Talent recommendation method and device based on talent hierarchical model, equipment and medium

Abstract

The invention relates to the technical field of machine learning, in particular to a talent recommendation method based on a talent hierarchical model, and the method comprises the steps: obtaining a talent resume, and extracting a talent information feature code; the talent information feature codes are input into a pre-trained talent hierarchical model, hierarchical labels of the talent resumes are output, the talent hierarchical model is obtained by training multiple pieces of sample data, and the sample data comprise the talent information feature codes and the hierarchical labels; the hierarchical label is determined based on the applicant behavior data of the applicant and the recruitment behavior data of the recruiter to the applicant; dividing the talent resumes into a plurality of resume sets according to the hierarchical labels; and screening out talent resumes meeting the requirements of the target user from the resume set associated with the target user, and recommending the talent resumes to the target user. The talent hierarchical model is trained by utilizing the behavior data of the recruiters and the applicants, so that the model can make more accurate judgment on the talents to perform hierarchical processing.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a talent recommendation method, device, electronic device and storage medium based on a talent stratification model. Background Art

[0002] With the rapid development of online recruitment platforms, the number of platform users continues to increase, and the matching needs of job seekers and companies are becoming increasingly urgent. In order to improve recruitment efficiency and optimize user experience, online recruitment platforms have gradually introduced the concept of talent stratification. Talent stratification refers to the division of job seekers into different levels based on their background, skills, experience and other characteristics, so that recruitment platforms can provide personalized services and recommendations based on the job seekers' levels. Accurate talent stratification can help platforms implement more effective search and recommendation strategies, improve matching and conversion rates in the recruitment process, and thus achieve efficient distribution and monetization of traffic.

[0003] Currently, recruitment platforms mainly use rule-based methods to stratify talents. Specifically, a set of specific rules is designed for each function, and talent stratification is performed based on the job seeker's label information (such as gender, age, education, school, years of work experience, etc.). Rule-based methods usually use predefined hard standards, which ensures a certain degree of accuracy in rule judgments, but also limits their coverage. Many job seekers may be excluded simply because they do not meet one of the rules, resulting in some excellent job seekers being misclassified, affecting the job seeker experience and the platform's recruitment efficiency.

[0004] Moreover, the job market and economic situation are changing dynamically, and the requirements for talents will also change accordingly. However, rule-based methods are difficult to perceive these changes in real time, and they need to rely on manual regular inspection and adjustment of rules, which is not only time-consuming and labor-intensive, but also easily causes the rules to lag behind market changes, thereby reducing the accuracy and effectiveness of stratification. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention proposes a talent recommendation method, device, electronic device and storage medium based on a talent stratification model to improve user satisfaction with the recommendation results.

[0006] According to one aspect of the present application, a talent recommendation method based on a talent stratification model is proposed, the method comprising:

[0007] Obtain talent resumes and extract talent information feature codes, wherein the talent information feature codes include one or more of basic information feature values, education information feature values, and work experience feature values; input the talent information feature codes into a pre-trained talent stratification model, and output stratification labels of the talent resumes, wherein the talent stratification model is trained using a plurality of sample data, wherein the sample data includes talent information feature codes and stratification labels, and the stratification labels are determined based on the application behavior data of the applicants and the recruitment behavior data of the applicants by the recruiters; divide the talent resumes into a plurality of resume sets according to the stratification labels; screen out talent resumes that meet the requirements of the target user from the resume set associated with the target user, and recommend them to the target user.

[0008] The talent recommendation method as described above, wherein the hierarchical labels are determined based on the applicant's application behavior data and the recruiter's recruitment behavior data of the applicant, including: calculating the recruiter's hand-reaching rate and / or handshake rate based on the applicant's application behavior data and the recruiter's recruitment behavior data of the applicant within a predetermined time period, wherein the hand-reaching rate is used to indicate the recruiter's satisfaction with the resume, and the handshake rate is used to indicate the satisfaction of both the recruiter and the applicant; determining the applicant's hierarchical labels based on the recruiter's hand-reaching rate and / or handshake rate, wherein the hierarchical labels include a first hierarchical label and a second hierarchical label.

