CN115526589B - Employment information processing method and equipment for crowd to be employment - Google Patents

Employment information processing method and equipment for crowd to be employment Download PDF

Info

Publication number
CN115526589B
CN115526589B CN202211128537.7A CN202211128537A CN115526589B CN 115526589 B CN115526589 B CN 115526589B CN 202211128537 A CN202211128537 A CN 202211128537A CN 115526589 B CN115526589 B CN 115526589B
Authority
CN
China
Prior art keywords
information
post
evaluation score
text
determining
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202211128537.7A
Other languages
Chinese (zh)
Other versions
CN115526589A (en
Inventor
单晓燕
郭志伟
郭宁
杨洋
刘科坊
王霆
宋纪宾
王丽雅
马海迪
金锋
王杲卿
张阿龙
刘传浩
李鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Talent Development Group Information Technology Co ltd
Original Assignee
Shandong Talent Development Group Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Talent Development Group Information Technology Co ltd filed Critical Shandong Talent Development Group Information Technology Co ltd
Priority to CN202211128537.7A priority Critical patent/CN115526589B/en
Publication of CN115526589A publication Critical patent/CN115526589A/en
Application granted granted Critical
Publication of CN115526589B publication Critical patent/CN115526589B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/10Office automation; Time management
    • G06Q10/105Human resources
    • G06Q10/1053Employment or hiring
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063Operations research, analysis or management
    • G06Q10/0639Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393Score-carding, benchmarking or key performance indicator [KPI] analysis

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application provides employment information processing method and equipment for a crowd to be employment, wherein the method acquires a plurality of post information; the post information at least comprises: post name, post responsibility text, post demand text. And determining a first evaluation score corresponding to each position information based on the recruiter information corresponding to each position information and resume information of the currently pending person. The first evaluation score represents the matching degree of the currently undetermined person and the post information. And determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model. And determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model. And determining at least one post information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the post information to the corresponding user terminal.

