Accurate matching recruitment system based on big data analysis

Abstract

The present invention relates to the technical field of data processing, and discloses a precise matching recruitment system based on big data analysis, comprising: loading description documents of all recruitment positions of an enterprise, and using the TF-IDF method to screen out a common word set and a unique word set of each recruitment position from the description documents; screening out a target position set from all recruitment positions according to the common word set; retrieving a job seeker's resume to be matched and a keyword set in the resume to be matched, and extracting the common word set and the unique word set of the target position, and inputting the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-position match is consistent, to obtain a judgment result; and summarizing all positions that are compatible with the job seeker from the target position set according to the judgment result; the present invention is conducive to avoiding repeated resume screening and interviews, and improving the accuracy and efficiency of enterprise recruitment.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a precise matching recruitment system based on big data analysis. Background Art

[0002] As the scale of enterprises expands and industry competition intensifies, the recruitment process has become increasingly complex and time-consuming, especially when it comes to large-scale recruitment and multi-position recruitment. The traditional recruitment process usually relies on manual screening by the human resources department, which is not only time-consuming and labor-intensive, but also difficult to achieve efficient and accurate matching of people and positions, especially in large-scale recruitment activities; and traditional recruitment methods are often unable to effectively identify and process recruitment information for multiple positions with similar functions, resulting in repeated resume screening, repeated interviews and other problems, which not only wastes the company's time and resources, but also reduces the efficiency and accuracy of recruitment.

[0003] In human resource management, how to quickly and accurately match the most suitable candidates for job requirements from a large number of job seekers has become the key to improving recruitment efficiency and competitiveness of enterprises. However, existing recruitment systems based on big data analysis mostly focus on improving resume matching rates. In the face of large-scale recruitment and multiple positions with similar functions, these systems often seem to be unable to effectively solve the problems of repeated resume screening and repeated interviews of candidates. At the same time, existing technologies lack intelligent means to accurately identify and match the relationship between multiple positions and candidates, making it difficult to meet the urgent needs of enterprises for efficient and accurate recruitment. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a precise matching recruitment system based on big data analysis.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A precise matching recruitment system based on big data analysis, the system comprising:

[0007] The feature screening module is used to load the description documents of all recruitment positions of the enterprise, and use the TF-IDF method to screen out the common word set and the unique word set of each recruitment position from the description documents; the common word set contains P common words, and the unique word set contains Q unique words, P, Q∈N + , N + is a positive integer;

[0008] The job screening module is used to screen out a target job set from all recruitment jobs based on the common word set; the target job set contains K target jobs with similar job functions, K∈N + , N + is a positive integer;

[0009] The matching judgment module is used to retrieve the job seeker's resume to be matched and the keyword set in the resume to be matched, and extract the common word set and unique word set of the target position, and input the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, to obtain a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer;

[0010] The person-job matching module is used to summarize all job sets that are suitable for job seekers from the target job set according to the judgment results; the job set contains R matching jobs that are suitable for job seekers, R∈N + , N + Is a positive integer.

[0011] Furthermore, before filtering out the common word sets and unique word sets for each recruitment position, including:

[0012] a1: Segment the description documents of all job positions and remove meaningless characters to obtain S preprocessed description documents. Each preprocessed description document includes M phrases, S, M∈N + , N + is a positive integer;

[0013] a2: Extract the jth phrase in the i-th preprocessed description document, i, j∈N + , N + is a positive integer;

[0014] a3: Input the j-th phrase into the pre-trained decision tree model for predicting information gain to obtain the information gain value of the j-th phrase;

[0015] a4: Compare the information gain value of the j-th word group with the preset information gain value threshold. If the information gain value is less than or equal to the information gain value threshold, remove the j-th word group from the i-th preprocessed description document, set i=i+1, and return to step a2; if the information gain value is greater than the information gain value threshold, retain the j-th word group in the i-th preprocessed description document, set i=i+1, and return to step a2;

[0016] a5: Repeat steps a2 to a4 above until i=M, then end the loop and obtain the i-th updated description document, which contains N phrases, and set j=j+1, and return to step a2; N≤M, N∈N + , N + is a positive integer;

[0017] a6: Repeat the above steps a2 to a5 until j=S, then end the loop and obtain S updated description documents.

[0018] Furthermore, the training logic of the decision tree model for predicting information gain is as follows:

[0019] Acquire historical information gain value training data, and divide the historical information gain value training data into an information gain value training set and an information gain value test set; the historical information gain value training data includes a plurality of phrases in vector form and their corresponding information gain values;

[0020] Construct a decision tree network, use the phrases in the form of vectors in the information gain value training set as the input of the decision tree network, and use the information gain values ​​in the information gain value training set as the output of the decision tree network, train the decision tree network, and obtain an initialized decision tree regression network;

[0021] The information gain value test set is used to verify the model of the initialized decision tree regression network, and the initialized decision tree regression network with an error value less than or equal to the preset error threshold is output as the decision tree model for predicting information gain.

