TW202143122A - Resume scoring method and system wherein the resume scoring system includes a plurality of regular expressions, a vector generation model, and a scoring model - Google Patents

Abstract

A resume scoring system includes a plurality of regular expressions, a vector generation model, and a scoring model. Each of the regular expressions has a predetermined keyword. The vector generation model is used to generate a text vector according to the texual content. The scoring model generates a score according to the results generated by the regular expressions and the text vector generated by the vector generation model. When receiving a resume related to a job applicant, for each regular expression, the system uses the regular expression based on the resume to obtain a keyword combination that includes the keyword content of the predetermined keyword corresponding to the regular expression in the resume, uses the vector generation model to generate a text vector corresponding to the resume, and uses the scoring model to generate a score based on each keyword combination and the text vector.

Description

Resume scoring method and system

The present invention relates to an office automation method, in particular to a method for automatically generating scores based on resumes.

In today's society, when general companies conduct job recruitment, the first pass will be based on the resume provided by job applicants. However, for talent recruitment companies, when they receive tens of thousands of resumes, they often The human resources department reviews the content of each resume one by one to select suitable job applicants. This method is not only laborious, but also may cause the problem of erroneous selection of resumes due to different opinions or human errors.

In view of this, it is bound to be necessary to propose a new solution to solve the current problems of excessive time-consuming and labor-intensive review of resumes and the possibility of misselection of resumes.

Therefore, the purpose of the present invention is to provide a resume scoring method that assists in evaluating resumes.

In addition, another object of the present invention is to provide a history scoring system that assists in evaluating history.

Therefore, the resume scoring method of the present invention is implemented by a server. The resume includes a content of academic experience filled in by a job applicant and an autobiographical content introduced by a person. The server includes multiple formal expressions, a vector generation model, and A scoring model, each of the regular expressions has a predetermined keyword, the vector generation model is used to generate a text vector based on a content composed of words, the scoring model is based on the result of the regular expression and the vector generation model The generated text vector generates a score. The resume scoring method includes one step (A), one step (B), and one step (C).

In this step (A), when the server receives the resume related to the job applicant, for each formal expression, the server uses the formal expression to obtain the content of the academic experience according to the content of the academic experience The keyword content corresponding to the predetermined keyword of the regular expression, wherein the predetermined keyword of each regular expression and its corresponding keyword content constitute a corresponding keyword combination.

In this step (B), the server uses the vector generation model to generate a text vector corresponding to the autobiographical content based on the autobiographical content.

In this step (C), the server uses the scoring model to generate a score corresponding to the history according to each keyword combination and the text vector.

Furthermore, the resume scoring system of the present invention is used to generate a score on a resume and connect it to a management terminal via a communication network. The resume includes the content of an academic experience filled in by a job applicant and the autobiographical content introduced by a person. The system includes a communication module, a storage module, and a processing module.

The communication module is connected to the communication network, the storage module stores a plurality of normal expressions, a vector generation model, and a scoring model, each of the normal expressions has a predetermined keyword, and the vector generation model is used for A text vector is generated based on a content composed of words, and the scoring model generates a score based on the result of the regular expression and the text vector generated by the vector generation model.

The processing module is electrically connected to the communication module and the storage module, wherein, when the processing module receives the resume related to the job applicant from the management terminal through the communication module, for each regular expression According to the content of the learning experience, the processing module uses the formal expression to obtain the keyword content corresponding to the predetermined keyword of the formal expression in the learning experience content, wherein the predetermined keyword of each formal expression and its The corresponding keyword content constitutes a corresponding keyword combination, and according to the autobiographical content, the vector generation model is used to generate a text vector corresponding to the autobiographical content, and the scoring model is used according to each keyword combination and the text vector Generate a score corresponding to the resume.

The effect of the present invention is that the server generates scores corresponding to the resumes, and the relevant staff can refer to the scores to screen job applicants, which not only saves the time and labor cost of reviewing each resume, but also avoids the problem of misselecting resumes caused by human factors. .

Before the present invention is described in detail, it should be noted that in the following description, similar elements are denoted by the same numbers.