Description

Employment information processing method and equipment for crowd to be employment
Technical Field
The application relates to the technical field of Internet, in particular to employment information processing method and equipment for a crowd waiting for employment.
Background
Along with the development of science and technology and the progress of society, the enterprises need to be equipped with proper talents to improve the industrial benefit. Even under the current age background that big data technology is developed rapidly, the matching between enterprises and talents is also that information is disjointed, for example, the enterprises cannot truly know talent ability, the talents cannot accurately obtain the requirements of the enterprise posts, and whether the enterprises are matched with own ability or not is determined in time. Under the conditions that current recruitment means are limited, graduates are increased dramatically year by year and young staff leave the staff rate high, enterprises only continuously recruit talents through the traditional means, and the time and the labor are very time-consuming and labor-consuming.
In addition, in the crowd to be employment, there are employment staff with working experience and technical ability, and the staff can not find proper work in a short time by means of own knowledge of posts and application means mastered by the staff. Moreover, recruitment information sent by different enterprises is difficult to understand, and a employment waiting person cannot accurately deliver a resume to a proper post, so that normal employment and long-term occupation planning of the employment waiting person are affected.
And the manual work is carried out to identify and screen the resume of the employment staff, and then the recruitment requirement of the enterprise is matched, so that the work is complex, the workload is large, and the labor cost is wasted. Therefore, a method for processing, identifying and screening employment information of employment staff is needed for reference by people to solve most of manpower cost.
Disclosure of Invention
The embodiment of the application provides employment information processing method and equipment for a crowd to be employment, which solve the problems that employment information is processed, identified and screened mainly by manpower, is complex in work, large in workload and wastes labor cost.
In one aspect, an embodiment of the present application provides a employment information processing method for a crowd to be employment, where the method includes:
and acquiring a plurality of post information. The post information at least comprises: post name, post responsibility text, post demand text. And determining a first evaluation score corresponding to each position information based on the recruiter information corresponding to each position information and resume information of the currently pending person. The first evaluation score represents the matching degree of the currently undetermined person and the post information. And determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model. And determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model. And determining at least one post information matched with the current undetermined person based on the second evaluation score and the matching probability value, so as to send the post information to the corresponding user terminal. The corresponding text is an evaluation text for the post. And generating a similarity matrix of the current undetermined person and each recruiter according to each cosine similarity. And determining a corresponding post score of the current undetermined person as a first evaluation score based on the evaluation text of each recruiter for each post and the corresponding cosine similarity in the similarity matrix.
In one implementation of the present application, the resume information of the current waiting person and the evaluation text corresponding to the first evaluation score are input into the lingo-meaning model, so that the lingo-meaning model determines the corresponding scoring matrix of the current waiting person and the recruiter corresponding to the first evaluation score. And performing factorization processing on the scoring matrix through the latent meaning model and a preset loss function until the preset loss function meets preset conditions so as to decompose the scoring matrix into a user scoring matrix and a post weight matrix. And determining the predictive evaluation scores of the currently undetermined personnel on each post based on the user scoring matrix and the post weight matrix. And taking the weighted average of the first evaluation score and the predicted evaluation score as a second evaluation score.
In one implementation of the present application, a plurality of resume samples and a plurality of post information samples are obtained. And respectively carrying out short text coding processing on the resume samples and the post information samples through the depth forest model to determine a first feature vector set corresponding to each resume sample and a second feature vector set corresponding to each post information sample, calculating the matching probability value of each feature vector in the first feature vector set and each feature vector in the second feature vector set until the matching probability value meets a preset sample probability table, and determining that training on the depth forest model is completed.
In one implementation of the present application, a second evaluation score greater than a first preset threshold is determined as the pending post score. And determining the information of the undetermined positions corresponding to the scores of the undetermined positions. And determining a matching probability value corresponding to the information of the to-be-positioned posts. And determining a third evaluation score corresponding to each piece of to-be-positioned post information according to the matching probability value corresponding to the to-be-positioned post information, the to-be-positioned post score corresponding to the to-be-positioned post information and the preset matching weight. And taking the post information corresponding to the third evaluation score larger than the second preset threshold value as the post information matched with the currently pending personnel.
In one implementation of the present application, a data dictionary of post information and corresponding resume information is generated. The data dictionary includes several triples of text. The triplet text includes: personnel name, employment capability text corresponding to resume information and post name. And respectively encoding each triplet text into vector groups through a preset TransH model, and projecting each vector group to a corresponding relation hyperplane so as to calculate the recommended score of each triplet text in the data dictionary according to a preset formula and the projection vector of each vector group in the relation hyperplane. The vector group comprises a personnel name vector, a employment capability text vector and a post name vector; the relational hyperplane is the hyperplane corresponding to the employment capability text vector in the vector group. And rejecting the triplet text with the recommended score smaller than the preset value, and updating the data dictionary to determine the post information according to the updated data dictionary.
In one implementation of the present application, a plurality of recruitment information and post keywords corresponding to the recruitment information are obtained. And sequentially combining the post keywords into post keywords according to the position sequence of the post keywords in the recruitment information, taking the post keywords as training samples, and inputting a bidirectional LSTM-CRF model to train the bidirectional LSTM-CRF model.
In one implementation of the present application, post keywords in the recruitment information are screened through a bi-directional LSTM-CRF model, so that the screened post keywords are used as post responsibility sample words. And determining post requirement sample words corresponding to each post responsibility text in recruitment information. And inputting the post responsibility sample words and the post requirement sample words into a keyword extraction model for training so as to obtain post information through the trained keyword extraction model. The keyword extraction model is a TextRank algorithm model.