[0022] Furthermore, the information gain value in the historical information gain value training data is obtained by processing using an information gain measurement method;

[0023] The processing logic of the information gain measurement method is as follows:

[0024] Obtain historical job matching sample data, which includes U job matching samples. Each job matching sample includes multiple feature words and their corresponding classification labels (target variables). The classification labels (target variables) include matching and non-matching. U∈N + , N + is a positive integer;

[0025] The initial entropy is calculated based on U job matching samples. The specific calculation formula is as follows:

[0026] ;

[0027] Where: represents the initial entropy, represents the number of categories of the classification label (target variable), Represents the probability of the rth category classification label (target variable);

[0028] Calculate the conditional entropy of the feature word, the specific calculation formula is as follows:

[0029] ;

[0030] Where: represents the conditional entropy, Indicates that it contains characteristic words The number of possible values ​​of Indicates that it contains characteristic words The value is The conditional probability of Indicates that it contains characteristic words The value is The entropy of the classification label (target variable) when ;

[0031] Calculate the information gain value, the specific calculation formula is as follows:

[0032] ;

[0033] Where: Represents the information gain value.

[0034] Furthermore, the filtering out of the common word set from the description document includes:

[0035] Get each phrase in the updated description document and calculate the DF value of each phrase;

[0036] Extract corresponding phrases whose DF values ​​are greater than or equal to a preset DF threshold, mark corresponding phrases whose DF values ​​are greater than or equal to the preset DF threshold as common words, obtain P common words, and merge the P common words into a common word set.

[0037] Furthermore, the specific word set of each recruitment position is screened out from the description document, including:

[0038] b1: Extract the updated description document corresponding to the g-th job position; and calculate the TF-IDF value of each phrase in the updated description document, g∈N + , N + is a positive integer;

[0039] b2: Compare the TF-IDF value of each phrase with the preset TF-IDF threshold. If the TF-IDF value is greater than or equal to the preset TF-IDF threshold, mark the corresponding phrase as a unique word; if the TF-IDF value is greater than the preset TF-IDF threshold, set g=g+1 and return to step b1;

[0040] b3: Count all the unique words of the g-th job position, obtain Q unique words of the g-th job position, merge the Q unique words of the g-th job position into a unique word set, then obtain the unique word set of the g-th job position, set g=g+1, and return to step b1;

[0041] b4: Repeat the above steps b1 to b3 until all recruitment positions are screened, then end the loop and obtain the unique word set for each recruitment position.

[0042] Furthermore, the target position set is selected from all recruitment positions, including:

[0043] Extract the updated description documents of all job positions and extract the number of occurrences of the common words in the updated description documents;

[0044] The number of occurrences of the common words is compared with the preset occurrence threshold. If the number of occurrences is greater than or equal to the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a target position with similar job functions; if the number of occurrences is less than the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a non-target position;

[0045] Count all corresponding recruitment positions whose number of appearances is greater than or equal to the preset number of appearances threshold, obtain K target positions with similar job functions, and merge the K target positions with similar job functions into a target job set.

[0046] Furthermore, the training logic of the convolutional neural network used to determine whether the person-job match is consistent is as follows:

[0047] Acquire historical person-job matching training data, and divide the historical person-job matching training data into a person-job matching training set and a person-job matching test set, wherein the historical person-job matching training data includes person-job matching feature data and corresponding annotation labels;

[0048] The person-job matching feature data includes the keyword set in the resume to be matched, the common word set and the unique word set of the target position; the marking label includes "0" and "1", wherein "0" indicates that the resume to be matched is not compatible with the target position, and "1" indicates that the resume to be matched is compatible with the target position;

[0049] The logic for generating the annotation labels in the historical person-job matching training data is as follows:

[0050] Calculate the similarity between the keyword set of the resume to be matched and the common word set of the target position to obtain a first similarity;

[0051] Calculate the similarity between the keyword set of the resume to be matched and the unique word set of the target position to obtain a second similarity;

[0052] Substituting the first similarity and the second similarity into a pre-built mathematical calculation model to obtain a matching coefficient between the resume to be matched and the target position;

[0053] The mathematical calculation model is expressed as follows:

[0054] ;

[0055] Where: Indicates the matching coefficient between the resume to be matched and the target position. represents the first similarity, represents the second similarity, and represents a weight factor greater than zero, ;

[0056] The matching coefficient is compared with the preset matching coefficient threshold. If the matching coefficient is greater than or equal to the matching coefficient threshold, it is determined that the resume to be matched is compatible with the corresponding target position; if the matching coefficient is less than the matching coefficient threshold, it is determined that the resume to be matched is not compatible with the corresponding target position;

[0057] Construct a convolutional neural network, use the job matching feature data in the job matching training set as the input of the convolutional neural network, and use the labeled labels in the job matching training set as the output of the convolutional neural network, train the convolutional neural network, and obtain an initial convolutional neural network;

[0058] The initial convolutional neural network model is verified using the person-job matching test set, and an initial convolutional neural network with a preset test accuracy is output as the convolutional neural network used to determine whether the person-job matching is consistent.

