A human resource enhancement recommendation method and device based on explicit feedback

Abstract

The application discloses a human resource enhancement recommendation method and device based on explicit feedback, and relates to the online recruitment technical field. The method comprises the following steps: obtaining target post information, resume text information delivered for the target post, and explicit feedback information of the recruitment party to the target post delivered resume; extracting resume abstract information and recruitment demand information, and extracting screening conditions from the recruitment demand information; generating a plurality of virtual resumes, performing vectorization processing on the plurality of virtual resumes, and generating a plurality of virtual resume vectors; querying real resume vectors matched with the virtual resume vectors from a real resume vector library related to the target post information as preliminary candidates; and filtering and sorting the preliminary candidates to obtain recommended candidates. The application greatly improves the human-post matching precision, and significantly improves the core indicators and user experience of the human resource recommendation platform.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online recruitment, in particular to a human resource enhanced recommendation method and device based on explicit feedback. BACKGROUND

[0002] User feedback in the online recruitment recommendation scenario includes the following two types of core data:

[0003] Type one: implicit feedback. Usually user behavior signals, including positive feedback behavior signals (such as: click / browse / view detail page, collect, dwell time, etc.), negative feedback signals (such as: not view, mark no longer recommended, etc.);

[0004] Type two: explicit feedback. Usually user-initiated text comments, such as job seekers marking "not considering positions in this industry", recruiters noting "too high salary", "lack of post-training experience with large models", etc.

[0005] Existing online recruitment recommendation systems have relatively maturely utilized implicit feedback, but the processing of explicit feedback must rely on deep semantic understanding and real-time calculation, otherwise the recommendation results cannot fully mine the explicit feedback information, for example:

[0006] (1) Unable to handle deep semantic understanding. The explicit feedback information of the recruiter is "need to have post-processing experience with large models", but it cannot be matched to candidates who master SFT, LoRA, DPO, RFT, etc.

[0007] (2) Unable to handle negative semantic understanding. The explicit feedback information of the recruiter is "not considering the Internet industry", which may recommend candidates from the Internet industry because of the keyword "Internet", i.e. ignoring the negative semantics of "not considering";

[0008] (3) Unable to handle associated semantic understanding. The explicit feedback information of the recruiter is "expecting too high salary", but cannot obtain the specific salary range of the recruiter;

[0009] (4) Unable to handle constraint degree semantic understanding of restriction conditions. The explicit feedback information of the recruiter contains constraint degree semantics such as "priority", "not considering", "only considering", etc., but the "priority" constraint recruitment condition should not be excluded as a restriction condition to exclude candidates who do not meet the priority condition.

[0010] Insufficient understanding and utilization of explicit feedback information will lead to subsequent recommendation results that do not take into account the explicit feedback of the user, indirectly leading to users questioning the effectiveness of the "explicit feedback" function, and ultimately leading to users abandoning explicit feedback. Such a vicious cycle forces the industry recommendation system to downgrade to a recommendation strategy dominated by implicit feedback.

[0011] Therefore, it is urgent to introduce an explicit feedback enhanced recommendation system scheme to fully mine the explicit feedback information of users to improve the matching degree of recommendation results and user experience. SUMMARY

[0012] In view of the above defects or deficiencies in the prior art, the present application provides a human resource enhanced recommendation method and device based on explicit feedback to solve the technical problems mentioned in the background art.

[0013] In one aspect of the present application, a human resource enhanced recommendation method based on explicit feedback is provided, comprising:

[0014] obtaining target post information, resume text information submitted for the target post, and explicit feedback information of the recruitment party on the submitted resumes for the target post;

[0015] extracting resume abstract information from the resume text information, generating recruitment demand information according to the resume text information and the explicit feedback information, extracting screening conditions from the recruitment demand information, adding the extracted screening conditions to a screening condition list, generating a plurality of virtual resumes according to the target post information, the resume abstract information, and the recruitment demand information, and performing vectorization processing on the plurality of virtual resumes to generate a plurality of virtual resume vectors;

[0016] querying real resume vectors matching the virtual resume vectors from a real resume vector library related to the target post information, and taking job seekers corresponding to the queried real resume vectors as preliminary candidates;

[0017] filtering out candidates who do not meet any screening condition in the screening condition list from the preliminary candidates to obtain recommended candidates, calculating matching features of each recommended candidate and the target post information, calculating a comprehensive matching degree of each recommended candidate and the target post information based on the matching features, sorting and outputting the recommended candidates according to the comprehensive matching degree, and obtaining a recommended candidate list.