In one implementation of the present application, feedback information of a user terminal is obtained. And determining the accuracy and recall rate of the post information according to the feedback information. And determining whether the post information is to-be-updated post information according to the accuracy rate and the recall rate. If yes, the recruiter information corresponding to the position information to be updated is removed, so that the first evaluation score and the second evaluation score are updated according to the updated recruiter information, and the position information to be updated is updated.
On the other hand, the embodiment of the application also provides employment information processing equipment for the crowd to be employment, which comprises:
at least one processor; and a memory communicatively coupled to the at least one processor. Wherein the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to:
and acquiring a plurality of post information. The post information at least comprises: post name, post responsibility text, post demand text. And determining a first evaluation score corresponding to each position information based on the recruiter information corresponding to each position information and resume information of the currently pending person. The first evaluation score represents the matching degree of the currently undetermined person and the post information. And determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model. And determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model. And determining at least one post information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the post information to the corresponding user terminal.
According to the method and the device, the first evaluation scores of the undetermined personnel to be employment and the historical recruiters corresponding to the post information are determined, and then the post information matched with the undetermined personnel is obtained through the lingering semantic model and the depth forest model, so that employment information recommendation is accurately performed. And employment information can be processed, identified and screened for reference by people so as to solve most of labor cost.
The method and the system can also utilize working experience and technical capability mastered by the employment staff to quickly find out proper posts, so that resume delivery of the employment staff is not blindly delivered, normal employment of the employment staff is ensured, and long-term occupation planning is established. And the method and the device realize convenient and accurate employment information processing and post information recommendation according with the characteristics of the personnel to be employment by combining a plurality of algorithms.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:
fig. 1 is a schematic flow chart of a employment information processing method for a crowd to be employment according to an embodiment of the present application;
Fig. 2 is a schematic structural diagram of employment information processing equipment for a crowd to be employment according to an embodiment of the present application.
Detailed Description
For the purposes, technical solutions and advantages of the present application, the technical solutions of the present application will be clearly and completely described below with reference to specific embodiments of the present application and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
Currently, the complex and various post information makes it difficult for many job seekers to find post information suitable for themselves. The employment crowd relates to students at the university, employment-losing employment-waiting staff, young employment-losing staff, elderly employment crowd, retired soldiers and the like. The employment number is large, the employment situation is severe, under the influence of social factors, staff changes greatly, the traditional recruitment information recruitment mode is subjective, the accuracy of accurately matching the posts is poor, and the current descriptions of all enterprises for the same posts are different and are not clear to position. Thus, it is difficult for the employment group to quickly find a proper position of the employment group. Recruitment is also difficult to quickly screen out suitable post response personnel among a large number of people waiting for employment.
Based on the above, the embodiment of the application provides employment information processing method and equipment for a crowd to be employment, which are used for solving the problems that employment information is processed, identified and screened mainly by manpower, the work is complex, the workload is large, and the labor cost is wasted, and the current personnel to be employment cannot quickly find out proper work according to the working experience and technical capability mastered by the personnel to be employment, so that the problem that the normal employment and long-term occupation planning of the personnel to be employment are influenced due to blind delivery of resume delivery is solved.
Various embodiments of the present application are described in detail below with reference to the accompanying drawings.
The embodiment of the application provides a employment information processing method for a crowd to be employment, as shown in fig. 1, the method may include steps S101-S105:
s101, the server acquires a plurality of post information.
The post information at least comprises: post name, post responsibility text, post demand text.
In embodiments of the present application, the post information may include post names, post responsibility texts, post requirement texts, e.g., attendant, post responsibility texts: welcome/pass/serve/sanitation; post demand text: physical health/appearance with good quality of qi/18-30 years old/star hotel/kender, mcdonald, must win guests/health card/shift.
It should be noted that, the server is merely an exemplary implementation subject of the employment information processing method for the crowd to be employment, and the implementation subject is not limited to the server, and the application is not limited thereto specifically.
In this embodiment of the present application, before the server obtains the plurality of post information, the method further includes:
the server acquires a plurality of recruitment information and post keywords corresponding to the recruitment information.
The server may acquire recruitment related text by using a crawler program or an octopus and a back collector, and perform data management on the related text, for example: removing special character data such as br, div, span and the like; duplicate data (one post ambiguous) is removed, missing data and supplemental incomplete data; and (3) treating places where the work description field is inconsistent with other fields of the post, removing misleading post description, and eliminating ambiguous data. Translating English data into Chinese data.
Then, the server sequentially combines the post keywords into post keywords according to the position sequence of the post keywords in the recruitment information, takes the post keywords as training samples, and inputs the two-way LSTM-CRF model to train the two-way LSTM-CRF model.
Specifically, the post description related text is segmented according to punctuation marks, the post requirements of the special character pattern= |post description are filtered regularly, then long sentences are segmented into short sentences, then entity dictionaries and word boundaries are obtained, and words at each position in the short sentences are converted into word vectors. Combining a plurality of word vectors to obtain a short sentence as the input of a model, and providing the short sentence to the model for learning
The step of obtaining the post information by the server specifically comprises the following steps:
firstly, the server screens position keywords in a plurality of recruitment information through a bidirectional LSTM-CRF model, so that the screened position keywords are used as position responsibility sample words.
For example, the recruitment information includes "handling law office cases, legal consultation, prosecution, evidence arrangement, and complaint, and participating in internal training of regular organization of law and courses of related activities, forums, training, etc. held together by external industry associations and institutions", and then the post responsibility sample word can be obtained through a bidirectional LSTM-CRF model: legal consultation drafting complaint participation training … … for treatment law case
Then, the server determines the post requirement sample words corresponding to the post responsibility texts in the recruitment information.
The post requirement sample word generation uses an automatic summary generation technology. The automated generation of the post requirements is described by giving a post requirement sample vocabulary B with |B| different posts, i.e
Figure BDA0003849924900000071
Wherein X is i Is a job, which describes the responsibility of the ith job, Y i Is a post requirement that describes the various capacity requirements of this post. Specifically, for each job role X i Assuming that it contains H d Individual words, i.e.
Figure BDA0003849924900000074
The post requirements typically contain multiple sentences to describe different capacity requirements, so each post requirement Y i Represented as
Y i ={y 1 ,y 2 ,...y N }。
Wherein y is i Is the j-th sentence. For example, give 5 post requirement sentences, i.eN=5, which corresponds to introduction of professional institutions, academia, work experiences, communication capabilities, thinking about the ability. Further assume that each y i Included
Figure BDA0003849924900000073
Individual words, i.e.
Figure BDA0003849924900000072
For example, the post responsibility sample word in the above example may correspond to the following post requirement sample word: law, law qualification certificate.
Then, the server inputs the post responsibility sample words and the post requirement sample words into a keyword extraction model for training so as to obtain post information through the trained keyword extraction model. The keyword extraction model is a TextRank algorithm model.
The server can sort the post responsibility sample words and post requirement sample words by a voting mechanism, sentences after sorting are post information samples, and further training of the keyword extraction model is completed, so that post information is extracted through the keyword extraction model after training is completed.
By the short text description identification mode, post information is generated, and the situation that the enterprise can not normally match with the employment staff capacity and posts due to the description difference of different posts can be avoided. And the text is processed, so that the calculation amount of a server can be saved, and the employment information processing time is shortened.
S102, the server determines a first evaluation score corresponding to each position information based on recruiter information corresponding to each position information and resume information of the currently pending person.
The first evaluation score represents the matching degree of the currently undetermined person and the post information.
In this embodiment of the present application, the server determines, based on recruiter information corresponding to each position information and resume information of the currently pending person, a first evaluation score corresponding to each position information, including:
firstly, the server determines cosine similarity between a text corresponding to the resume information and a text corresponding to each recruiter information according to resume information of a current undetermined person and information of each recruiter.
The corresponding text is an evaluation text for the post.
The resume information of the pending person can be resume related information uploaded by the pending person, and the server determines cosine similarity of the recruiter information and the resume information. The server can determine the similarity between the resume of the undetermined person and the information of each recruiter through a collaborative filtering recommendation algorithm.
Specifically, the server encodes the evaluation text of the posts in the resume information and the evaluation text of the posts by each recruiter in the recruiter information, for example: work overtime is more, and the codes are two vector work overtime [1, 0] and more [2,0,0]; the working competition pressure is large, encoded as working competition [1,2,0], the pressure is large [1,2,1] … …, and then the cosine similarity of the corresponding text is calculated by the following cosine similarity calculation company:
Figure BDA0003849924900000091
wherein cosA is cosine similarity of evaluation text J corresponding to resume information and evaluation text Z corresponding to recruiter information, J x To evaluate the xth word vector of text J, Z x To evaluate the xth word vector of text Z, y is the total number of word vectors.
And then, the server generates a similarity matrix of the currently pending person and each recruiter according to each cosine similarity.
The server may build a similarity matrix to simplify the amount of computation when computing the evaluation scores. Specifically, after the server obtains the cosine similarity between the evaluation text obtained according to the resume information of the current pending person and the evaluation text of the recruiter for the same post, the current pending person and the recruiter can be used as rows of the matrix, and the post can be used as columns of the matrix, so that the cosine similarity matrix of the evaluation text of the current pending person and the recruiter for each post can be obtained. The recruiter in the similarity matrix may be a recruiter with cosine similarity smaller than a preset value, after the recruiter in the similarity matrix is removed, for example, the recruiter information with cosine similarity smaller than a is removed when the preset value is a, and the recruiter close to the current undetermined person and having similar preference is reserved. The method and the system can write cosine similarity between all recruiters and the current undetermined person in the similarity matrix, and reject recruiter information with cosine similarity smaller than a before the first evaluation score is generated subsequently.
Then, the server determines a position score corresponding to the current pending person as a first evaluation score based on the evaluation text of each recruiter for each position and the corresponding cosine similarity in the similarity matrix.
The server can determine the evaluation score corresponding to the evaluation text according to the evaluation text of each recruiter for each post, and the server can determine the evaluation score corresponding to the evaluation text by using a preset model or a preset text comparison table (text and evaluation score comparison table). The server determines a similarity matrix generated by each historical recruiter and the pending person, and then calculates the similarity with the historical recruiter when the current pending person evaluates each post by the following formula.
Figure BDA0003849924900000092
Wherein A is 1 A is the similarity of the evaluation of the third post by the pending person to the first historic recruiter, a 1 For the evaluation score of the first post by the pending person, a 2 For the evaluation score of the undetermined person on the second post, b 1 A first post evaluation score for a first historic recruiter, b 2 A rating score for the second post for the first historic recruiter.
A first evaluation score is calculated according to the following formula, taking three historical recruiters as an example:
Figure BDA0003849924900000101
wherein T is 3 For the first evaluation score of the pending person for the third station,
Figure BDA0003849924900000102