[0059] Furthermore, the process of summarizing all the positions that are suitable for the job seeker from the target position set includes:

[0060] Collect all target positions that match the resumes to be matched in the target position collection;

[0061] All target positions that match the resumes to be matched are marked as matching positions that match the job seekers, and R matching positions are obtained.

[0062] A precise matching recruitment method based on big data analysis is implemented according to the precise matching recruitment system based on big data analysis described above, and the method comprises:

[0063] Load the description documents of all the recruitment positions of the enterprise, and use the TF-IDF method to filter out the common word set and the unique word set of each recruitment position from the description documents; the common word set contains P common words, and the unique word set contains Q unique words, P, Q∈N + , N + is a positive integer;

[0064] The target position set is selected from all recruitment positions based on the common word set; the target position set contains K target positions with similar job functions, K∈N + , N+ is a positive integer;

[0065] Retrieving the job seeker's resume to be matched and the keyword set in the resume to be matched, as well as extracting the common word set and the unique word set of the target position, inputting the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, and obtaining a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer;

[0066] According to the judgment result, a set of all positions that are suitable for the job seeker is summarized from the target position set; the position set contains R matching positions that are suitable for the job seeker, R∈N + , N + Is a positive integer.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] (1) The present application discloses a precise matching recruitment system based on big data analysis, comprising: loading description documents of all recruitment positions of an enterprise, using the TF-IDF method to filter out common word sets and unique word sets of each recruitment position from the description documents; filtering out a target position set from all recruitment positions based on the common word sets; retrieving the job seeker's resume to be matched and the keyword set in the resume to be matched, and extracting the common word set and unique word set of the target position, inputting the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, and obtaining a judgment result; summarizing all the positions that are compatible with the job seeker from the target position set based on the judgment result; based on the above content, the present invention can automatically process multiple positions with similar functions, avoid repeated screening of resumes and repeated interviews of candidates, and significantly reduce redundant work in human resource management; compared with the traditional "one-to-one" recruitment method, the present system can handle the needs of multiple positions at one time, greatly improving the efficiency and speed of the recruitment process.

[0069] (2) This application discloses a precise matching recruitment system based on big data analysis. By speeding up the recruitment process and ensuring matching accuracy, enterprises can find suitable employees more quickly and shorten the recruitment cycle. It is suitable for the multi-position recruitment needs of large-scale enterprises and can effectively handle the different needs of multiple recruitment departments, ensuring that enterprises can quickly obtain the required talents. In addition, by reducing the repeated screening of resumes and the number of interviews, it greatly saves recruitment time and human resource costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1A module structure diagram of the precise matching recruitment system based on big data analysis provided by the present invention;

[0071] Figure 2 A flow chart of the precise matching recruitment method based on big data analysis provided by the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] Example 1

[0074] See also Figure 1 As shown, this embodiment discloses a precise matching recruitment system based on big data analysis, and the system includes:

[0075] The feature screening module 101 is used to load the description documents of all the recruitment positions of the enterprise, and use the TF-IDF method to screen out the common word set and the unique word set of each recruitment position from the description documents; the common word set includes P common words, and the unique word set includes Q unique words, P, Q∈N + , N + is a positive integer;

[0076] It should be understood that: enterprises, especially large enterprises, generally publish a large number of recruitment information for different positions in recruitment software. When you search by enterprise name and click on the enterprise information link, you will find a large number of description documents about various positions in the enterprise. In each description document, the specific conditions required for each recruitment position are defined, including but not limited to education, age, salary and skills. Therefore, by loading the description documents of all recruitment positions of a certain enterprise in the recruitment software, you can obtain all the specific conditions required for each recruitment position of the enterprise.

[0077] Specifically, before screening out the common word sets and unique word sets for each recruitment position, including:

[0078] a1: Segment the description documents of all job positions and remove meaningless characters to obtain S preprocessed description documents. Each preprocessed description document includes M phrases, S, M∈N + , N + is a positive integer;

[0079] It should be noted that: the tokenization of the description documents for all recruitment positions is implemented based on existing tokenization tools, such as the Jieba tokenization software, etc.; the meaningless characters include meaningless words and phrases (including but not limited to repeated words, stop words removed, onomatopoeic words, etc.) and meaningful symbols (such as: "!", ",", ".", etc.);

[0080] a2: Extract the j-th phrase from the i-th preprocessed description document, where i, j ∈ N + , N + is a positive integer;

[0081] a3: Input the j-th phrase into the pre-trained decision tree model for predicting information gain to obtain the information gain value of the j-th phrase;

[0082] Specifically, the training logic of the decision tree model for predicting information gain is as follows:

[0083] Obtain historical information gain value training data, and divide the historical information gain value training data into an information gain value training set and an information gain value test set; the historical information gain value training data includes multiple phrases in vector form and their corresponding information gain values;

[0084] It should be noted that: the phrase in vector form is obtained after text vector conversion through the Word2Vec bag-of-words model; the phrase in vector form in the historical information gain value training data is obtained by preprocessing the description documents in the recruitment positions, for details, see the relevant description above;