[0018] In another aspect of the present application, a human resource enhanced recommendation device based on explicit feedback is also provided, comprising:

[0019] A first module is configured to obtain target post information, resume text information submitted for the target post, and explicit feedback information of the recruitment party on the submitted resumes for the target post;

[0020] The second module is configured to extract resume abstract information from the resume text information, generate recruitment demand information according to the resume text information and the explicit feedback information, extract screening conditions from the recruitment demand information, and add the extracted screening conditions to a screening condition list; and generate a plurality of virtual resumes according to the target post information, the resume abstract information and the recruitment demand information, and perform vectorization processing on the plurality of virtual resumes to generate a plurality of virtual resume vectors.

[0021] The third module is configured to query real resume vectors matching the virtual resume vectors from a real resume vector library related to the target post information, and take job seekers corresponding to the queried real resume vectors as preliminary candidates.

[0022] The fourth module is configured to filter out people not satisfying any screening condition in the screening condition list from the preliminary candidates to obtain recommended candidates, calculate a matching feature of each recommended candidate and the target post information, calculate a comprehensive matching degree of each recommended candidate and the target post information based on the matching feature, sort and output the recommended candidates according to the comprehensive matching degree, and obtain a recommended candidate list.

[0023] The human resource enhanced recommendation method and device based on explicit feedback provided by the application significantly improve the depth understanding and utilization of explicit feedback information of a user by a human resource recommendation system, break through the semantic gap through offline generation of virtual resumes and online vector retrieval technology, greatly improve the human-post matching accuracy, and significantly improve the core indicators and user experience of a human resource recommendation platform. BRIEF DESCRIPTION OF DRAWINGS

[0024] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the attached drawings:

[0025] Figure 1 is a system logic diagram of the human resource enhanced recommendation method based on explicit feedback provided by an embodiment of the application;

[0026] Figure 2 is a flowchart of the human resource enhanced recommendation method based on explicit feedback provided by an embodiment of the application;

[0027] Figure 3 is a flowchart of offline user portrait updating provided by an embodiment of the application;

[0028] Figure 4 is a structure diagram of the human resource enhanced recommendation device based on explicit feedback provided by an embodiment of the application. DETAILED DESCRIPTION

[0029] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of the present application.

[0030] The terms used in the embodiments of the present application are only for the purpose of describing the specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0031] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present application to describe the acquisition modules, these acquisition modules should not be limited to these terms. These terms are only used to distinguish the acquisition modules from each other.

[0032] Referring to Figure 1 The human resource enhancement recommendation method based on explicit feedback of the present application deeply analyzes the user explicit feedback text, generates structured screening conditions and virtual resume vectors offline, adds semantic recall links based on virtual resumes online, and improves the utilization accuracy and matching degree of the recommendation system for explicit feedback in two stages.

[0033] The method can be preferably divided into two parts of offline stage and online stage:

[0034] Offline stage: the user feedback understanding module deeply understands the user explicit feedback (which can also include implicit feedback) to continuously update the user portrait. The update of the user portrait can be decoupled from the online service link of the recommendation system and executed asynchronously at regular intervals. Different update frequencies can also be adopted according to the importance of the user feedback. Important user feedback (for example: user expects to change) is triggered by user feedback behavior and updated in time, and user feedback of less importance can be updated regularly.

[0035] Online stage: the updated user portrait of the online application optimizes the recommendation effect. Specifically, based on the user portrait, four stages of recall, filtering, sorting, and rearrangement are performed. Among them, the recall stage is to screen out a candidate set of ten thousand levels from a candidate pool of one hundred million levels. The filtering stage is to screen out a candidate set of one thousand levels from the candidate set of ten thousand levels that meet the filtering conditions. The sorting stage is to estimate and sort the matching degree of the candidate set of one thousand levels to select a candidate set of one hundred levels. The rearrangement stage is to generate the final recommendation list considering factors such as list diversity. After the recommendation result list after the four stages is displayed to the user, the explicit feedback of the user is collected through the product function, and the implicit feedback of the user behavior is recorded through data burying (optional). All collected user explicit feedback (which can also include implicit feedback) is recorded in the user feedback database for subsequent offline user portrait update.