[0009] The talent recommendation method as described above further includes: calculating an excellent quantitative value A based on the recruiter's hand-reaching rate and handshake rate; sorting according to the size of the excellent quantitative value A; determining the stratification labels of the first 10%-40% of the applicants as the first stratification labels; and determining the stratification labels of the last 10%-40% of the applicants as the second stratification labels.

[0010] In the talent recommendation method described above, the excellent quantitative value A is expressed by the following formula:

[0011] Excellent quantitative value A = α × recruiter's hand-reaching rate + β × handshake rate;

[0012] Among them, α is the proportion of recruiters' hand-reaching rate, β is the proportion of handshake rate, α+β=1, the recruiters' hand-reaching rate is the ratio of the number of recruiters' hands reaching out to the number of resumes exposed, and the handshake rate is the ratio of the number of handshakes to the number of resumes submitted.

[0013] As described above, the talent recommendation method extracts talent information feature codes from talent resumes, including: extracting relevant data of target feature names from the resumes; parsing relevant data of target feature names and configuring corresponding feature values; and combining multiple feature values ​​into talent information feature codes according to preset rules.

[0014] The talent recommendation method as described above further includes: obtaining talent resumes and filtering out poor quality resumes from the talent resumes based on the resume quality score Q.

[0015] In the talent recommendation method as described above, the resume quality score Q is calculated using one or more of the following features in the resume: whether the content in the resume is too little; whether the sentences in the resume are fluent; whether there are garbled characters in the resume; whether the work experience is filled in the resume; whether the educational experience is filled in the resume, wherein different features are assigned different weight values.

[0016] According to another aspect of the present application, a talent recommendation device based on a talent stratification model is proposed, comprising: a feature extraction module, used to obtain talent resumes and extract talent information feature codes, wherein the talent information feature codes include one or more of basic information feature values, education information feature values ​​and work experience feature values; a talent stratification module, used to input the talent information feature codes into a pre-trained talent stratification model, and output stratification labels of the talent resumes, wherein the talent stratification model is trained using multiple sample data, wherein the sample data includes talent information feature codes and stratification labels, and wherein the stratification labels are determined based on the application behavior data of the applicant and the recruitment behavior data of the applicant by the recruiter; a resume division module, used to divide the talent resumes into multiple resume sets according to the stratification labels; a recommendation module, used to screen out talent resumes that meet the requirements of the target user from the resume set associated with the target user, and recommend them to the target user.

[0017] According to another aspect of the present application, an electronic device is proposed, which includes a processor and a memory, wherein a computer program instruction set is stored in the memory, and when the processor executes the computer program instruction set in the memory, the talent recommendation method based on the talent stratification model as described above is implemented.

[0018] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a computer program instruction set is stored on the computer-readable storage medium, and when the computer program instruction set is executed by a processor, the talent recommendation method based on the talent stratification model as described above is implemented.

[0019] The talent stratification method based on the machine learning model of this application can automatically learn from a large amount of historical data and explore a certain relationship between static information and dynamic information. The behavioral data of recruiters and applicants can better reflect the popularity of talents. Using popularity to stratify talents can make more accurate judgments. In addition, this application is also friendly to the talent classification of new users. When a new user registers on the platform, he may not have generated enough behavioral data (such as submitting resumes, interview records, etc.). In this case, this application can use the static information characteristics of new users (such as educational background, work experience, skills, etc.) to stratify talents. In this way, even in the absence of dynamic behavioral data, the system can still effectively classify and label users to ensure that new users can be quickly identified and processed by the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The preferred embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, wherein:

[0021] Figure 1 It is a flowchart of a talent recommendation method based on a talent stratification model according to an embodiment of the present application.

[0022] Figure 2 It is a flowchart of a method for determining hierarchical labels according to an embodiment of the present application.

[0023] Figure 3 It is a flow chart of the method for determining talent stratification based on rules.

[0024] Figure 4 It is a flow chart of a method for training a talent stratification model according to an embodiment of the present application.

[0025] Figure 5 It is a flowchart of a method for online talent stratification according to an embodiment of the present application.