[0085] Among them, the information gain value in the historical information gain value training data is obtained by processing with the information gain measurement method;

[0086] It should be noted that: the information gain measurement method is used to represent the degree of reduction in uncertainty after introducing feature words, and it can help eliminate worthless words in the description documents; further explanation is that in the subsequent target position screening process, high-frequency phrases (i.e., common words that appear in multiple documents) need to be used to screen out target positions with similar job functions, but some words are of no value. For example, the stop words "de", "shi", "you". Suppose target positions with similar job functions are screened out based on these stop words, and the final screening result will definitely not be representative and applicable because these phrases cannot reflect the functional attributes of the positions; of course, through the above description, it can be understood that stop words have been removed in the actual processing, but in addition to stop words, there are still some neutral or general words that cannot be processed. If these neutral or general words are not further processed, the final screening result will also be inaccurate;

[0087] For example, in some job description documents or resumes, there are usually some description sentences such as "possessing XXX technology" or "being good at XXXX software" etc. Among them, the words "possessing", "technology", "being good at" and "software" often appear in various description documents. If they are used as common words to filter out target positions with similar job functions, the obtained screening results are still not accurate enough, which means that there is no functional similarity or job complementarity between the selected target positions, which in turn leads to the accuracy of subsequent job matching;

[0088] Specifically, the processing logic of the information gain measurement method is as follows:

[0089] Obtain historical job matching sample data, which includes U job matching samples. Each job matching sample includes multiple feature words and their corresponding classification labels (target variables). The classification labels (target variables) include matching and non-matching. U∈N + , N + is a positive integer;

[0090] It should be noted that: the characteristic words of the historical job matching sample data are collected and recorded in advance by the technical staff, and the classification labels (target variables) in the historical job matching sample data are manually classified and labeled by the technical staff;

[0091] The initial entropy is calculated based on U job matching samples. The specific calculation formula is as follows:

[0092] ;

[0093] Where: represents the initial entropy, represents the number of categories of the classification label (target variable), Represents the probability of the rth category classification label (target variable);

[0094] Calculate the conditional entropy of the feature word, the specific calculation formula is as follows:

[0095] ;

[0096] Where: represents the conditional entropy, Indicates that it contains characteristic words The number of possible values ​​of Indicates that it contains characteristic words The value is The conditional probability of Indicates that it contains characteristic words The value is The entropy of the classification label (target variable) when ;

[0097] Calculate the information gain value, the specific calculation formula is as follows:

[0098] ;

[0099] Where: represents the information gain value;

[0100] For example, it is assumed that there are 10 job matching samples for which the technicians have determined whether the resume of the job seeker matches a certain job. The specific data is shown in Table 1 below:

[0101] Table 1: Historical job matching sample data table

[0102] Sample No. Whether it contains the characteristic word "good at" Match (Y or N) 1 yes Y 2 yes Y 3 yes Y 4 yes N 5 yes N 6 yes N 7 no Y 8 no Y 9 no N 10 no N

[0103] According to Table 1 above, there are a total of 10 job matching samples, of which 5 are matches (Y) and 5 are mismatches (N). Therefore, the probability of matching is: , the probability of mismatch is: ; In the formula, Represents the probability that the classification label (target variable) category is matched, It represents the probability that the classification label (target variable) category is mismatched. Substituting it into the initial entropy calculation formula, we have: ;

[0104] In addition, according to the above table, among the 6 job matching samples containing the feature word "excellence", there are 3 matches and 3 mismatches; therefore, it can be seen that the conditional probability of containing the feature word "excellence" and being a match is: , and the conditional probability of containing the feature word "good at" and not matching is: In addition, we know that among the 4 job matching samples that do not contain the feature word "excellence", there are 3 matches and 3 mismatches. Therefore, the conditional probability of not containing the feature word "excellence" and being a match is: , and the conditional probability of not containing the feature word "good at" and not matching is: ; Further, substituting the above data into the calculation of conditional entropy, it can be concluded that the conditional entropy containing the feature word "good at" is: The conditional entropy without the feature word "good at" is: ; Finally, calculate the overall conditional entropy. The overall conditional entropy is weighted according to the probability of the feature word X appearing or not. Since the feature word "good at" appears 6 times in 10 job matching samples, and the feature word "good at" does not appear 4 times in 10 job matching samples, the probability of containing the feature word "good at" is 0.6, and the probability of containing the feature word "good at" is 0.4; Finally, the overall conditional entropy is: ;

[0105] In summary, the information gain value is finally calculated and obtained 1-1=0; It can be understood that when the information gain value is larger, it means that the introduction of feature words will reduce the uncertainty of classification. On the contrary, when the information gain value is smaller, it means that the introduction of feature words has little impact on classification. It can be further understood that the word "good at" does not have value and influence, and cannot highlight the characteristics of the recruitment position, so it should be eliminated in subsequent processing;