[0036] Referring to Figure 2 The method of the embodiment includes steps S101-S105. To ensure the response speed of the human resource recommendation platform, steps S101-S103 are user portrait updating steps, preferably implemented in an offline stage, and steps S104-S105 are steps of recall and sorting according to the latest user portrait, preferably implemented in an online stage.

[0037] The steps of the method will be described in detail below. The method is for the scenario of recommending candidates to employers, and the explicit feedback comes from employers.

[0038] Step S101, obtaining target post information, resume text information delivered for the target post, and explicit feedback information of the employer on the delivered resume for the target post.

[0039] Specifically, the target post information, the resume text information delivered by the job seeker for the target post, and the explicit feedback information of the employer on the delivered resume for the target post are obtained through the human resource recommendation system platform. It should be pointed out that the embodiment can also obtain explicit feedback information and implicit feedback information at the same time, but the key point of the embodiment is the use of deep semantic understanding of explicit feedback information.

[0040] The explicit feedback information in the method includes text comments actively input by the user, such as the employer's remarks of “too high salary” and “lack of artificial intelligence debugging experience”. The implicit feedback information in the method includes user behavior signals, such as positive feedback behavior signals of click / view / detail page, collection, and dwell time, or negative feedback behavior signals of not viewing and marking no longer recommended.

[0041] Step S102, extract resume summary information from the resume text information; generate recruitment demand information according to the resume text information and the explicit feedback information, extract the screening conditions from the recruitment demand information, and add the extracted screening conditions to the screening condition list; generate a plurality of virtual resumes according to the target post information, the resume summary information and the recruitment demand information, and perform vectorization processing on the plurality of virtual resumes to generate a plurality of virtual resume vectors.

[0042] This step is used to realize the update of the user portrait, which includes two parts of the screening condition list and the virtual resume list.

[0043] The screening condition is a restriction condition that must be met by the human resource recommendation platform to recommend candidates to the recruitment party, for example: the candidate that the target post hopes to find needs to meet the condition of "expected annual salary less than or equal to 100,000 yuan", which can be decomposed into multiple filtering conditions such as { "param": "expected salary", "op": "<=", "value": 100000}.

[0044] The virtual resume is a virtual resume that meets the recruitment demand generated by a text generation model (for example: a large language model) according to the target post information, the resume summary information and the recruitment demand information. It does not exist in reality, but describes a target resume example that meets the recruitment demand. The virtual resume plays an important role in this method, which converts the user's vague explicit feedback information into a text description of an ideal candidate containing specific information or quantitative values, thereby breaking through the limitations of keyword matching.

[0045] Referring to Figure 3 , the specific execution process of step S102 is as follows (the serial number is only used to distinguish the steps, and the order of the steps is not limited) :

[0046] (1) Generate a resume summary

[0047] Obtain real resume text data for the target post from the user feedback data according to the resume content field, and obtain target post information from the user feedback data according to the post ID field. Call the resume text summary generation module to generate resume summary information according to the resume text information, which is used to assist in generating virtual resumes in subsequent steps.

[0048] For example, a general large language model or a fine-tuned large model is used to generate a resume summary in combination with a prompt engineering:

[0049] "Prompt words: You are a professional resume analyst. Please generate a concise summary text according to the full text of the user-provided resume. The requirements are as follows:

[0050] = 1 \* GB3 ① Core element extraction

[0051] Must-have information: Professional identity (e.g., "Full-stack engineer")

[0052] Key qualifications: Highest education + prestigious school / experience (e.g., "Beida computer master, former Tencent architect")

[0053] Core skills: Select 3 most relevant technical stacks / abilities (e.g., "Master React and SpringCloud")

[0054] Career highlights: 1 quantitative achievement (e.g., "Led project with 2 million users")