Referring to FIG. 1, a first embodiment of the history scoring system of the present invention is implemented by a server 1. The server 1 is connected to a management terminal 2 through a communication network 100, and includes a communication module 11 and a storage Module 12, and a processing module 13.

The communication module 11 is connected to the management terminal 2 through the communication network 100.

The storage module 12 stores multiple regular expressions, multiple historical histories, a vector generation model, and a scoring model. Each regular expression has a predetermined keyword, and each historical record has a correlation with a scoring model. The autobiographical content introduced by the historical job seeker, multiple keyword combinations related to the historical job seeker’s learning experience, a text vector related to the autobiographical content, and a score related to the historical resume. The vector generation model is used A text vector is generated based on a content composed of words, and the scoring model generates a score based on the result of the regular expression and the text vector generated by the vector generation model.

The processing module 13 is electrically connected to the communication module 11 and the storage module 12, and generates a score corresponding to a resume based on a resume related to a job seeker, wherein the resume contains the content of an academic experience filled in by the job seeker , And an autobiographical content related to the job applicant’s personal introduction.

Referring to FIG. 2, FIG. 3, and FIG. 4, the history scoring method of the present invention includes a vector generation model training program, a scoring model training program, and a scoring program.

Referring to Figure 2, the vector generation model training program includes a step 301, a step 302, a step 303, a step 304, and a step 305, and illustrates how the processing module 13 adjusts and refines the vector generation based on the historical histories Model.

In step 301, the processing module 13 uses a deep learning algorithm based on the autobiographical content and text vectors of the historical histories to establish a second training model for generating a text vector based on a content composed of text , Such as training models such as BERT or XLNet.

In this step 302, the processing module 13 uses the second training model to generate a plurality of training text vectors corresponding to each historical record according to the autobiographical content corresponding to each historical record.

In step 303, for each history history, the processing module 13 determines whether the similarity between the text vector and the training text vector corresponding to the history history is greater than a preset threshold. In the first embodiment, the processing module 13 obtains the similarity between the text vector of the historical history and the training text vector according to a similarity comparison algorithm, such as cosine similarity. When the processing module 13 determines that the degree of similarity is not greater than the preset threshold, the processing module 13 adjusts the second training model and re-executes the step 302, that is, the step 304. When the processing module 13 determines When the similarity is greater than the preset threshold, the processing module 13 uses the second training model as the vector generation model for generating a text vector based on a content composed of text, that is, step 305.

Referring to FIG. 3, the scoring model training procedure includes a step 401, a step 402, a step 403, a step 404, and a step 405, and illustrates how the processing module 13 adjusts and refines the scoring model according to the historical histories.

In step 401, the processing module 13 divides the stored history history into a training subset and a test subset, wherein the history history included in the training subset and the test subset included The number of these historical histories can be equal or different.

In step 402, the processing module 13 uses a machine learning algorithm, such as logistic regression (Logistic Regression), according to the keyword combination, the text vector, and the score corresponding to each historical record in the training subset. ), Random Forest, Gradient Boosting, Artificial Neural Network, etc., to establish a first training model that generates a training score based on the keyword combinations and the text vector .

In step 403, the processing module 13 determines whether the first training model is over-engineered according to the keyword combination, the text vector, and the score corresponding to each historical record in the training subset and the test subset. When it is judged that the first training model is over-fitted or under-fitted, the processing module 13 adjusts the first training model and re-determines whether the adjusted first training model is over-fitted or under-fitted. Insufficient synthesis, that is, in step 404, when it is determined that the first training model is not over-fitted or under-fitted, the processing module 13 uses the first training model as the result generated according to the regular expression and the vector The text vector generated by the generative model generates a score of the scoring model, that is, step 405.

Referring to FIG. 4, the scoring procedure includes a step 501, a step 502, a step 503, and a step 504, and illustrates how the processing module 13 generates the score corresponding to the history according to the history.