[0026] Figure 6 A talent recommendation device based on a talent stratification model according to an embodiment of the present application.

[0027] Figure 7 It is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] In the following detailed description, reference may be made to the various specification drawings that are part of the present application and are used to illustrate specific embodiments of the present application. In the accompanying drawings, similar reference numerals describe substantially similar components in different figures. The various specific embodiments of the present application are described below in sufficient detail so that a person of ordinary skill in the art with relevant knowledge and skills in the art can implement the technical solutions of the present application. It should be understood that other embodiments may also be used or structural, logical or electrical changes may be made to the embodiments of the present application.

[0030] The talent stratification model of the present application can be a machine learning model. Machine learning is a branch of artificial intelligence. Machine learning algorithms can analyze input data, identify patterns, rules and relationships in the data, and make predictions, classifications, clustering, regression and other decisions or tasks based on these learned knowledge. The talent stratification model of the present application can utilize any one of the GBDT (Gradient Boosting Decision Tree) algorithm, logistic regression algorithm (LogisticRegression, LR) and deep learning in the machine learning algorithm. In particular, the GBDT algorithm, by constructing multiple decision trees and iteratively improving the model, can efficiently fit complex data patterns, and the training is efficient and interpretable.

[0031] Figure 1 FIG. 1 is a flow chart of a talent recommendation method based on a talent stratification model according to an embodiment of the present application. Figure 1 As shown, the method includes:

[0032] S110, obtaining resumes of talents and filtering out inferior resumes from the resumes of talents based on resume quality scores Q;

[0033] S120, extracting a talent information feature code from the filtered talent resume, wherein the talent information feature code includes one or more of a basic information feature value, an education information feature value, and a work experience feature value;

[0034] S130, inputting the talent information feature code into a pre-trained talent stratification model, and outputting a stratification label of the talent resume, wherein the talent stratification model is trained using a plurality of sample data, wherein the sample data includes the talent information feature code and the stratification label, and the stratification label is determined based on the applicant's application behavior data and the recruiter's recruitment behavior data on the applicant;

[0035] S140, dividing the talent resumes into multiple resume sets according to the hierarchical labels;

[0036] S150, selecting a resume that meets the requirements of the target user from the resume collection associated with the target user, and recommending the resume to the target user.

[0037] This application divides the information of recruiters and applicants on the recruitment platform into two categories: one is static information, such as basic personal information (such as name, age, contact information), education information (such as education, school, major) and work experience (such as company, position, years of work); the other is dynamic information, such as applicants' application behavior data (such as downloading resumes, active chatting, collecting resumes and submitting resumes) and recruiters' recruitment behavior data (such as downloading resumes, active chatting, collecting resumes and employment status).

[0038] Traditional rule-based stratification methods usually rely on artificially set rules and conditions (such as education, years of work experience, etc.). These rules may be too simple or subjective and cannot capture the complexity of talent information. The talent stratification method based on machine learning models can automatically learn from a large amount of historical data, and use the stratification labels determined by dynamic information and the talent information feature codes representing static information as sample data to input into the talent stratification model. The talent stratification model can explore a certain relationship between static information and dynamic information. The behavioral data of recruiters and applicants can better reflect the popularity of talents. Using popularity to stratify talents can make more accurate judgments.

[0039] In addition, this application is also friendly to the talent classification of new users. When a new user registers on the platform, he or she may not have generated enough behavioral data (such as submitted resumes, interview records, etc.). In this case, this application can use the static information characteristics of new users (such as educational background, work experience, skills, etc.) to stratify talents. In this way, even in the absence of dynamic behavioral data, the system can still effectively classify and label users to ensure that new users can be quickly identified and processed by the platform.

[0040] By stratifying talents, the platform can make personalized recommendations based on different levels of labels. For example, high-potential candidates may be recommended to companies or positions with higher requirements, while candidates with medium and low potential are matched with other suitable positions. In this way, the recommendation results are more in line with the needs and expectations of users (such as recruiters or job seekers), improving the accuracy and effectiveness of recommendations. This personalized matching mechanism can greatly improve the interview success rate and the recruitment rate, enhance the satisfaction of recruiters and job seekers, and increase users' dependence and stickiness on the platform.