[0106] Construct a decision tree network, use the phrases in the form of vectors in the information gain value training set as the input of the decision tree network, and use the information gain values ​​in the information gain value training set as the output of the decision tree network, train the decision tree network, and obtain an initialized decision tree regression network;

[0107] The information gain value test set is used to verify the model of the initialized decision tree regression network, and the initialized decision tree regression network with an error value less than or equal to a preset error threshold is output as a decision tree model for predicting information gain;

[0108] a4: Compare the information gain value of the j-th word group with the preset information gain value threshold. If the information gain value is less than or equal to the information gain value threshold, remove the j-th word group from the i-th preprocessed description document, set i=i+1, and return to step a2; if the information gain value is greater than the information gain value threshold, retain the j-th word group in the i-th preprocessed description document, set i=i+1, and return to step a2;

[0109] a5: Repeat steps a2 to a4 above until i=M, then end the loop and obtain the i-th updated description document, which contains N phrases, and set j=j+1, and return to step a2; N≤M, N∈N + , N + is a positive integer;

[0110] It can be understood that: compared with the pre-processed description document, the retained phrases in the updated description document can better highlight the functional attributes of the corresponding recruitment position, which is conducive to accurately obtaining the pre-processed description document, and further conducive to improving the accuracy of subsequent person-job matching;

[0111] a6: Repeat the above steps a2 to a5 until j=S, then end the loop and obtain S updated description documents.

[0112] In implementation, the filtering out of the common word set from the description document includes:

[0113] Get each phrase in the updated description document and calculate the DF value (i.e. document frequency) of each phrase;

[0114] It should be understood that the calculation formula of DF value (i.e. document frequency) is: ; Where: Indicates the number of documents containing the tth phrase; represents the document frequency of the t-th phrase;

[0115] Extract corresponding phrases whose DF values ​​are greater than or equal to a preset DF threshold, mark corresponding phrases whose DF values ​​are greater than or equal to the preset DF threshold as common words, obtain P common words, and merge the P common words into a common word set.

[0116] In implementation, the unique word set for each job post is screened out from the description document, including:

[0117] b1: Extract the updated description document corresponding to the g-th job position; and calculate the TF-IDF value of each phrase in the updated description document, g∈N + , N + is a positive integer;

[0118] It should be noted that the calculation formula of TF-IDF value is: ; Where: represents the TF-IDF value of the t-th phrase, represents the frequency of the t-th phrase in a single updated description document, represents the inverse document frequency of the t-th term; where, ; Where: Represents the total number of all updated description documents. represents the logarithmic function;

[0119] b2: Compare the TF-IDF value of each phrase with the preset TF-IDF threshold. If the TF-IDF value is greater than or equal to the preset TF-IDF threshold, mark the corresponding phrase as a unique word; if the TF-IDF value is greater than the preset TF-IDF threshold, set g=g+1 and return to step b1;

[0120] b3: Count all the unique words of the g-th job position, obtain Q unique words of the g-th job position, merge the Q unique words of the g-th job position into a unique word set, then obtain the unique word set of the g-th job position, set g=g+1, and return to step b1;

[0121] b4: Repeat the above steps b1 to b3 until all recruitment positions are screened (assuming that the number of all recruitment positions is Z, that is, g=Z), then end the loop and obtain the unique word set for each recruitment position.

[0122] The job screening module 102 is used to screen out a target job set from all the recruitment jobs based on the common word set; the target job set includes K target jobs with similar job functions, K∈N + , N + is a positive integer;

[0123] In practice, the target positions are selected from all recruitment positions, including:

[0124] Extract the updated description documents of all job positions and extract the number of occurrences of the common words in the updated description documents;

[0125] The number of occurrences of the common words is compared with the preset occurrence threshold. If the number of occurrences is greater than or equal to the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a target position with similar job functions; if the number of occurrences is less than the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a non-target position;

[0126] Count all corresponding recruitment positions whose number of appearances is greater than or equal to the preset number of appearances threshold, obtain K target positions with similar functions, and merge the K target positions with similar functions into a target position set;

[0127] For example, assume that a certain recruiting company has three recruiting positions, namely Y1, Y2 and Y3, and the description documents of the three recruiting positions are processed through the above content to obtain a common word set, and the common word set includes five common words, namely {"Master", "phython", "C++", "MATLAB", "SQL"}, and assume that four common words appear in the updated description documents of the recruitment positions Y1 and Y2, namely "Master", "phython", "C++" and "MATLAB", and assume that the preset occurrence quantity threshold is 3, therefore, the recruitment positions Y1 and Y2 are marked as target positions with similar job functions, and only one common word "Master" appears in the updated description document of the recruitment position Y3, then the recruitment position Y3 is not called the target position with similar job functions.