In step 501, when the processing module 13 receives the resume from the management terminal 2 and related to the job applicant through the communication module 11, for each regular expression, the processing module 13 according to The academic experience content of the resume uses the formal expression to obtain the keyword content corresponding to the predetermined keyword of the formal expression in the academic experience content, wherein the predetermined keyword of each formal expression and its corresponding keyword content Form a corresponding keyword combination. In the first embodiment, the predetermined keywords of the formal expressions are "highest education", "work experience", and "English ability" respectively, and the processing module 13 obtains the key according to the formal expressions. The content of the words correspond to "Master of Electrical Engineering of Chenggong University", "3 years of TSMC R&D Department", and "870 points on the TOEIC test", and "Highest Education: Master of Electrical Engineering of Chenggong University" is a keyword corresponding to the "highest degree". Similarly, the keyword combinations corresponding to "work experience" and "English ability" are "work experience: 3 years in TSMC R&D department" and "English ability: TOEIC test 870 points" respectively.

In step 502, the processing module 13 uses the vector generation model to generate a text vector corresponding to the autobiographical content based on the autobiographical content of the resume. It is worth mentioning that the vector generation model may be the vector generation model confirmed in step 305.

In step 503, the processing module 13 uses the scoring model to generate a score corresponding to the history according to each keyword combination and the text vector. It is worth mentioning that the scoring model may be the scoring model confirmed in step 405.

In the step 504, the processing module 13 stores the autobiographical content of the history, each keyword combination, the text vector, and the score is one of the history history. In this way, the number of historical histories can be accumulated to make the training samples of the scoring model and the vector generation model more diverse, and the obtained scoring model and the vector can be refined by continuously tracking and adjusting the training and testing samples Generate the model.

Referring to FIG. 5, a second embodiment of the resume scoring method of the present invention is similar to the first embodiment, and the similarities are not repeated here. The difference is that in the second embodiment, the storage module 12 also stores a summary generation model for generating a summary based on a content composed of text, and each history history stored in the storage module 12 also has a history related to the autobiographical content of the corresponding history history Abstract, and before step 504, a summary program is also included, which explains how the processing module 13 generates a summary corresponding to the autobiographical content of the resume based on the autobiographical content of the resume, and includes a step 601, a step 602, and a step 602. Step 603, a step 604, a step 605, and a step 606.

In this step 601, the processing module 13 uses a deep learning algorithm, such as a recurrent neural network, based on the autobiographical content and the abstracts of the historical histories to create a text-based content generation An abstract training model, such as GPT-2 or Transformer.

In step 602, the processing module 13 uses the training model to generate a plurality of training summaries corresponding to each historical record according to the autobiographical content corresponding to each historical record.

In this step 603, for each history history, the processing module 13 determines whether the similarity between the summary and the training summary corresponding to the history history is greater than another preset threshold. In this embodiment, the processing module 13 uses a similarity comparison algorithm, such as cosine similarity, to obtain the similarity between the summary of the historical history and the training summary. When it is judged that the similarity is not greater than the another preset threshold, the processing module 13 adjusts the training model and re-executes the step 602, that is, the step 604, when it is judged that the similarity is greater than the another preset threshold At this time, the processing module 13 determines that the training model is a summary generation model for generating a summary based on a content composed of text, that is, step 605.

In step 606, when the processing module 13 receives the resume related to the job applicant from the management terminal 2 through the communication module 11, the processing module 13 uses the autobiographical content of the resume The abstract generation model generates a pair of abstracts that should be autobiographical content.

In addition, in the second embodiment, in the step 504, the processing module 13 stores the autobiographical content of the history, each keyword combination, the text vector, the score, and the summary as the history history One of them.

In summary, the history scoring method of the present invention uses the processing module 13 to generate scores corresponding to the history based on the regular expressions, the vector generation model, and the scoring model, thereby saving relevant staff The time and labor cost for evaluating and screening each resume, and because the processing module generates the score according to the scoring model, it also avoids problems caused by different opinions or human errors, such as resumes. In addition, the processing module 13 generates a summary corresponding to the autobiographical content according to the summary generation model, so that when the reviewer is interested in the history of certain scores, he can use the The summary quickly learns about the job applicant, and saves the time and labor cost of reading the entire resume, so it can indeed achieve the purpose of the invention.