[0055] = 2 \* GB3 ② Content organization logic

[0056] "Identity positioning" -> "Core qualifications" -> "Technical / ability labels" -> "Career highlights" -> "Job positioning"

[0057] Example structure:

[0058] "[Identity]. [Education / prestigious background]. [3 core skill labels]. Achieved [quantitative achievement]. [Job direction / value proposition]"

[0059] = 3 \* GB3 ③ Strict constraints

[0060] Word limit of no more than 400 words (must be strictly truncated)

[0061] Prohibit bullet points, points, tables, and other formats

[0062] Reject subjective evaluations (e.g., "excellent candidate")

[0063] Maintain third-person objective narrative

[0064] = 4 \* GB3 ④ Special handling principles

[0065] Ambiguous information: Take the highest level of expression (e.g., take the highest education)

[0066] No quantitative achievements: Change to describe core project roles

[0067] Information conflict: Prefer to adopt the "work experience" chapter.

[0068] Input: Full text of the resume.

[0069] (2) Generate recruitment demand information

[0070] If the user feedback data is explicit feedback data, the recruitment demand generation module (such as a fine-tuned large language model) is called to generate recruitment demand information based on the resume text information and explicit feedback information. The recruitment demand information in this method is a further clarification of the ambiguous explicit feedback information and a supplement to the recruitment position. It contains more specific information or quantitative values than explicit feedback information. For example: the explicit feedback information is "salary requirement is too high", and the recruitment demand information can be more specific "annual salary below 100000 yuan". Since part of the explicit feedback information needs to be combined with the specific candidate resume content to obtain complete quantitative information, the recruitment demand information must rely on the resume text information and explicit feedback information to generate jointly.

[0071] For example, a general large language model or a fine-tuned large model is used to generate recruitment demand information in combination with prompt engineering:

[0072] "Prompt words: You are a recruitment demand analyst, responsible for generating independent recruitment demands based on user feedback and resume information as a supplement to the recruitment position. The output demand must be concise, complete, and independent of external context.

[0073] Input:

[0074] `cv field`: resume full text string, containing candidate information (such as job expectations, salary range, industry preferences, skills, etc.). If it is empty or invalid, ignore the resume part.

[0075] `feedback field`: user feedback content string, indicating comments or problem points on the candidate (such as "expecting salary too high" or "industry mismatch").

[0076] Output:

[0077] `job_requirement field`: recruitment demand string, format is complete, independent phrase or sentence (such as "candidate expects annual salary not to exceed 6w").

[0078] Requirements:

[0079] Self-contained: The demand can be directly used for JD without additional explanation.

[0080] Supplement: As a supplement to the JD, focus on the limitations or screening conditions in the feedback.

[0081] Concise: Up to 20 words, avoid verbosity.

[0082] Neutrality: Use objective language (such as "not considering the internet industry" instead of "candidate industry not suitable").

[0083] (3) Extract screening conditions

[0084] Extract filtering conditions from the recruitment demand information, such as: { "param": "expected salary", "op": "<=", "value": 100000}, and add the extracted filtering conditions to the current filtering condition list to update the filtering condition list.

[0085] For example, a general large language model or a fine-tuned large model can be used, combined with prompting engineering to extract filtering conditions:

[0086] "enter:

[0087] Job requirements job_requirement_list= "Candidates expect an annual salary of no more than 60,000 RMB. No internet or finance industries considered. Priority given to candidates in Haidian District. Expected annual salary not exceeding 50,000 RMB. Priority given to candidates with experience in express delivery sorting. Age not lower than 30 years old."

[0088] Output:

[0089] The feedback_filter_rule_list is a list of filter rules extracted from user feedback.

[0090] { "param": "Salary", "op": "<=", "value": 50000},

[0091] { "param": "Industry", "op": "!=", "value": "Internet"},

[0092] { "param": "Industry", "op": "!=", "value": "Finance"}

[0093] ]”

[0094] (4) Generate a virtual resume

[0095] Multiple virtual resumes are generated based on the target job information, resume summary information, and recruitment requirements. These virtual resumes are then vectorized to generate virtual resume vectors. These virtual resume vectors are used in the online phase to perform vector retrieval on the real resume vectors to obtain candidates corresponding to real resumes that match the virtual resumes.