[0041] Figure 2 FIG. 1 is a flow chart of a method for determining a hierarchical label according to an embodiment of the present application. Figure 2 As shown, the method includes:

[0042] S210, calculating the recruiter's hand-reaching rate and / or handshake rate based on the applicant's application behavior data and the recruiter's recruitment behavior data of the applicant within a predetermined time period, wherein the hand-reaching rate is used to indicate the recruiter's satisfaction with the resume, and the handshake rate is used to indicate the satisfaction of both the recruiter and the applicant;

[0043] S220, calculating an excellent quantitative value A according to the recruiter's hand-reaching rate and handshake rate;

[0044] S230, sorting according to the size of the excellent quantization value A;

[0045] S240, determining the stratification labels of the first 10%-40% of the applicants as the first stratification labels;

[0046] S250: Determine the stratified labels of the last 10%-40% of the applicants as the second stratified labels.

[0047] The recruiter's hand-in rate means that when HR sees a resume, the following behaviors can be regarded as a hand-in: downloading the resume, actively chatting, collecting the resume, etc. The HR hand-in rate reflects the HR's satisfaction with the resume. A high HR hand-in rate indicates an excellent resume.

[0048] The HR reaching rate is expressed by the following formula:

[0049] HR reaching out rate = (number of HR reaching out) / (number of resume exposures).

[0050] The handshake rate means that when a job seeker actively submits a resume, a handshake is formed when the HR has the following behaviors: downloading the resume, actively chatting, collecting the resume, etc. A handshake is a mutual satisfaction between the HR and the job seeker, and a high handshake rate reflects an excellent resume.

[0051] The handshake rate is expressed as follows:

[0052] Handshake rate = (number of handshakes) / (number of job applications submitted).

[0053] This application uses the behavioral data of recruiters and applicants (hand-reaching rate and handshake rate) to evaluate the excellence of resumes, avoiding the bias of subjective judgment and more objectively reflecting the popularity and excellence of applicants. The hand-reaching rate and handshake rate directly reflect the quality of interaction between recruiters and applicants. A high hand-reaching rate indicates that the resume is attractive, while a high handshake rate means that both parties are satisfied. Through these two indicators, the potential and value of applicants can be accurately evaluated.

[0054] The top 10%-40% of the ranked candidates are labeled as the first layer label, indicating the best or most potential candidates. The samples of the first layer label are used as positive samples to train the model. The bottom 10%-40% of the ranked candidates are labeled as the second layer label, indicating the candidates with relatively low potential. The samples of the second layer label can be used as negative samples to train the model.

[0055] According to one embodiment of the present application, the excellent quantization value A is expressed by the following formula:

[0056] Excellent quantitative value A = α × recruiter's hand-reaching rate + β × handshake rate;

[0057] Among them, α is the proportion of the recruiter's hand-holding rate, β is the proportion of the handshake rate, α+β=1, and the β value is greater than the α value. Because the handshake rate data is more sparse, and the handshake rate can better reflect the two-way satisfaction, the β value is greater than the α value.

[0058] According to one embodiment of the present application, extracting talent information feature codes from talent resumes includes: extracting relevant data of target feature names from the resumes; parsing relevant data of target feature names and configuring corresponding feature values; and combining multiple feature values ​​into talent information feature codes according to preset rules.

[0059] This application can use natural language processing (NLP) technology to scan the resume text and identify the content related to these target feature names. For example, for the feature name of "education background", the system will extract relevant data such as the school name, major, degree, and study start and end dates in the resume. After extracting the data related to the target feature name, the data will be parsed, and the corresponding feature values ​​will be identified and configured. The feature value is a standardized expression of the information corresponding to the feature name. For example: for the degree information in "education background", the system will parse the applicant's highest degree, such as the standardized feature value of "undergraduate" configuration is "4", the standardized feature value of "master" configuration is "5" or the standardized feature value of "doctor" configuration is "6". After parsing and configuring the feature value, the system combines multiple feature values ​​into a talent information feature code according to preset rules. The code is a compressed representation of the applicant's key information, which is used for subsequent analysis and hierarchical processing.