However, the above are only examples of the present invention. When the scope of implementation of the present invention cannot be limited by this, all simple equivalent changes and modifications made in accordance with the scope of the patent application of the present invention and the content of the patent specification still belong to Within the scope covered by the patent of the present invention.

1: Server 100: Communication network 11: Communication module 12: Storage module 13: Processing module 2: Management side 301~305: steps 401~405: Steps 501~504: Steps 601~606: steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, in which: Figure 1 is a block diagram illustrating a first embodiment of the resume scoring system of the present invention connected to a management terminal via a communication network; Figure 2 is a flowchart illustrating a vector generation model training program executed by the first embodiment of the resume scoring method of the present invention; Fig. 3 is a flowchart illustrating a scoring model training procedure performed by the first embodiment of the resume scoring method of the present invention; Fig. 4 is a flowchart illustrating a scoring procedure of the resume scoring method of the present invention executed by the first embodiment; and FIG. 5 is a flowchart illustrating a summary procedure of the resume scoring method of the present invention executed by a second embodiment.

1: Server

100: Communication network

11: Communication module

12: Storage module

13: Processing module

2: Management side

Claims (12)

A resume scoring method implemented by a server. The resume includes content of an academic experience filled in by a job applicant and autobiographical content introduced by a person. The server includes multiple formal expressions, a vector generation model, and a scoring model , Each regular expression has a predetermined keyword, the vector generation model is used to generate a text vector based on a content composed of text, and the scoring model is based on the result of the regular expression and the text generated by the vector generation model The vector generates a score. The resume scoring method includes the following steps: (A) When the server receives the resume related to the job applicant, for each formal expression, the server uses the formal expression according to the academic experience content to obtain the academic experience content corresponding to the formal expression The keyword content of the predetermined keyword of the formula, wherein the predetermined keyword of each regular expression and its corresponding keyword content constitute a corresponding keyword combination; (B) The server uses the vector generation model to generate a pair of text vectors corresponding to the autobiographical content based on the autobiographical content; and (C) The server uses the scoring model to generate a score corresponding to the resume based on each keyword combination and the text vector. The resume scoring method as described in claim 1, further includes the following steps: (D) The server stores the autobiographical content of the history, each keyword combination, the text vector, and the score into a history history. The resume scoring method described in claim 2 further includes the following steps: (E) The server divides the stored history history into a training subset and a test subset; (F) According to the keyword combination, the text vector, and the score corresponding to each historical record in the training subset, the server uses a machine learning algorithm to create a generation based on the keyword combination and the text vector The first training model of a training score; (G) The server judges whether the first training model is overfitted or underfitted according to the keyword combination, the text vector, and the score corresponding to each historical record in the training subset and the test subset ； (H) When the server determines that the first training model is overfitted or underfitted, the server adjusts the first training model and performs step (G) again; and (I) When the server determines that the first training model is not overfitted or underfitted, the server uses the first training model as the scoring model when performing the step (C). The resume scoring method described in claim 2 further includes the following steps: (J) The server uses a deep learning algorithm based on the autobiographical content and text vectors of multiple historical histories to establish a second training model for generating a text vector based on a content composed of text; (K) The server uses the second training model to generate multiple training text vectors corresponding to each historical record according to the autobiographical content corresponding to each historical record; (L) For each historical record, the server determines whether the similarity between the text vector and the training text vector corresponding to the historical record is greater than a preset threshold; (M) When the server side judges that the similarity is not greater than the preset threshold, the server side adjusts the second training model and repeats the step (K); and (N) When the server side judges that the similarity is greater than the preset threshold, the server side uses the second training model as the vector generation model when performing the step (B). The resume scoring method according to claim 1, wherein the server further includes a summary generation model for generating a summary based on a content composed of text, and after the step (C), the following is further included step: (O) According to the autobiographical content, the server uses the abstract generation model to generate a summary of the autobiographical content. According to the history scoring method described in claim 5, the server further includes a plurality of history history, each history history includes an autobiographical content and a summary of the corresponding history history, wherein, before step (O), return It includes the following steps: (P) The server uses a deep learning algorithm based on the autobiographical content and the abstracts of the historical histories to establish a training model that generates an abstract based on a content composed of text; (Q) The server uses