By list MK As filtering criteria, define the set DS The TF-IDF value of the feature keyword is reduced in dimension H , ,in m For collection DS In s total, n for MK The total number of members. h for KW middle k TF-IDF value, column subscript j for k exist MK If ,but .

[0018] The model M s The matrix C As the training set data of the KNN algorithm, its category index list D As the label of the training set data.

[0019] In order to achieve the best prediction results, the iterative method is used to calculate the optimal number of neighboring vectors required by the KNN algorithm ( k value).

[0020] The matrix H As the solution set of the KNN algorithm, the inferred k The value is used as the number of neighboring vectors required by the algorithm to predict the set DS The categories in .

[0021] In order to make the above contents of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The foregoing contents of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the accompanying drawings;

[0023] Figure 1 The structure diagram of the job characteristics evaluation system based on machine learning

[0024] Figure 2 Workflow diagram of the job characteristics evaluation system based on machine learning

[0025] Figure 3 Flowchart of principal component analysis

[0026] Figure 4 Flowchart of clustering and principal feature analysis

[0027] Figure 5 The main feature analysis results of this system for the instance sample data

[0028] Figure 6 Flowchart of hierarchical analysis

[0029] Figure 7 The classification results of this system for example samples

[0030] Figure 8 Flowchart of classification analysis

[0031] Fig. 9 Iterative calculation process of the nearest vector number (k value)

[0032] Fig.10 This system is aimed at the new sample set of the embodiment DS Classification results predicted by KNN algorithm DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0034] In order to solve the technical problem that the existing technology is difficult to quickly and accurately identify job characteristics from the text content of job recruitment released by enterprises by means of setting weights, this application proposes a job characteristic evaluation model, which uses a machine learning algorithm to analyze the collected enterprise recruitment data, so as to quickly quantify the evaluation of job characteristics and the classification and grading of jobs. It is conducive to reducing the impact of human factors on job characteristic results and improving the objectivity and accuracy of job research.

[0035] The job characteristic evaluation model provided in the embodiment of the present application is applicable to a terminal device. By way of example, the terminal device includes but is not limited to: a tablet computer, a laptop computer, a desktop computer, etc.

[0036] like Figure 1 , 2 Shown are respectively a structural diagram of the job characteristic evaluation system based on machine learning of the present application and a workflow diagram of the job characteristic evaluation system based on machine learning of the present application.

[0037] Example 1

[0038] Taking several representative recruitment websites (such as Lagou.com, Liepin.com and BOSS Direct Recruitment, etc.) and corporate official websites (such as Alibaba's 1+6+n Recruitment, Tencent Recruitment, Baidu Recruitment, ByteDance Recruitment, etc.), the big data job recruitment information that can be queried from June 2023 to August 2023 is taken as an example.

[0039] In the first step, the data collection and ETL processing module first automatically obtains and parses the job recruitment information publicly released on the website through web crawlers. Each piece of recruitment information collected is regarded as a sample data that has not been evaluated and analyzed by this system.

[0040] In the second step, the data collection and ETL processing module cleans the collected data to remove invalid or redundant information; then extracts data items and selects four data items, including job title, job release time, job requirements, and job description; finally, converts the extracted data items into Json data format and loads them into local files for storage. After ETL processing (including data cleaning, data format organization and unification), the collected sample data becomes the training set required for the evaluation model. TS .

[0041] Step 3: Evaluate the model module on the training set TS First, read the training set from the local file TS The data in TS There are many similar or identical text data in the text, but not all text data are equally important, so we use word frequency-inverse document frequency ( Term Frequency-Inverse Document Frequency , referred to as TF-IDF) extraction TS The feature keywords in the feature set form a key-value pair relationship with their TF-IDF values, and the feature extraction results are defined as feature sets. KW , feature set KW Each row of feature vectors in the training set TS The feature extraction result of a sample data in the training set TS Taking a sample data with a job title of "Big Data Direction (Algorithm / AutoML / Engine)" as an example, the key-value pair format of the extracted feature keywords and their TF-IDF values ​​is as follows:

[0042] [('Engine', 0.9864606051551998), ('Data', 0.6375142237853333), ('Data mining', 0.35220081257066665), ('Algorithm', 0.3476476599652), ('Frontier technology',0.3277663930533333), ('Field', 0.2886557622618667), ('Capability', 0.2632182975024), ('Insight', 0.26150221572666665), ('Technology', 0.25170438285706664), ('Solution', 0.22847890289893333), ('Scenario', 0.21669669617146667), ('Competitiveness', 0.18213209452586668), ('Machine Learning', 0.15939690003866666), ('Core', 0.15606742800826667), ('Strategy', 0.1533928529528), ('Open Source', 0.14412846931466666), ('Signpost', 0.14190108152), ('Time Series', 0.13649488007733335), ('Graph', 0.1312031285524), ('Industry', 0.12374666787066665)]. The feature extraction result of such a sample data becomes the feature set KW Each row of the feature vector in .