[0128] The matching judgment module 103 is used to retrieve the job seeker's resume to be matched and the keyword set in the resume to be matched, and extract the common word set and the unique word set of the target position, and input the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, to obtain a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer;

[0129] It should be noted that the keyword set in the resume to be matched is obtained by using a word segmentation tool (such as jieb word segmentation software) and removing meaningless characters;

[0130] For example, assuming that the text description of the resume to be matched is "graduated from XX University, bachelor's degree, capable of phython programming, good at using phython and SQL", assuming that after word segmentation processing, "graduated from / XX / University, bachelor's degree, capable of / phython / programming / , good at / using / phython / and / SQL" is obtained. According to the above processing, meaningless characters, neutral words or common words will be removed. Assuming that the final phrases only have "phython", "C++", "MATLAB" and "SQL", these three phrases are used as keywords in the resume to be matched, and the three keywords are combined into a keyword set;

[0131] Specifically, the training logic of the convolutional neural network used to determine whether the person-job match is consistent is as follows:

[0132] Acquire historical person-job matching training data, and divide the historical person-job matching training data into a person-job matching training set and a person-job matching test set, wherein the historical person-job matching training data includes person-job matching feature data and corresponding annotation labels;

[0133] The person-job matching feature data includes the keyword set in the resume to be matched, the common word set and the unique word set of the target position; the marking label includes "0" and "1", wherein "0" indicates that the resume to be matched is not compatible with the target position, and "1" indicates that the resume to be matched is compatible with the target position;

[0134] The logic for obtaining the person-job matching feature data in the historical person-job matching training data is described in the above related content, which will not be repeated here.

[0135] The logic for generating the annotation labels in the historical person-job matching training data is as follows:

[0136] Calculate the similarity between the keyword set of the resume to be matched and the common word set of the target position to obtain a first similarity;

[0137] Calculate the similarity between the keyword set of the resume to be matched and the unique word set of the target position to obtain a second similarity;

[0138] It should be noted that: the first similarity and the second similarity are both calculated using the Jaccard similarity algorithm, and the expression of the Jaccard similarity algorithm is: ; Where: = or , represents the keyword set of the resume to be matched in vectorized form, Represents the common word set or unique word set of the target position in vectorized form;

[0139] Substituting the first similarity and the second similarity into a pre-built mathematical calculation model to obtain a matching coefficient between the resume to be matched and the target position;

[0140] The mathematical calculation model is expressed as follows:

[0141] ;

[0142] Where: Indicates the matching coefficient between the resume to be matched and the target position. represents the first similarity, represents the second similarity, and represents a weight factor greater than zero, ;

[0143] The matching coefficient is compared with the preset matching coefficient threshold. If the matching coefficient is greater than or equal to the matching coefficient threshold, it is determined that the resume to be matched is compatible with the corresponding target position; if the matching coefficient is less than the matching coefficient threshold, it is determined that the resume to be matched is not compatible with the corresponding target position.

[0144] Construct a convolutional neural network, use the job matching feature data in the job matching training set as the input of the convolutional neural network, and use the labeled labels in the job matching training set as the output of the convolutional neural network, train the convolutional neural network, and obtain an initial convolutional neural network;

[0145] The initial convolutional neural network model is verified using the person-job matching test set, and an initial convolutional neural network with a preset test accuracy is output as the convolutional neural network used to determine whether the person-job matching is consistent.

[0146] The person-job matching module 104 is used to summarize all job sets that are suitable for the job seeker from the target job set according to the judgment result; the job set contains R matching jobs that are suitable for the job seeker, R∈N + , N + is a positive integer;

[0147] In practice, the above-mentioned process of summarizing all the positions that are suitable for the job seeker from the target positions includes:

[0148] Collect all target positions that match the resumes to be matched in the target position collection;

[0149] Mark all target positions that match the resume to be matched, and the matching positions that match the job seeker, and get R matching positions;

[0150] It can be understood that, through the common word set and the unique word set of each recruitment position, the present invention can dig out the commonalities and characteristics of different recruitment positions in the same enterprise; thus, it is convenient for recruiters to understand which recruitment positions the same job seeker can adapt to in the same enterprise, so that they can conduct interviews and communications for multiple positions at the same time for the same job seeker, which greatly reduces the number of repeated screening and interviews by recruiters. For example, there are usually multiple recruiters in an enterprise, and each is responsible for a different recruitment position. Suppose recruiter A communicates or interviews with job seeker W1 for recruitment position Y1, and finally finds that he is not suitable for recruitment position Y1, but because recruitment position Y1 is different from recruitment position Y1, Due to the functional similarity of Position Y2, Recruiter B communicated or interviewed Job Applicant W1 again for Position Y2. Even if the second interview found that Job Applicant W1 was qualified for Position Y2, this is undoubtedly an inefficient recruitment behavior. If the second interview is still not satisfactory, another communication or interview is conducted for other positions. This will undoubtedly aggravate the waste of recruitment resources, resulting in extremely low recruitment efficiency and long recruitment cycle. If this happens when the company is short of labor, this will undoubtedly bring significant losses to the company. Therefore, compared with the existing technology, this case can greatly reduce the occurrence of repeated screening and interviews, which is conducive to improving the accuracy and timeliness of corporate recruitment.