the training model to generate multiple training summaries corresponding to each historical record according to the autobiographical content corresponding to each historical record; (R) For each historical record, the server judges whether the similarity between the summary and the training summary corresponding to the historical record is greater than another preset threshold; (S) When the server side judges that the similarity is not greater than the other preset threshold, the server side adjusts the training model and repeats the step (Q); and (T) When the server determines that the similarity is greater than the other preset threshold, the server determines that the training model is the summary generation model. A resume scoring system for generating a score on a resume and connecting it to a management terminal via a communication network. The resume includes a content of academic experience filled in by a job applicant and an autobiographical content introduced by a person. The resume scoring system includes: A communication module, connected to the communication network; A storage module stores a plurality of regular expressions, a vector generation model, and a scoring model. Each regular expression has a predetermined keyword. The vector generation model is used to generate a A text vector, where the scoring model generates a score based on the result generated by the regular expression and the text vector generated by the vector generation model; and A processing module, electrically connected to the communication module and the storage module; Wherein, when the processing module receives the resume related to the job applicant from the management terminal through the communication module, for each formal expression, the processing module uses the formal expression according to the content of the academic experience Obtain the keyword content corresponding to the predetermined keyword of the regular expression in the learning experience content, wherein the predetermined keyword of each regular expression and its corresponding keyword content constitute a corresponding keyword combination, and according to the autobiography For content, use the vector generation model to generate a text vector corresponding to the autobiographical content, and use the scoring model to generate a score corresponding to the resume based on each keyword combination and the text vector. The history scoring system according to claim 7, wherein the processing module stores the autobiographical content of the history, each keyword combination, the text vector, and the score in the storage module to form a history history. The history scoring system according to claim 8, wherein the processing module divides the stored multiple history history into a training subset and a test subset by the storage module, and according to each of the training subsets The keyword combination, the text vector, and the score corresponding to the historical resume are used to establish a first training model that generates a training score based on the keyword combination and the text vector using a machine learning algorithm, and based on the training The keyword combination, the text vector, and the score corresponding to each historical resume in the subset and the test subset are used to determine whether the first training model is overfitted or underfitted. When it is judged that the first training model is overfitting When fitting or underfitting, adjust the first training model and re-judge whether it is overfitting or underfitting, and when it is judged that the first training model is not overfitting or underfitting, the first training model The model serves as the scoring model. The history scoring system according to claim 8, wherein the processing module uses a deep learning algorithm to establish a system based on the autobiographical content and text vectors of multiple history history stored in the storage module. The constituted content generates a second training model of text vectors, and according to the autobiographical content corresponding to each history history, the second training model is used to generate multiple training text vectors corresponding to each history history. For each history history , The processing module determines whether the similarity between the text vector and the training text vector corresponding to the historical record is greater than a preset threshold, and when it is determined that the similarity is not greater than the preset threshold, adjusts the second training model The training text vectors are regenerated and the similarity is judged. When the similarity is judged to be greater than the preset threshold, the second training model is used as the vector generation model. The resume scoring system according to claim 7, wherein the storage module further stores a summary generation model, the summary generation model is used to generate a summary based on a content composed of text, and the processing module is based on the autobiographical content, Use this abstract generation model to generate a pair of autobiographical content abstracts. The history scoring system according to claim 11, wherein the storage module also stores a plurality of history history, each history history includes an autobiographical content and a summary of the corresponding history history, and the processing module is based on the After the autobiographical content and the abstracts, a deep learning algorithm is used to establish a training model that generates an abstract based on a content composed of text, and according to the autobiographical content corresponding to each historical resume, the training model is used to generate a corresponding A plurality of training summaries of a history history, for each history history, the processing module determines whether the similarity of the summary corresponding to the history history and the training summary is greater than another preset threshold, when it is judged that the similarity is not greater than When the other preset threshold value, the training model is adjusted and the training summaries are regenerated and the similarity is judged. When the judgement similarity is greater than the other preset threshold value, the training model is used as the summary generation model.

Images