[0096] For example, a virtual resume can be generated by using a general large language model or a fine-tuned large model, combined with prompting engineering:

[0097] "Role:

[0098] You are a senior recruitment consultant. You need to create virtual resumes that are precisely matched to the target positions based on the job description, the information of the candidates who intend to apply for the positions, and the recruitment requirements.

[0099] Generation requirements:

[0100] 1. Job Matching: In-depth analysis of the core requirements of the job description (salary range / skills / experience, etc.)

[0101] 2. Soft skills: Summarize and refer to the resumes of candidates who are interested in the job posting.

[0102] 3. Mandatory Requirements: 100% fulfillment of all recruitment requirements outlined in the job posting.

[0103] 4. Output format: No more than 400 words; plain text description, disallowing bullet points, periods, tables, and other formatting.

[0104] This step ultimately yields offline all user profile data related to the target position, thus preparing for subsequent online retrieval and matching of real resumes.

[0105] Step S103: Query the real resume vector that matches the virtual resume vector from the real resume vector library related to the target job information, and take the job seekers corresponding to the queried real resume vectors as preliminary candidates.

[0106] Specifically, the HR system retrieves, filters, sorts, and rearranges real resumes based on user profile data. These four stages form the final recommendation list, which is then displayed to recruiters. Implicit user feedback is recorded through data tracking, while explicit user feedback is collected through product features. All collected implicit and explicit user feedback is stored in a user feedback database for subsequent offline user profile updates.

[0107] Virtual resume vectors are used in the online recall phase. The recall phase involves filtering tens of thousands of candidates from a pool of hundreds of millions of candidates. The preferred approach is to use a multi-path recall strategy fusion scheme. Each recall strategy is based on the user profile of the job posting and selects a subset of candidates from the pool of hundreds of millions of job seekers.

[0108] Recall strategies often rely on job descriptions / content profiles or behavioral information / behavioral profiles. For example:

[0109] Job Content Keyword Retrieval: Analyze the hard requirements in the job content (such as: bachelor's degree, preferred work location is Beijing, experience in "LoRA fine-tuning"), and based on these hard requirements, call a search engine based on inverted index to find a list of job seekers who meet the requirements.

[0110] Job content vector recall: Use the double tower model based on the job content vector to call the vector retrieval system based on vector index to find the list of job seekers matching the job content vector.

[0111] Collaborative filtering recall: Use platform behavior data to calculate the similarity between job postings (such as using the similarity of candidate lists associated with two job postings), and recommend candidates that similar job postings have contacted to the current job posting.

[0112] The embodiment further increases the use of feedback semantic recall strategy based on the above recall strategy, that is, according to the virtual resume vector, the real resume matching the virtual resume vector is obtained through vector retrieval, and the corresponding job seeker is determined through the real resume to be the preliminary candidate. Since the virtual resume integrates the target job information, the resume abstract information and the recruitment demand information, the candidate set that matches the recruitment demand can be obtained by searching for real resumes similar to the virtual resume. In addition, since the virtual resume vector is generated offline, the online service stage of the feedback semantic recall only needs to call several times of vector retrieval, and the overall service delay will not be significantly increased, so as to ensure the real-time performance and user experience of the recommendation system.

[0113] Step S104, filtering out the person who does not meet any screening condition in the screening condition list from the preliminary candidate to obtain the recommended candidate, calculating the matching feature of each recommended candidate and the target job information, calculating the comprehensive matching degree of each recommended candidate and the target job information based on the matching feature, sorting and outputting the recommended candidate according to the comprehensive matching degree to obtain the recommended candidate list.

[0114] After the recall stage, the filtering stage is entered. The purpose of the filtering stage is to filter out the thousand-level candidate set that meets the filtering condition from the ten-thousand-level candidate set, so as to eliminate the candidate who does not meet the screening condition. Specifically, the screening condition list of the user portrait is read, and whether each candidate in the preliminary candidate meets all the screening conditions in the screening condition list is checked. If any screening condition in the screening condition list does not meet, the candidate is filtered out from the preliminary candidate list.