[0151] Example 2

[0152] See also Figure 2 As shown, this embodiment discloses a precise matching recruitment method based on big data analysis, and the method includes:

[0153] S201: Load the description documents of all the recruitment positions of the enterprise, and use the TF-IDF method to filter out the common word set and the unique word set of each recruitment position from the description documents; the common word set contains P common words, and the unique word set contains Q unique words, P, Q∈N + , N + is a positive integer;

[0154] S202: Filter out a target position set from all recruitment positions based on the common word set; the target position set includes K target positions with similar job functions, K∈N + , N + is a positive integer;

[0155] S203: Retrieving the job seeker's resume to be matched and the keyword set in the resume to be matched, and extracting the common word set and the unique word set of the target position, inputting the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, and obtaining a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer;

[0156] S204: According to the judgment result, a set of positions that are suitable for the job seeker is summarized from the target position set; the set of positions contains R matching positions that are suitable for the job seeker, R∈N + , N + Is a positive integer.

[0157] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

[0158] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired network or a wireless network. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD) or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0159] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0160] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0161] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0162] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0164] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0165] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A precise matching recruitment system based on big data analysis, characterized in that: The system comprises: The feature screening module is used to load the description documents of all recruitment positions of the enterprise, and use the TF-IDF method to screen out the common word set and the unique word set of each recruitment position from the description documents; the common word set contains P common words, and the unique word set contains Q unique words, P, Q∈N + , N + is a positive integer; The step of filtering out a common word set from the description document includes: Get each phrase in the updated description document and calculate the DF value of each phrase; Extract corresponding phrases whose DF values ​​are greater than or equal to a preset DF threshold, mark corresponding phrases whose DF values ​​are greater than or equal to the preset DF threshold as common words, obtain P common words, and merge the P common words into a common word set; The method of filtering out a unique word set for each job post from the description document includes: b1: Extract the updated description document corresponding to the g-th job position; and calculate the TF-IDF value of each phrase in the updated description document, g∈N + , N + is a positive integer; b2: Compare the TF-IDF value of each phrase with the preset TF-IDF threshold. If the TF-IDF value is greater than or equal to the preset TF-IDF threshold, mark the corresponding phrase as a unique word; if the TF-IDF value is greater than the preset TF-IDF threshold, set g=g+1 and return to step b1; b3: Count all the unique words of the g-th job position, obtain Q unique words of the g-th job position, merge the Q unique words of the g-th job position into a unique word set, then obtain the unique word set of the g-th job position, set g=g+1, and return to step b1; b4: Repeat steps b1 to b3 until all recruitment positions are screened, then end the loop and obtain a unique word set for each recruitment position; The job screening module is used to screen out a target job set from all recruitment jobs based on the common word set; the target job set contains K target jobs with similar job functions, K∈N + , N + is a positive integer; The target job set is selected from all recruitment positions, including: Extract the updated description documents of all job positions and extract the number of occurrences of the common words in the updated description documents; The number of occurrences of the common words is compared with the preset occurrence threshold. If the number of occurrences is greater than or equal to the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a target position with similar job functions; if the number of occurrences is less than the preset occurrence threshold, the recruitment position corresponding to the updated description document is marked as a non-target position; Count all corresponding recruitment positions whose appearance number is greater than or equal to the preset appearance number threshold, obtain K target positions with similar functions, and merge the K target positions with similar functions into a target position set; The matching judgment module is used to retrieve the job seeker's resume to be matched and the keyword set in the resume to be matched, and extract the common word set and unique word set of the target position, and input the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, to obtain a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer; The training logic of the convolutional neural network used to determine whether the person-job match is consistent is as follows: Acquire historical person-job matching training data, and divide the historical person-job matching training data into a person-job matching training set and a person-job matching test set, wherein the historical person-job matching training data includes person-job matching feature data and corresponding annotation labels; The person-job matching feature data includes a keyword set in the resume to be matched, a common word set and a unique word set of the target position; the marking label includes "0" and "1", wherein "0" indicates that the resume to be matched is not compatible with the target position, and "1" indicates that the resume to be matched is compatible with the target position; The logic for generating the annotation labels in the historical person-job matching training data is as follows: Calculate the similarity between the keyword set of the resume to be matched and the common word set of the target position to obtain a first similarity; Calculate the similarity between the keyword set of the resume to be matched and the unique word set of the target position to obtain a second similarity; Substituting the first similarity and the second similarity into a pre-built mathematical calculation model to obtain a matching coefficient between the resume to be matched and the target position; The mathematical calculation model is expressed as follows: ； Where: Indicates the matching coefficient between the resume to be matched and the target position. represents the first similarity, represents the second similarity, and represents a weight factor greater than zero, ; The matching coefficient is compared with the preset matching coefficient threshold. If the matching coefficient is greater than or equal to the matching coefficient threshold, it is determined that the resume to be matched is compatible with the corresponding target position; if the matching coefficient is less than the matching coefficient threshold, it is determined that the resume to be matched is not compatible with the corresponding target position; Construct a convolutional neural network, use the job matching feature data in the job matching training set as the input of the convolutional neural network, and use the labeled labels in the job matching training set as the output of the convolutional neural network, train the convolutional neural network, and obtain an initial convolutional neural network; Use the person-job matching test set to verify the model of the initial convolutional neural network, and output an initial convolutional neural network with a test accuracy greater than or equal to the preset test accuracy as the convolutional neural network used to determine whether the person-job matching is consistent; The person-job matching module is used to summarize all job sets that are suitable for job seekers from the target job set according to the judgment results; the job set contains R matching jobs that are suitable for job seekers, R∈N + , N + Is a positive integer.