[0060] According to one embodiment of the present application, the resume quality score Q is calculated using one or more of the following features in the resume: whether the content in the resume is too little; whether the sentences in the resume are fluent; whether there are garbled characters in the resume; whether the resume contains work experience; whether the resume contains educational experience, wherein different features are assigned different weight values.

[0061] The resume quality score Q can be calculated using the following formula:

[0062] ResumeQualityScore=w1*f1+…+w n *f n

[0063] Among them, w1, w2, … w n is the weight of each feature, f1, f2, ... f n is the score of each feature.

[0064] For example, whether the content in the resume is too small (f1): Check the total number of words or total information in the resume. If the content of the resume is too small (for example, below the preset threshold), the score of this feature is low; whether the sentences in the resume are fluent (f2): Use natural language processing (NLP) technology to analyze the sentence structure and grammar in the resume. If it is detected that the sentence is not fluent or there are obvious grammatical errors, the score of this feature will be reduced; whether there is garbled code (f3): Check whether there are garbled characters or unparseable text in the resume content. If there is garbled code, the score of this feature is low, etc.

[0065] Once the resume quality score is calculated, the system compares the score with a preset threshold. The threshold is set by the recruitment platform based on specific business needs and quality standards. If the resume quality score is lower than the threshold, the system marks it as a poor quality resume. The purpose of marking poor quality resumes is to improve the efficiency of talent classification and reduce the time wasted by the model on low-quality resumes.

[0066] The above describes the implementation methods and advantages of the embodiments of the present application through multiple embodiments. This application describes the specific processing process of the embodiments of the present application in detail with reference to specific examples.

[0067] Figure 3 This is a flow chart of the method for determining talent stratification based on rules. Figure 3 As shown, the method for determining talent stratification based on rules includes:

[0068] First, obtain the resumes of talents, which may come from recruitment platforms or other resume databases;

[0069] Then, relevant tags are extracted from the talent resume. These tags include information such as the candidate's gender, age, graduation school, education, years of work experience, etc. This information is a preliminary screening of the candidate's basic situation.

[0070] Then, according to the specific needs of each function, a set of talent stratification rules is designed for each functional position (such as algorithm engineer). The system will use these rules to analyze and judge the extracted labels. Taking the algorithm engineer in the figure as an example, the rules may include:

[0071] Bachelor degree or above;

[0072] Graduated from a 211 or higher school;

[0073] More than 2 years of working experience;

[0074] Have working experience in large Internet companies.

[0075] Finally, the talent resumes are stratified according to preset rules and labels in the talent resumes. Talent stratification can be divided into selected talents and ordinary talents.

[0076] This application uses a pre-trained large model to perform talent stratification. The model is trained based on the behavioral data of recruiters and applicants, and the user's talent resume information is input into the model to output the talent label. The specific implementation of this application is mainly divided into two parts: one: offline model training; two: online talent stratification. The following is a detailed introduction to these two parts.

[0077] Figure 4 1 is a flow chart of a method for training a talent stratification model according to an embodiment of the present application. Figure 4 As shown, this application takes the screening of selected talents as an example to introduce the process of talent stratification model training.

[0078] Hierarchical label definition

[0079] Hierarchical label definition, namely, the definition of selected talent labels: In the solution of this application, labels are divided into two categories, 0 represents non-selected talents; 1 represents selected talents. Label classification is performed based on the excellent quantitative value A, and the excellent quantitative value A is expressed using the following formula:

[0080] Excellent quantitative value A = α × recruiter's hand-reaching rate + β × handshake rate;

[0081] Among them, α is the proportion of recruiters' hand-reaching rate, β is the proportion of handshake rate, α+β=1, the recruiters' hand-reaching rate is the ratio of the number of recruiters' hands reaching out to the number of resumes exposed, and the handshake rate is the ratio of the number of handshakes to the number of resumes submitted.

[0082] Sample range: Select job seekers who have been active within 3 months and have registered for more than 1 month.