[0043] The fourth step is Figure 3 As shown, the evaluation model evaluates the feature set KW Perform principal component analysis to achieve dimensionality reduction. KW The key names of the feature vectors in the word frequency statistics are sorted according to the word frequency statistics results. KW Mapped into a full matrix containing only TF-IDF values W ,matrix W Each row vector in represents a sample data TF-IDF value vector; then the matrix W Perform principal component analysis, extract characteristic keywords with factor coefficient values ​​greater than 0 from each principal component, and accumulate the factor coefficients of the same characteristic keywords; sort the characteristic keywords in descending order based on the accumulated factor coefficients, and calculate the factor coefficients according to the formula The calculation result of is used to truncate the ranking of the characteristic keywords, where n is the total number of factors, and the truncated result is defined as an ordered list of feature keywords MK;Finally, an ordered list based on the characteristic keywords MK , from the matrix W The characteristic keywords that have a great influence on the main components are selected, and the results after screening are defined as the characteristic dimension reduction matrix C The comparison of the results before and after the dimensionality reduction of the sample data of the big data job embodiment by this system is shown in Table 1.

[0044] Table 1 Comparison before and after dimensionality reduction

[0045] Step 5: Figure 4 As shown, the evaluation model reduces the dimension of the feature matrix C Perform cluster analysis on the job sample data and perform main feature analysis on the job categories after clustering. First, reduce the feature dimension matrix C As the input of K-means algorithm, get the clustering result Ms Output; then remove the duplicate elements in its category index value list to obtain the clustered category index value, and divide the training set into TS Divide into category sample sets, that is, ,in cs It is a set divided by category, and the members in the set come from the matrix C , m is the total number of categories. TS The results after clustering are shown in Table 2. cs Perform principal component analysis, filter out the feature words with the largest factor loadings from the principal component analysis results, add up the factor loadings of the same feature words, and define the screening results as cs List of main features FK , and list the main features according to the rule of factor loading from large to small FK The factor loading calculation formula is as follows: ,in, Eigenvalue is the eigenvalue of the covariance matrix, Eigenvector For the corresponding Eigenvalue The unit eigenvector of sqrt is the square root operation. FK The elements in come from the results of principal component analysis, so the proportion of the elements represents the contribution rate to the principal component. The results of the main feature analysis of the job categories after clustering the big data job implementation example by this system are as follows: Figure 5 shown.

[0046] Table 2 Sample training set of big data job implementation examples TS Category statistics after clustering

[0047] Step 6: Figure 6 As shown, the evaluation model is used for the category set cs First, extract cs In the main features list FK The TF-IDF value and contribution rate corresponding to the top main feature words in the TF-IDF value vector are defined as and contribution rate vector ; Then define the level factor , grading standards ,in . Divide the category set according to the level classification standard cs Internal job level. The results of this system's hierarchical analysis of the job categories after clustering the big data job implementation example are as follows: Figure 7 shown.

[0048] Step 7: Figure 8 As shown in Figure 2, the evaluation model performs classification analysis on the newly collected job samples. First, the newly collected samples are defined as a set DS ,gather DS The data structure is similar to the training set defined in the first step. TS agree; then DS Perform keyword extraction based on TF-IDF features and define the extraction results as DS The feature set DW ,gather DW In data structure and feature collection KW Next, use the list in step 4 MK As filtering criteria, define the set DS The feature reduction matrix H In order to achieve the best classification prediction results, the iterative method is used to calculate the optimal number of neighboring vectors required by the K-Nearest Neighbor (KNN) algorithm ( k value), such as Fig. 9 As shown; Finally, the dimension reduction matrix in the fourth step C As the training set data of the KNN algorithm, the matrix H As the solution set of the KNN algorithm, the inferred k The value is used as the number of neighboring vectors required by the algorithm to predict the set DS The system performs a new sample set on the embodiment. DS The classification results predicted by the KNN algorithm are as follows Fig.10 shown.

[0049] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method and system for evaluating job characteristics based on machine learning, characterized in that: The model and system include: Job recruitment data collection, the collected data is regarded as sample data to be analyzed; Perform extract-transform-load (ETL) processing on sample data; Perform feature extraction, principal component analysis, clustering, and grading on sample data after ETL processing; Classify the new sample data according to the analysis results after clustering.