2. The accurate matching recruitment system based on big data analysis according to claim 1 is characterized in that: Before screening out the common and unique word sets for each job position, including: a1: Segment the description documents of all job positions and remove meaningless characters to obtain S preprocessed description documents. Each preprocessed description document includes M phrases, S, M∈N + , N + is a positive integer; a2: Extract the jth phrase in the i-th preprocessed description document, i, j∈N + , N + is a positive integer; a3: Input the j-th phrase into the pre-trained decision tree model for predicting information gain to obtain the information gain value of the j-th phrase; a4: Compare the information gain value of the j-th word group with the preset information gain value threshold. If the information gain value is less than or equal to the information gain value threshold, remove the j-th word group from the i-th preprocessed description document, set i=i+1, and return to step a2; if the information gain value is greater than the information gain value threshold, retain the j-th word group in the i-th preprocessed description document, set i=i+1, and return to step a2; a5: Repeat steps a2 to a4 above until i=M, then end the loop and obtain the i-th updated description document, which contains N phrases, and set j=j+1, and return to step a2; N≤M, N∈N + , N + is a positive integer; a6: Repeat the above steps a2 to a5 until j=S, then end the loop and obtain S updated description documents.

3. The accurate matching recruitment system based on big data analysis according to claim 2 is characterized in that: The training logic of the decision tree model for predicting information gain is as follows: Acquire historical information gain value training data, and divide the historical information gain value training data into an information gain value training set and an information gain value test set; the historical information gain value training data includes a plurality of phrases in vector form and their corresponding information gain values; Construct a decision tree network, use the phrases in the form of vectors in the information gain value training set as the input of the decision tree network, and use the information gain values ​​in the information gain value training set as the output of the decision tree network, train the decision tree network, and obtain an initialized decision tree regression network; The information gain value test set is used to verify the model of the initialized decision tree regression network, and the initialized decision tree regression network with an error value less than or equal to the preset error threshold is output as the decision tree model for predicting information gain.

4. The accurate matching recruitment system based on big data analysis according to claim 3 is characterized in that: The information gain value in the historical information gain value training data is obtained by processing using an information gain measurement method; The processing logic of the information gain measurement method is as follows: Obtain historical job matching sample data, the historical job matching sample data includes U job matching samples, each job matching sample includes multiple feature words and their corresponding classification labels, the classification labels include matching and non-matching, U∈N + , N + is a positive integer; The initial entropy is calculated based on U job matching samples. The specific calculation formula is as follows: ； Where: represents the initial entropy, Indicates the number of categories of classification labels, Represents the probability of the rth category classification label; Calculate the conditional entropy of the feature word, the specific calculation formula is as follows: ； Where: represents the conditional entropy, Indicates that it contains characteristic words The number of possible values ​​of Indicates that it contains characteristic words The value is The conditional probability of Indicates that it contains characteristic words The value is , the entropy of the classification label; Calculate the information gain value, the specific calculation formula is as follows: ； Where: Represents the information gain value.

5. The accurate matching recruitment system based on big data analysis according to claim 4 is characterized in that: The above-mentioned process of summarizing all the positions that are suitable for the job seeker from the target positions includes: Collect all target positions that match the resumes to be matched in the target position collection; All target positions that match the resumes to be matched are marked as matching positions that match the job seekers, and R matching positions are obtained.

6. A precise matching recruitment method based on big data analysis, characterized in that: The method is implemented based on the accurate matching recruitment system based on big data analysis as described in any one of claims 1 to 5, and the method comprises: Load the description documents of all the recruitment positions of the enterprise, and use the TF-IDF method to filter out the common word set and the unique word set of each recruitment position from the description documents; the common word set contains P common words, and the unique word set contains Q unique words, P, Q∈N + , N + is a positive integer; The target position set is selected from all recruitment positions based on the common word set; the target position set contains K target positions with similar job functions, K∈N + , N + is a positive integer; Retrieving the job seeker's resume to be matched and the keyword set in the resume to be matched, as well as extracting the common word set and the unique word set of the target position, inputting the keyword set, the common word set and the unique word set into a convolutional neural network pre-trained for judging whether the person-job match is consistent, and obtaining a judgment result; the judgment result includes one of the following: the resume to be matched is not compatible with the target position, or the resume to be matched is compatible with the target position; the keyword set includes L keywords, L∈N + , N + is a positive integer; According to the judgment result, a set of all positions that are suitable for the job seeker is summarized from the target position set; the position set contains R matching positions that are suitable for the job seeker, R∈N + , N + Is a positive integer.

Images