[0083] Positive samples (label=1): the excellent quantitative values ​​A are sorted from high to low, and the top 30% of the highest ranking scores and the samples whose reaching rate and handshake rate meet the threshold (the threshold is determined according to the actual business) are taken as positive samples.

[0084] Negative samples (label = 0): The excellent quantitative values ​​A are sorted from high to low, and the last 30% of the resumes with low ranking scores are selected as negative samples.

[0085] Feature Selection

[0086] Extract static features within the sample range, namely, talent information feature coding. Static features are mainly extracted from job seekers' resumes. Some of the extracted features are shown in Table 1 below:

[0087] Table 1

[0088]

[0089]

[0090] Sample generation

[0091] Bind the static feature information of job seekers with hierarchical labels to generate sample data, as shown in Table 2:

[0092] Table 2

[0093]

[0094] Model Training

[0095] The sample data is input into the machine learning model for training. This application selects the GBDT model. GBDT is an iterative decision tree algorithm with efficient training and strong interpretability.

[0096] First, initialize the GBDT model. Configure the hyperparameters of the GBDT model, such as the number of trees (n_estimators), the depth of each tree (max_depth), the learning rate (learning_rate), etc. These parameters need to be adjusted according to the actual situation to obtain the best model performance.

[0097] Then, the sample data is used for model training. The model will train a series of decision trees in a step-by-step iterative manner, and each tree will fit the prediction residual of the previous tree, thereby gradually reducing the error.

[0098] Finally, model evaluation. The performance of the model is evaluated by making predictions on the test set. Indicators such as accuracy, recall, and F1-score can be used to measure the classification effect of the model. If the performance of the model does not meet expectations, the model can be improved by adjusting hyperparameters, reselecting features, or using more training data until the accuracy of the model's stratification results meets the application requirements.

[0099] Figure 5 1 is a flow chart of a method for online talent stratification according to an embodiment of the present application. Figure 5 As shown, online talent methods include:

[0100] (1) Resume information: Enter resume information. If it is for a new user, who only has resume information but no behavioral data, the talent stratification model can still be used for talent stratification.

[0101] (2) Filtering of poor-quality resumes: The resume quality score Q is determined based on the resume quality scoring system of the recruitment platform. If the score is lower than the threshold (determined according to business requirements), it is considered a poor-quality resume. The quality score is obtained through linear weighting, and the weight can be adjusted according to the actual application. The main features are: 1. Whether the content is too little; 2. Whether the sentence is fluent; 3. Whether there is garbled code; 4. Whether there is work experience (non-fresh graduates); 5. Whether the educational experience is filled in, etc. The resume quality score Q can be calculated using the following formula:

[0102] ResumeQualityScore=w1*f1+…+w n *f n

[0103] Among them, w1, w2, … w n is the weight of each feature, f1, f2, ... f n is the score of each feature.

[0104] (3) Feature extraction: Extract features from the filtered resumes and generate prediction samples. The online feature extraction and sample construction are consistent with the offline part. The final format of the prediction sample is the same as the sample data in Table 2.

[0105] (4) Model stratification: Based on the prediction samples constructed in the previous step, the trained talent stratification model (such as the GBDT model) is used to determine whether the candidate is a selected talent.

[0106] (5) Output results: Output results to the database. According to the output hierarchical results, the talent resumes are divided into multiple resume sets. Based on the hierarchical multiple resumes, the online recruitment platform will provide different business strategies such as search and recommendation, so as to achieve efficient distribution of platform traffic and traffic monetization. For example, recruiters with relatively strong technical and financial strength hope to attract outstanding talents, and outstanding talents also hope to work in strong companies. Therefore, the resumes are collected as resumes of selected talents and recommended to strong companies according to recruitment needs. The recommended results are more in line with the needs and expectations of users (recruiters or job seekers), which will increase the hand-holding rate and handshake rate, and improve the user satisfaction of job seekers and recruiters.

[0107] Compared with the results of talent stratification based on rules, talent stratification based on talent stratification models can improve the reaching-out rate and handshake rate in recommendation effects.