for similarity of the evaluation of the third position by the pending person and the second historic recruiter,/for the evaluation of the third position by the pending person and the second historic recruiter >
Figure BDA0003849924900000103
B for similarity of the pending person to the third post's evaluation of the third historic recruiter 3 A third post evaluation score for the first historic recruiter, c 3 A third post evaluation score for the second historical recruiter, d 3 And a third post evaluation score for the third historical recruiter.
After the first evaluation scores of the undetermined personnel on each post are obtained according to collaborative filtering recommendation, a certain difference exists between the recommended post and an expected result, so that the server further processes employment information through the following method.
S103, the server determines a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model.
In this embodiment of the present application, the server determines, according to the first evaluation score and a preset lingering semantic model, a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score, including:
firstly, the server inputs resume information of the currently pending person and an evaluation text corresponding to the first evaluation score into the lingo-meaning model so that the lingo-meaning model determines a scoring matrix corresponding to the currently pending person and the recruiter corresponding to the first evaluation score.
The server can input the resume information and the evaluation text corresponding to the scores of the first evaluation scores on all posts, namely the evaluation text corresponding to the resume information, of the undetermined personnel on all posts, wherein the evaluation text can be generated through the resume information and the obtained first evaluation scores. The generation means that when the resume information of the undetermined person does not have the evaluation text of the pair X position, the first evaluation score is 5 on the pair X position, the corresponding historical recruiter m is 5 on the pair X position, and the evaluation text of m can be used as the evaluation text of the undetermined person. And further, processing the evaluation text through the latent semantic model, and obtaining the score of the undetermined personnel on each post through the pre-trained latent semantic model.
The latent model can determine the interest degree of the personnel to be determined in the post hidden class and the hidden class weight of the post, and the objective function of the latent model is continuously optimized through the loss function constructed by the punishment factors of the latent model, so that the high-quality evaluation score output by the latent model is obtained.
And then, the server carries out factorization processing on the scoring matrix through the latent semantic model and a preset loss function until the preset loss function meets preset conditions so as to decompose the scoring matrix into a user scoring matrix and a post weight matrix.
After the scoring matrix is obtained, the scoring matrix can be decomposed into a user scoring matrix and a post weight matrix according to the interestingness of the personnel to be determined to the post hidden class and the post hidden class weight. In order to ensure the minimum error during decomposition, the decomposition is performed by setting a loss function, for example, the random gradient descent method is used for minimizing, solving the loss function, decomposing the scoring matrix, and determining that the decomposition is completed under the condition that the loss function is smaller than a set value.
Then, the server determines predictive rating scores for the posts of the currently pending person based on the user rating matrix and the post weight matrix.
The server can calculate the product by using the user scoring matrix and the post weight matrix, and calculate the predictive evaluation score by the following formula:
Figure BDA0003849924900000111
wherein r is ui Predictive evaluation score for recruitment position i for person u, p uk Scoring the interest level user of the kth hidden class of the post for the person u, which corresponds to a user scoring matrix; q ki The effect of the kth hidden class on position i is scored for a weight that corresponds to the position weight matrix.
Then, the server takes the weighted average of the first evaluation score and the predicted evaluation score as the second evaluation score.
After the server obtains the predictive evaluation score outputted by the latent semantic model, a weighted average of the first evaluation score and the predictive evaluation score corresponding to the first evaluation score is obtained, and the weighted average is used as a second evaluation score. The weights corresponding to the first evaluation score and the predicted evaluation score in the weighted average may be set by the user, which is not specifically limited in this application. The implicit behavior of the user is introduced into the scoring consideration of all posts, so that more accurate recommendation results can be obtained when posts are recommended.
S104, the server determines a matching probability value corresponding to the resume information and each post information through a pre-trained deep forest model.
In this embodiment of the present application, before determining, by the server, the matching probability value corresponding to the resume information and each post information through the pre-trained deep forest model, the method further includes:
the server acquires a plurality of resume samples and a plurality of post information samples. And then, the server respectively carries out short text coding processing on each resume sample and each post information sample through the depth forest model so as to determine a first feature vector set corresponding to each resume sample and a second feature vector set corresponding to each post information sample, calculates the matching probability value of each feature vector in the first feature vector set and each feature vector in the second feature vector set until the matching probability value meets a preset sample probability table, and determines that training on the depth forest model is completed.
The server needs to train the depth forest model to obtain the depth forest model fused with implicit features, and the first feature vector set is a feature vector obtained from a profile sample and contains user latent features. The second set of feature vectors consists of feature vectors of post information samples, including post lingering features. And matching the feature vectors in the first feature vector set with the feature vectors in the second feature vector set respectively, wherein the matching can be performed by vector cross multiplication, and the calculation result is used as a matching probability value of each feature vector.
The preset sample probability table may be a probability table corresponding to the resume sample and the post information sample, for example, the matching probability value of each feature vector in the resume sample and each feature vector in the post information sample is preset, training is performed by the depth forest model until the matching probability value of each feature vector in the resume sample and each feature vector in the post information sample obtained by training is matched with the probability value in the preset sample probability table, so as to obtain the depth forest model after training is completed.
S105, the server determines at least one position information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the position information to the corresponding user terminal.
In this embodiment of the present application, the server determines, based on the second evaluation score and the matching probability value, at least one post information that the currently pending person matches, including:
first, the server determines a second rating score greater than a first preset threshold as a to-be-located post score.
The server can reject the second evaluation score which is too small, and the recommendation accuracy when the employment information is recommended is ensured.