[0115] After the filtering stage, the sorting stage is entered. The purpose of the sorting stage is to estimate and sort the matching degree of the thousand-level candidate set, and to select the hundred-level candidate set, that is, to evaluate the matching degree of the candidate after the filtering stage with the job posting one by one, and to sort according to the matching degree from high to low.

[0116] Exemplarily, the matching features of the candidate and the target post are calculated, including but not limited to: work location matching degree, salary expectation matching degree, skill matching degree, industry experience matching degree, and the like; based on the matching features, a ranking model is used to calculate the comprehensive matching degree of the candidate and the target post. The ranking model includes but is not limited to traditional statistical learning model, deep learning model, multi-objective optimization, sequence modeling, and the like; the recommended candidate is ranked according to the obtained comprehensive matching degree, and a recommended candidate list after ranking is output.

[0117] Further, after the ranking stage, a rearrangement stage is entered. The rearrangement stage is to generate a final recommendation list by considering factors such as list diversity in the hundred-level candidate set, that is, to output a final recommendation list according to business rules and user experience. The business rules are rules generally followed by the industry, for example, based on the scattering strategy to avoid too large a proportion of candidates from the same enterprise / school. The user experience refers to the user's own settings and preferences, for example, according to the configured number k of candidates per page of the recommendation list, the current top-k candidates are returned.

[0118] The human resource enhanced recommendation method based on explicit feedback of the embodiment improves the utilization accuracy and matching degree of the recommendation system for explicit feedback through deep analysis of the explicit feedback text of the user, offline generation of structured screening conditions and virtual resume vectors, online semantic recall link based on virtual resumes, and two-stage improvement of the utilization accuracy and matching degree of the recommendation system for explicit feedback, thereby greatly improving the human-post matching accuracy and significantly improving the core indicators and user experience of the human resource recommendation platform.

[0119] Referring to Figure 4 Another embodiment of the present application also provides a human resource enhanced recommendation device 200 based on explicit feedback, which comprises a first module 201, a second module 202, a third module 203, and a fourth module 204, and the human resource enhanced recommendation device 200 based on explicit feedback can execute the human resource enhanced recommendation method based on explicit feedback in the method embodiment.

[0120] Specifically, the human resource enhanced recommendation device 200 based on explicit feedback comprises:

[0121] The first module 201 is configured to obtain target post information, resume text information for a target post, and explicit feedback information of a recruitment party for a target post;

[0122] The second module 202 is configured to extract resume abstract information from the resume text information; generate recruitment demand information based on the resume text information and the explicit feedback information, extract screening conditions from the recruitment demand information, and add the extracted screening conditions to a screening condition list; generate a plurality of virtual resumes based on the target post information, the resume abstract information, and the recruitment demand information, and perform vectorization processing on the plurality of virtual resumes to generate a plurality of virtual resume vectors.

[0123] The third module 203 is configured to query real resume vectors matched with the virtual resume vector from a real resume vector library related to the target post information, and take a job seeker corresponding to the queried real resume vector as a preliminary candidate;

[0124] The fourth module 204 is configured to filter out a person who does not meet any screening condition in the screening condition list from the preliminary candidate to obtain a recommended candidate, calculate a matching feature of each recommended candidate with the target post information, calculate a comprehensive matching degree of each recommended candidate with the target post information based on the matching feature, sort and output the recommended candidate according to the comprehensive matching degree, and obtain a recommended candidate list.

[0125] Further, the second module 202 is further configured to input the resume text information, the explicit feedback information and the prompt word into the fine-tuned general large language model to obtain output recruitment demand information.

[0126] Further, the recruitment demand information includes quantified data related to the resume text information and the explicit feedback information.

[0127] Further, the resume abstract information includes professional identity information, seniority information, skill information and professional highlight information.

[0128] Further, the device further comprises a fifth module configured to rearrange the recommended candidate list according to a preset business rule and a parameter setting of the user on the recommended candidate list.

[0129] It should be noted that the human resource enhancement recommendation device 200 based on explicit feedback provided in the embodiment corresponds to a technical solution that can be used to execute each method embodiment, and the implementation principle and technical effects thereof are similar to those of the method, which will not be described here.