[0108] Experimental comparison of rule-based stratification and model-based stratification

[0109] Control group: No stratification of talents was performed, and talents were recommended only based on the recruitment needs of users (recruiters);

[0110] Experimental Group 1: Talent stratification is performed using rules. The talent stratification is performed into multiple resume sets according to the stratification labels. Talents are recommended based on the user's needs in the resume sets associated with the user (recruiter).

[0111] Experimental Group 2: Talent stratification is performed using the talent stratification model of this application, and the resumes are divided into multiple resume sets according to the stratification labels. In the resume sets associated with the user (recruiter), talents are recommended according to the user's needs.

[0112] In the recommendation system, different AB experiments are conducted on each group of experimental users. When the number of people in the experimental group reaches the minimum sample size, the experiment is ended and a statistical evaluation is performed on the effects of each group.

[0113] The experimental results are shown in Table 3 below:

[0114] Table 3

[0115]

[0116]

[0117] Experimental conclusion: By stratifying talents, the reaching rate is significantly improved. In Experiment 1, the reaching rate is increased by 36.4% compared with the control group, in Experiment 2, the reaching rate is increased by 100% compared with the control group, and in Experiment 2, the reaching rate is increased by 46.7% compared with Experiment 1. Therefore, using the talent stratification model to stratify talents can significantly improve the reaching rate. In addition, there is no need to manually set the stratification rules, which improves the stratification efficiency.

[0118] Corresponding to the method embodiment of the present application, the present application also provides a talent recommendation device based on a talent stratification model, such as Figure 6 As shown, the talent recommendation device 100 based on the talent stratification model includes:

[0119] A feature extraction module 610 is used to obtain a resume of a talent in a resume database and extract a talent information feature code, wherein the talent information feature code includes one or more of a basic information feature value, an education information feature value, and a work experience feature value;

[0120] The talent stratification module 620 is used to input the talent information feature code into the pre-trained talent stratification model and output the stratification label of the talent resume. The talent stratification model is trained using a plurality of sample data. The sample data includes the talent information feature code and the stratification label. The stratification label is determined based on the application behavior data of the applicant and the recruitment behavior data of the applicant by the recruiter.

[0121] A resume division module 630, configured to divide the resumes of talents into multiple resume sets according to the hierarchical labels;

[0122] The recommendation module 640 is used to screen out the resumes of talents that meet the requirements of the target user from the resume collection associated with the target user, and recommend them to the target user.

[0123] Figure 7 1 is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. The electronic device can be implemented as a server or other various terminal devices, such as a desktop personal computer, a tablet computer, a laptop computer, a mobile phone, etc., including a processor 601 and a memory 602. The memory 602 stores a program instruction set. When the processor 601 executes the program instruction set on the memory 602, any of the aforementioned recommendation information generation methods is implemented.

[0124] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present invention.

[0125] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 602 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 602 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 602 is a non-volatile solid-state memory.

[0126] The memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the recommendation information generation method provided by the present invention.

[0127] In one example, the electronic device may further include a communication interface 603 and a bus 604. The processor 601, the memory 602, and the communication interface 603 are connected via the bus 604 and communicate with each other.

[0128] The communication interface 603 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiment of the present invention.

[0129] Bus 604 includes hardware, software or both, and the components of online data flow billing equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 604 may include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0130] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement any of the talent recommendation methods based on the talent stratification model in the aforementioned embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in combination with an instruction execution system, device, and equipment. The storage medium can be a transient computer-readable storage medium or a non-transient computer-readable storage medium. Non-transient computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical disks based on CD, DVD, or Blu-ray technology, and persistent solid-state memories such as flash memory, solid-state drives, and the like.

[0131] The present invention also provides a computer program product, which includes a set of computer program instructions, and when the computer program instruction set is executed by a processor, it implements any talent recommendation method based on a talent stratification model in the aforementioned embodiments. The computer program product includes but is not limited to application installation packages published on websites and application stores, application plug-ins, and small programs that can be run in certain applications.

[0132] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0133] The above embodiments are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the scope of the present invention. Therefore, all equivalent technical solutions should also fall within the scope of the present invention.