Then, the server determines the undetermined post information corresponding to the undetermined post score.
The to-be-positioned position score corresponds to position information, which is to-be-positioned position information.
Meanwhile, the server determines a matching probability value corresponding to the information of the to-be-positioned posts.
And then, the server determines a third evaluation score corresponding to each piece of to-be-positioned post information according to the matching probability value corresponding to the to-be-positioned post information, the to-be-positioned post score corresponding to the to-be-positioned post information and the preset matching weight.
The preset matching weights are set by the user, for example, the matching probability value and the weight of the score of the position to be determined are respectively 0.6 and 0.4, and then the third evaluation score is: y=0.6u+0.4p, wherein the third evaluation score is Y, the matching probability value is U, and the undetermined post score is P.
And finally, the server takes the post information corresponding to the third evaluation score larger than the second preset threshold value as the post information matched with the currently undetermined person.
The second preset threshold is set by the user, which is not specifically limited in this application.
In this embodiment of the present application, before the server sends the post information to the corresponding user terminal, the method further includes:
the server generates a data dictionary of post information and corresponding resume information.
The data dictionary includes several triples of text. The triplet text includes: personnel name, employment capability text corresponding to resume information and post name.
The data dictionary may be as follows: zhang Sant educational experience\t method management; zhang Sant professional skills t lawyer.
And then, the server codes each triplet text into vector groups through a preset TransH model, and projects each vector group to a corresponding relation hyperplane so as to calculate the recommended score of each triplet text in the data dictionary according to a preset formula and the projection vector of each vector group in the relation hyperplane.
The vector group comprises a personnel name vector, a employment capability text vector and a post name vector; the relational hyperplane is the hyperplane corresponding to the employment capability text vector in the vector group.
Step 1, dividing a head part, a relation part and a tail part; step 2, converting the index tensor into an embedded tensor; step 3, embedding the head and the tail of the project entity into the associated hyperplane; and 4, calculating the correlation hyperplane similarity score, namely the recommendation score. In the embodiment of the application, the personnel name vector and the post name vector in the data dictionary are projected onto the hyperplane corresponding to the employment capability text vector along the normal vector of the hyperplane, wherein the projected vector can be expressed as,
h =h-w r T *h*w r ,t =t-w r T *t*w r
Wherein h is For projected person name vector, h is person name vector, w r Is the hyperplane normal vector, t And t is a post name vector. The recommendation score calculation formula:
Figure BDA0003849924900000141
where f is the recommendation score and r is the employment capability text vector.
And then, the server eliminates the triplet text with the recommended score smaller than the preset value, and updates the data dictionary to determine the post information according to the updated data dictionary.
In this embodiment of the present application, after the server sends the post information to the corresponding user terminal, the method further includes:
firstly, a server acquires feedback information of a user terminal.
The user terminal may be a mobile phone, a tablet computer, or other devices of the user, which is not particularly limited in this application. The feedback information may be an accurate result of recommended post information fed back by the user through the user terminal, e.g. "post information does not fit into professional planning", "post information does not fit into my interests", etc.
And then, the server determines the accuracy and recall rate of the post information according to the feedback information.
Wherein, TP: TP is a positive class and is determined to be a positive class; FP: FP is a negative class and is determined to be a positive class; FN: FN is a positive class and is determined to be a negative class; TN: TN is a negative class and is determined to be a negative class; accuracy = TP/(tp+fp); recall = TP/(tp+fn).
And then, the server determines whether the post information is to-be-updated post information according to the accuracy rate and the recall rate.
And under the condition that the position information is determined to be the position information to be updated, removing recruiter information corresponding to the position information to be updated, so as to update the first evaluation score and the second evaluation score according to the updated recruiter information, and update the position information to be updated.
Figure BDA0003849924900000151
And under the condition that F1 is smaller than a fourth preset threshold value, determining the post information as the post information to be updated.
According to the method and the device, the first evaluation scores of the undetermined personnel to be employment and the historical recruiters corresponding to the post information are determined, and then the post information matched with the undetermined personnel is obtained through the lingering semantic model and the depth forest model, so that employment information recommendation is accurately performed. The method and the system can quickly find out the proper posts by utilizing the working experience and technical capability mastered by the employment staff, so that the resume delivery of the employment staff is not blindly delivered any more, the normal employment of the employment staff is ensured, and long-term occupation planning is established.
The employment information can be processed, identified and screened for reference by people, so that most of labor cost is solved. In addition, the method and the device can provide corresponding training courses according to resume information of the undetermined personnel, and quicken employment of the undetermined personnel.
Fig. 2 is a schematic structural diagram of employment information processing equipment for a crowd to be employment according to an embodiment of the present application, where the equipment includes:
at least one processor; and a memory communicatively coupled to the at least one processor. Wherein the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to:
and acquiring a plurality of post information. The post information at least comprises: post name, post responsibility text, post demand text. And determining a first evaluation score corresponding to each position information based on the recruiter information corresponding to each position information and resume information of the currently pending person. The first evaluation score represents the matching degree of the currently undetermined person and the post information. And determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model. And determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model. And determining at least one post information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the post information to the corresponding user terminal.
All embodiments in the application are described in a progressive manner, and identical and similar parts of all embodiments are mutually referred, so that each embodiment mainly describes differences from other embodiments. In particular, for the apparatus embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments in part.
The devices and the methods provided in the embodiments of the present application are in one-to-one correspondence, so that the devices also have similar beneficial technical effects as the corresponding methods, and since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices are not described here again.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.