[0130] The above description is only the preferred embodiment of the present application. Those skilled in the art should understand that the disclosed range of the present application is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the disclosed concept. For example, the above features are replaced with the technical features disclosed in the present application (but not limited to) having similar functions to form a technical solution.

Claims

1. A human resource enhancement recommendation method based on explicit feedback, characterized in that, include: Obtain information about the target job, the text information of the resumes submitted for the target job, and the explicit feedback information from the recruiter regarding the resumes submitted for the target job; Extract resume summary information from the resume text information; generate recruitment demand information based on the resume text information and the explicit feedback information; extract screening conditions from the recruitment demand information; and add the extracted screening conditions to the screening condition list. Based on the target job information, the resume summary information, and the recruitment requirements information, several virtual resumes are generated, and the several virtual resumes are vectorized to generate several virtual resume vectors. Query the real resume vectors that match the virtual resume vectors from the real resume vector library related to the target job information, and take the job seekers corresponding to the real resume vectors found as preliminary candidates. Filter out candidates from the preliminary candidates who do not meet any of the screening criteria in the screening criteria list to obtain recommended candidates. Calculate the matching features between each recommended candidate and the target job information. Calculate the comprehensive matching degree between each recommended candidate and the target job information based on the matching features. Sort and output the recommended candidates according to the comprehensive matching degree to obtain a list of recommended candidates.

2. The human resource enhancement recommendation method based on explicit feedback according to claim 1, characterized in that, The steps for generating recruitment demand information based on the resume text information and the explicit feedback information include: The resume text, explicit feedback, and prompts are input into a finely tuned general-purpose large language model to obtain the output recruitment demand information.

3. The human resource enhancement recommendation method based on explicit feedback according to claim 2, characterized in that, The recruitment demand information includes quantitative data related to resume text information and explicit feedback information.

4. The human resource enhancement recommendation method based on explicit feedback according to claim 1, characterized in that, The resume summary information includes professional identity information, qualifications information, skills information, and professional highlights information.

5. The human resource enhancement recommendation method based on explicit feedback according to claim 1, characterized in that, Also includes: The recommended candidate list is rearranged according to preset business rules and user parameter settings for the recommended candidate list.

6. A human resource enhancement recommendation device based on explicit feedback, characterized in that, include: The first module is used to obtain information about the target job, the text information of the resumes submitted for the target job, and the explicit feedback information from the recruiter regarding the resumes submitted for the target job. The second module is used to extract resume summary information from the resume text information; generate recruitment demand information based on the resume text information and the explicit feedback information; extract screening conditions from the recruitment demand information; and add the extracted screening conditions to the screening condition list. Based on the target job information, the resume summary information, and the recruitment requirements information, several virtual resumes are generated, and the several virtual resumes are vectorized to generate several virtual resume vectors. The third module is used to query real resume vectors that match the virtual resume vectors from a real resume vector library related to the target job information, and to take the job seekers corresponding to the real resume vectors found as preliminary candidates. The fourth module is used to filter out candidates who do not meet any of the screening conditions in the screening condition list from the preliminary candidates to obtain recommended candidates, calculate the matching features between each recommended candidate and the target job information, calculate the comprehensive matching degree between each recommended candidate and the target job information based on the matching features, and sort and output the recommended candidates according to the comprehensive matching degree to obtain a list of recommended candidates.

7. The human resource enhancement recommendation device based on explicit feedback according to claim 6, characterized in that, The second module is further used for: The resume text, explicit feedback, and prompts are input into a finely tuned general-purpose large language model to obtain the output recruitment demand information.

8. The human resource enhancement recommendation device based on explicit feedback according to claim 7, characterized in that, The recruitment demand information includes quantitative data related to resume text information and explicit feedback information.

9. A human resource enhancement recommendation device based on explicit feedback according to claim 6, characterized in that, The resume summary information includes professional identity information, qualifications information, skills information, and professional highlights information.

10. A human resource enhancement recommendation device based on explicit feedback according to claim 6, characterized in that, Also includes: The fifth module is used to rearrange the recommended candidate list according to preset business rules and user parameter settings for the recommended candidate list.

Images