2. The model and system according to claim 1, characterized in that: The job recruitment data collection includes using a web crawler to automatically obtain and analyze the job recruitment data publicly released on the website.

3. The model and system according to claim 2, characterized in that: The ETL processing of sample data includes data cleaning function, which is used to remove invalid or redundant information; data format processing function, which is used to convert data into a unified format; and data storage function, which is used to store data from different sources. And ETL processing is connected with data collection in the workflow.

4. The model and system according to claim 3, characterized in that: The feature extraction of the sample data after ETL processing includes defining a training set according to the sample data. ,in s Sample data for job recruitment information. n is the total number of samples. Using word frequency-inverse document frequency ( Term Frequency-Inverse Document Frequency , referred to as TF-IDF) extraction TS Feature keywords, define feature sets KW , in k is the characteristic keyword, w is the TF-IDF value of the feature keyword, , n for s The total number of characteristic keywords, m For collection TS middle s And feature extraction is connected with ETL processing in the workflow.

5. The model and system according to claim 4, characterized in that: The principal component analysis of the sample data after feature extraction includes: KW of k Perform word frequency statistics and define the statistical results as an ordered list , ,in n for KW In k The total number of k is the characteristic keyword. The characteristic matrix is ​​defined according to the word frequency statistics. W , in m For collection TS middle s The total number of n For List K middle k The total number of w for k TF-IDF value, column subscript j for k exist K If ,but . For the full matrix of feature keywords W Perform principal component analysis and extract factors with positive coefficients. k , and for the same k After removing duplicates, the factor coefficients are accumulated and defined as an ordered list , ,in n is the total number of factors with positive coefficients, NK Truncate and remove the ones that have little influence on the main components k , defined as an ordered list , . Define the feature dimension reduction matrix based on the truncation result in m For collection TS middle s The total number of t for MK The total number of members, c for MK middle k TF-IDF value, column subscript j for k exist MK If ,but . And principal component analysis is connected with feature extraction in the workflow.

6. The model and system according to claim 5, characterized in that: The clustering of the sample data after principal component analysis includes: C Perform K-means analysis to obtain the training set TS The clustering results , whose category index list ,in n For collection TS middle s The total number of d is the category index obtained after K-means analysis. D Remove duplicate elements and obtain m Index value. m The index value of the training set TS Divide into m A set of class samples, namely ,in cs It is a set divided by category, and the members in the set come from the matrix C , m is the total number of categories. cs Perform principal component analysis and define cs Ordered list of main features ,in n is the total number of principal components, fk is the characteristic keyword, ; . For the list FK The same fk After removing duplicates, the factor loadings are accumulated. The factor loading calculation formula is as follows: ,in, Eigenvalue is the eigenvalue of the covariance matrix, Eigenvector For the corresponding Eigenvalue The unit eigenvector of sqrt is the square root operation. And cluster analysis is connected with principal component analysis in the workflow.

7. The model and system according to claim 6, characterized in that: The classification of the clustered sample data includes extracting cs exist FK Center front n indivual w , defined as and ,in r for fk The factor loading of ; Define the level factor , grading standards ,in . And hierarchical analysis is connected to cluster analysis in the workflow.

8. The model and system according to claim 6, characterized in that: The classification of new sample data according to the analysis results after clustering includes defining the new job information samples as a set ,in s is the text data of the job information, n is the total number of new samples. DS The collection with clustering stage TS The structure is consistent. DS Perform keyword extraction based on TF-IDF features and define the extraction results as DS The feature set DW ,gather DW In data structure and feature collection KW Consistent. List MK As filtering criteria, define the set DS The TF-IDF value of the feature keyword is reduced in dimension H , in m For collection DS In s total, n for MK The total number of members. h for DW middle k TF-IDF value, column subscript j for k exist MK If ,but . The model M s The matrix C As the training set data of K-Nearest Neighbor (KNN) algorithm, its category index list D As the label of the training set data. The optimal number of neighboring vectors required by the KNN algorithm is calculated using an iterative method. k Value. H As the solution set data of the KNN algorithm, the inferred k The value is used as the number of neighboring vectors required by the algorithm to predict the set DS And classification analysis is connected with cluster analysis in the workflow.

9. A job characteristic evaluation system based on machine learning, characterized in that: The method applied to any one of claims 1 to 8 above comprises a data collection and ETL processing module, an evaluation model module, and a new data sample classification module; The data collection and ETL processing module is used to crawl job recruitment information from recruitment websites and corporate websites, and clean, format and unify the collected data; The evaluation model module is used to cluster, grade and analyze the characteristics of job sample data based on machine learning algorithms; The new data sample classification module is used to automatically classify new job data samples.

Images