Claims (10)

1. A employment information processing method for a crowd to be employment, the method comprising:
acquiring a plurality of post information; the post information at least comprises: post name, post responsibility text and post demand text;
determining a first evaluation score corresponding to each position information based on recruiter information corresponding to each position information and resume information of a currently pending person; the first evaluation score represents the matching degree of the currently undetermined person and each post information; the first evaluation score is obtained by calculation according to a similarity matrix generated by cosine similarity corresponding to the recruiter information and the resume information and an evaluation text of the recruiter;
determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingering semantic model; the second evaluation score is obtained according to a weighted average value of the first evaluation score and the predicted evaluation score; the predictive evaluation score is calculated through a user scoring matrix and a post weight matrix corresponding to the lingering semantic model; the user scoring matrix and the post weight matrix are obtained by decomposing the scoring matrix for the latent semantic model; the scoring matrix is determined for the lingo semantic model based on the resume information and the corresponding evaluation text of the first scoring value;
Determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model; the depth forest model is obtained based on training of a plurality of resume samples and a plurality of post information samples and a preset sample probability table; the sample probability table comprises preset matching probability values of each characteristic vector in the resume samples and each characteristic vector in the post information samples;
and determining at least one piece of post information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the post information to a corresponding user terminal.
2. The method of claim 1, wherein determining a first evaluation score corresponding to each of the post information based on the recruiter information corresponding to each of the post information and resume information for the currently pending person, comprises:
according to the resume information of the currently pending person and the recruiter information, determining cosine similarity between the text corresponding to the resume information and the text corresponding to the recruiter information; the corresponding text is an evaluation text for the post;
Generating a similarity matrix of the current undetermined person and each recruiter according to each cosine similarity;
and determining a post score corresponding to the current undetermined person as the first evaluation score based on the evaluation text of each recruiter for each post and the corresponding cosine similarity in the similarity matrix.
3. The method according to claim 1, wherein determining a corresponding second evaluation score of the undetermined person corresponding to the first evaluation score according to the first evaluation score and a preset lingo-semantic model specifically comprises:
inputting resume information of the current undetermined person and an evaluation text corresponding to the first evaluation score into the lingo-semantic model so that the lingo-semantic model determines a scoring matrix corresponding to the current undetermined person and the recruiter corresponding to the first evaluation score;
performing factorization processing on the scoring matrix through the latent semantic model and a preset loss function until the preset loss function meets preset conditions so as to decompose the scoring matrix into a user scoring matrix and a post weight matrix;
based on the user scoring matrix and the post weight matrix, determining the predictive evaluation scores of the currently pending personnel on each post;
And taking a weighted average of the first evaluation score and the predicted evaluation score as the second evaluation score.
4. The method of claim 1, wherein prior to determining the matching probability values for the resume information and each of the post information by a pre-trained deep forest model, the method further comprises:
acquiring a plurality of resume samples and a plurality of post information samples;
and respectively carrying out short text coding processing on each resume sample and each post information sample through the depth forest model to determine a first feature vector set corresponding to each resume sample and a second feature vector set corresponding to each post information sample, calculating a matching probability value of each feature vector in the first feature vector set and each feature vector in the second feature vector set until the matching probability value meets a preset sample probability table, and determining that training on the depth forest model is completed.
5. The method of claim 4, wherein determining at least one of the post information for the currently pending person to match based on the second evaluation score and the match probability value, comprises:
Determining the second evaluation score which is larger than a first preset threshold value as a to-be-positioned post score;
determining the information of the to-be-positioned posts corresponding to the score of the to-be-positioned posts; and
determining the matching probability value corresponding to the information of the to-be-positioned post;
determining a third evaluation score corresponding to each piece of to-be-positioned position information according to the matching probability value corresponding to the to-be-positioned position information, the to-be-positioned position score corresponding to the to-be-positioned position information and a preset matching weight;
and taking the post information corresponding to the third evaluation score larger than a second preset threshold as the post information matched with the currently pending person.
6. The method of claim 5, wherein prior to transmitting the post information to the corresponding user terminal, the method further comprises:
generating a data dictionary of the post information and the corresponding resume information; the data dictionary comprises a plurality of triples of text; the triplet text includes: personnel names, employment capability texts corresponding to the resume information and post names;
encoding each triplet text into a vector group through a preset TransH model, and projecting each vector group to a corresponding relational hyperplane so as to calculate a recommended score of each triplet text in the data dictionary according to a preset formula and a projection vector of each vector group in the relational hyperplane; the vector group comprises a personnel name vector, a employment capability text vector and a post name vector; the relation hyperplane is a hyperplane corresponding to the employment capability text vector in the vector group;
And eliminating the triplet text with the recommended score smaller than a preset value, and updating the data dictionary to determine the post information according to the updated data dictionary.
7. The method of claim 1, wherein prior to obtaining the plurality of post information, the method further comprises:
acquiring a plurality of recruitment information and post keywords corresponding to the recruitment information;
and sequentially combining the post keywords into post keywords according to the position sequence of the post keywords in the recruitment information, taking the post keywords as training samples, and inputting a bidirectional LSTM-CRF model to train the bidirectional LSTM-CRF model.
8. The method of claim 7, wherein the method further comprises:
screening a plurality of post keywords in the recruitment information through the bidirectional LSTM-CRF model, and taking the screened post keywords as post responsibility sample words;
determining post demand sample words corresponding to the post responsibility texts in the recruitment information;
inputting the post responsibility sample words and the post requirement sample words into a keyword extraction model for training so as to acquire the post information through the trained keyword extraction model; the keyword extraction model is a TextRank algorithm model.
9. The method of claim 1, wherein after transmitting the post information to the corresponding user terminal, the method further comprises:
acquiring feedback information of the user terminal;
determining the accuracy and recall rate of the post information according to the feedback information;
determining whether the post information is to-be-updated post information according to the accuracy rate and the recall rate;
if yes, rejecting the recruiter information corresponding to the position information to be updated, so as to update the first evaluation score and the second evaluation score according to the updated recruiter information, and update the position information to be updated.
10. A employment information processing apparatus for a crowd to be employment, the apparatus comprising:
at least one processor; the method comprises the steps of,
a memory communicatively coupled to the at least one processor; wherein,,
the memory stores instructions executable by the at least one processor to enable the at least one processor to:
acquiring a plurality of post information; the post information at least comprises: post name, post responsibility text and post demand text;
Determining a first evaluation score corresponding to each position information based on the historical recruiter information corresponding to each position information and the resume information of the currently pending person; the first evaluation score represents the matching degree of the currently undetermined person and each post information; the first evaluation score is obtained by calculation according to a similarity matrix generated by cosine similarity corresponding to the recruiter information and the resume information and an evaluation text of the recruiter;
determining a corresponding second evaluation score of the currently pending person according to the first evaluation score and a preset lingering semantic model; the second evaluation score is obtained according to a weighted average value of the first evaluation score and the predicted evaluation score; the predictive evaluation score is calculated through a user scoring matrix and a post weight matrix corresponding to the lingering semantic model; the user scoring matrix and the post weight matrix are obtained by decomposing the scoring matrix for the latent semantic model; the scoring matrix is determined for the lingo semantic model based on the resume information and the corresponding evaluation text of the first scoring value;
determining a matching probability value corresponding to the resume information and each post information through a pre-trained depth forest model; the depth forest model is obtained based on training of a plurality of resume samples and a plurality of post information samples and a preset sample probability table; the sample probability table comprises preset matching probability values of each characteristic vector in the resume samples and each characteristic vector in the post information samples;
And determining at least one piece of post information matched with the currently pending person based on the second evaluation score and the matching probability value, so as to send the post information to a corresponding user terminal.
CN202211128537.7A 2022-09-16 2022-09-16 Employment information processing method and equipment for crowd to be employment Active CN115526589B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211128537.7A CN115526589B (en) 2022-09-16 2022-09-16 Employment information processing method and equipment for crowd to be employment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211128537.7A CN115526589B (en) 2022-09-16 2022-09-16 Employment information processing method and equipment for crowd to be employment