Claims

1. A talent recommendation method based on a talent stratification model, comprising: Obtaining a talent resume and extracting a talent information feature code, wherein the talent information feature code includes one or more of a basic information feature value, an education information feature value, and a work experience feature value; Inputting the talent information feature code into a pre-trained talent stratification model, and outputting a stratification label of the talent resume, wherein the talent stratification model is trained using a plurality of sample data, wherein the sample data includes the talent information feature code and the stratification label, and the stratification label is determined based on the applicant's application behavior data and the recruiter's recruitment behavior data on the applicant; Dividing the talent resumes into multiple resume sets according to the hierarchical labels; Filter out the resumes of talents that meet the requirements of the target user from the resume collection associated with the target user and recommend them to the target user.

2. The talent recommendation method according to claim 1, wherein the hierarchical labels are determined based on the applicant's application behavior data and the recruiter's recruitment behavior data of the applicant, including: Calculate the recruiter's hand-reaching rate and / or handshake rate based on the applicant's application behavior data and the recruiter's recruitment behavior data of the applicant within a predetermined time period, wherein the hand-reaching rate is used to indicate the recruiter's satisfaction with the resume, and the handshake rate is used to indicate the satisfaction of both the recruiter and the applicant; The hierarchical labels of the applicants are determined according to the hand reaching rate and / or handshake rate of the recruiters, and the hierarchical labels include a first hierarchical label and a second hierarchical label.

3. The talent recommendation method according to claim 2, further comprising: The excellent quantitative value A is calculated based on the recruiter's hand-reaching rate and handshake rate; Sorting according to the size of the excellent quantization value A; Determine the stratification labels of the top 10%-40% of applicants as the first stratification labels; The stratification labels of the last 10-40% of the applicants are determined as the second stratification labels.

4. According to the talent recommendation method of claim 3, the excellent quantitative value A is expressed by the following formula: Excellent quantitative value A = α × recruiter's hand-reaching rate + β × handshake rate; in, α is the proportion of recruiters' hand-reaching rate, β is the proportion of handshake rate, α+β=1, the recruiters' hand-reaching rate is the ratio of the number of recruiters' hands reaching out to the number of resumes exposed, and the handshake rate is the ratio of the number of handshakes to the number of resumes submitted.

5. According to the talent recommendation method of claim 1, extracting the talent information feature code from the talent resume comprises: Extract relevant data of target feature name from resume; Parse the relevant data of the target feature name and configure the corresponding feature value; as well as Combine multiple feature values ​​into talent information feature codes according to preset rules.

6. The talent recommendation method according to claim 1, further comprising: Obtain talent resumes and filter out low-quality resumes from the talent resumes based on the resume quality score Q.

7. The talent recommendation method according to claim 6, wherein the resume quality score Q is calculated using one or more of the following features in the resume: Is there too little content in the resume? Whether the sentences in the resume are fluent; Is there any garbled characters in the resume? Whether to fill in the work experience in the resume; Whether to fill in the educational experience in the resume, in, Different features are assigned different weights.

8. A talent recommendation device based on a talent stratification model, comprising: A feature extraction module, used to obtain a talent resume and extract a talent information feature code, wherein the talent information feature code includes one or more of a basic information feature value, an education information feature value, and a work experience feature value; A talent stratification module, used for inputting the talent information feature code into a pre-trained talent stratification model, and outputting a stratification label of the talent resume, wherein the talent stratification model is trained using a plurality of sample data, wherein the sample data includes the talent information feature code and the stratification label, and the stratification label is determined based on the applicant's application behavior data and the recruiter's recruitment behavior data on the applicant; A resume division module, used to divide the talent resumes into multiple resume sets according to the hierarchical labels; The recommendation module is used to filter out the resumes that meet the requirements of the target user from the resume collection associated with the target user and recommend them to the target user.

9. An electronic device comprising a processor and a memory, wherein a computer program instruction set is stored in the memory, and when the processor executes the computer program instruction set in the memory, the talent recommendation method based on the talent stratification model described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program instruction set, and when the computer program instruction set is executed by a processor, the talent recommendation method based on the talent stratification model described in any one of claims 1 to 7 is implemented.

Images