Publications (2)

Publication Number Publication Date
CN115526589A CN115526589A (en) 2022-12-27
CN115526589B true CN115526589B (en) 2023-06-06

Family

ID=84697873

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211128537.7A Active CN115526589B (en) 2022-09-16 2022-09-16 Employment information processing method and equipment for crowd to be employment

Country Status (1)

Country Link
CN (1) CN115526589B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117196550B (en) * 2023-09-22 2024-05-28 蔓悦科技(宁波)有限公司 Talent and enterprise supply and demand matching method and system

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110633960A (en) * 2019-09-25 2019-12-31 重庆市重点产业人力资源服务有限公司 Human resource intelligent matching and recommending method based on big data
CN111144723A (en) * 2019-12-17 2020-05-12 埃摩森网络科技(上海)有限公司 Method and system for recommending people's job matching and storage medium
CN111105209B (en) * 2019-12-17 2023-07-21 上海沃锐企业发展有限公司 Job resume matching method and device suitable for person post matching recommendation system
CN113435841A (en) * 2021-06-24 2021-09-24 浙江工贸职业技术学院 Talent intelligent matching recruitment system based on big data
CN114819924B (en) * 2022-06-28 2022-09-23 杭银消费金融股份有限公司 Enterprise information push processing method and device based on portrait analysis

Also Published As

Publication number Publication date
CN115526589A (en) 2022-12-27

Similar Documents

Publication Publication Date Title
Neculoiu et al. Learning text similarity with siamese recurrent networks
CN108021616B (en) Community question-answer expert recommendation method based on recurrent neural network
CN109635291B (en) Recommendation method for fusing scoring information and article content based on collaborative training
Min et al. Question answering through transfer learning from large fine-grained supervision data
US20180181544A1 (en) Systems for Automatically Extracting Job Skills from an Electronic Document
CN113435203A (en) Multi-modal named entity recognition method and device and electronic equipment
US20200210957A1 (en) Classification of job titles via machine learning
US20210182496A1 (en) Machine learning techniques for analyzing textual content
CN111626616A (en) Crowdsourcing task recommendation method
CN106503123A (en) A kind of deep learning intelligent response system based on computer cloud data
CN111738002A (en) Ancient text field named entity identification method and system based on Lattice LSTM
Althagafi et al. Arabic tweets sentiment analysis about online learning during COVID-19 in Saudi Arabia
Ezen-Can et al. Understanding Student Language: An Unsupervised Dialogue Act Classification Approach.
CN108052504A (en) Mathematics subjective item answers the structure analysis method and system of result
CN113627797B (en) Method, device, computer equipment and storage medium for generating staff member portrait
CN115329200A (en) Teaching resource recommendation method based on knowledge graph and user similarity
CN115526589B (en) Employment information processing method and equipment for crowd to be employment
JP2016153931A (en) Information processing method, information processing device, and information processing program
CN109408803A (en) A method of it semantic understanding for subjective item natural language and corrects
Wahde et al. DAISY: an implementation of five core principles for transparent and accountable conversational AI
KR20200015054A (en) System for evaluating supply mock exam automatically
Zhao et al. What are students thinking and feeling? Understanding them from social data mining
Amiri et al. Spotting spurious data with neural networks
Okhdar et al. English vocabulary learning through recommender system based on sentence complexity and vocabulary difficulty
CN113486649B (en) Text comment generation method and electronic device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant