LLM-based position description generation method and apparatus, and electronic device

Abstract

The invention discloses an LLM-based position description generation method and apparatus, and an electronic device. The method comprises the steps of obtaining recruitment demand information input by a recruiter; obtaining enterprise portrait attributes of recruiters to obtain an enterprise information set and a function label set; the enterprise information set comprises multiple enterprise information labels and label values under the enterprise information labels; the function label set comprises multiple function labels and label values under the function labels; performing function estimation according to the recruitment demand information to obtain a first function corresponding to the to-be-recruitment post; constructing a position description candidate sample set according to the first function and the enterprise information set; according to the position description candidate sample set and the selection strategy, obtaining a position description target sample; on the basis of a preset prompt template, constructing a position description prompt according to the recruitment demand information, the position description target sample, the function label set and the enterprise information set; and inputting the constructed position description prompt into the LLM model to generate the position description.

Description

Technical Field

[0001] The present application relates to the technical field of computers and the Internet, and in particular to a method, device and electronic device for generating a job description based on LLM. Background Art

[0002] In the corporate recruitment process, whenever a job vacancy needs to be filled, the first task is to carefully write a detailed job description (Job Description, referred to as JD), and then accurately publish it through major recruitment platforms. At present, many recruitment platforms still require HR to manually fill in job descriptions, but HR is not the initiator of job recruitment needs and cannot deeply understand recruitment needs; even if HR uses a large model to generate job descriptions, due to the small amount of input information and the impact of online calls on response time, it is impossible to use a model with large parameters, so the overall quality of the generated job description is poor; in addition, because the information filled in by HR is limited, for example, most users' descriptions are "I have a sales position, requiring a college degree or above and more than 5 years of work experience", which is relatively simple and single, so the job descriptions generated by various companies using large models are not much different. Summary of the invention

[0003] The purpose of this application is to provide a method, device and electronic device for generating job descriptions based on LLM.

[0004] The present application embodiment provides a method for generating a job description based on LLM, including:

[0005] Obtain the recruitment demand information input by the recruiter for the recruitment position;

[0006] Obtain the enterprise portrait attributes of the recruiter to obtain an enterprise information set and a function label set; wherein the enterprise information set includes multiple enterprise information labels and label values ​​under each enterprise information label; the function label set includes multiple function labels and label values ​​under each function label;

[0007] Estimating functions based on recruitment demand information to obtain the first functions corresponding to the positions to be recruited;

[0008] According to the first function and the enterprise information set, a candidate sample set of job descriptions is constructed;

[0009] According to the candidate sample set and selection strategy of job description, the target sample of job description is obtained;

[0010] Based on the preset prompt template, a job description prompt is constructed according to the recruitment demand information, job description target sample, function label set and enterprise information set;

[0011] Input the constructed job description prompt into the LLM model to generate the job description for the position to be recruited.

[0012] Furthermore, the acquiring of the enterprise portrait attributes of the recruiter to obtain the enterprise information set and the function label set includes:

[0013] Get historical job descriptions posted by recruiters;

[0014] The acquired historical job descriptions are screened to obtain core historical job descriptions with functional tags;

[0015] Count the core historical job descriptions to obtain multiple candidate label values ​​under each functional label;

[0016] Calculate the first probability of each candidate label value under each functional label; wherein the first probability is the number of historical job descriptions containing the current candidate label value under the corresponding functional label divided by the total number of historical job descriptions published by the recruiter;

[0017] The first probabilities corresponding to the multiple candidate tag values ​​are sorted to obtain a target tag value; wherein the target tag value is the candidate tag value corresponding to the maximum first probability.

[0018] Furthermore, the step of obtaining a target job description sample based on the job description candidate sample set and the selection strategy includes:

[0019] Obtain the current number of resume submissions and current exposures of job seekers for the job positions published by the job descriptions generated using the current job description candidate samples within a preset time period;

[0020] According to the current number of resume submissions and the current number of exposures, the reward value obtained by the job description generated by the current job description candidate sample under the first function is obtained;

[0021] Get the first total exposure count of the candidate sample for the current job description under the first function;

[0022] Obtain the second total exposure times of the job descriptions generated by all job description candidate samples under the first function;

[0023] A score for each job description candidate sample is obtained according to the reward value, the first total exposure number, and the second total exposure number.

[0024] Furthermore, the step of obtaining a target job description sample based on the job description candidate sample set and the selection strategy further includes:

[0025] Normalizing the score of each job description candidate sample to obtain a normalized score of each job description candidate sample; wherein the sum of the normalized scores of each job description candidate sample is set to 100;

[0026] According to the normalized score of each candidate job description sample, construct each numerical range under each candidate job description sample; wherein each numerical range is connected to each other and is between 0 and 100;

[0027] Get a random number between 0 and 100. If the random number is within the corresponding numerical range, select the corresponding job description candidate sample as a high-quality job description sample.

[0028] Furthermore, the function estimation is performed according to the recruitment demand information to obtain the first function corresponding to the position to be recruited, including:

[0029] Classify the job demand information based on the pre-built multi-classification model to obtain the estimated scores of multiple candidate functions;

[0030] The estimated scores of multiple candidate functions are sorted, and the candidate function corresponding to the maximum estimated score is selected as the first function corresponding to the position to be recruited.

[0031] Furthermore, after inputting the constructed job description prompt into the LLM model to generate the job description of the position to be recruited, the method further includes:

[0032] Receive the recruitment demand information after adjusting the recruitment positions input by the recruiter;

[0033] Based on the constructed job description prompt, the adjusted recruitment demand information is input into the LLM model to generate the optimized job description of the position to be recruited.

[0034] Furthermore, the enterprise information set includes one or more of the following tags and corresponding tag values: enterprise nature, enterprise scale, enterprise industry, enterprise establishment time or enterprise location city;

[0035] The functional tag set includes one or more of the following tags and corresponding tag values: work system or welfare benefits.

[0036] The embodiment of the present application provides a job description generation device based on LLM, including:

[0037] The first acquisition module is used to acquire the recruitment demand information input by the recruiter for the recruitment position;

[0038] The second acquisition module is used to acquire the enterprise portrait attributes of the recruiter to obtain an enterprise information set and a function label set; wherein the enterprise information set includes multiple enterprise information labels and label values ​​under each enterprise information label; the function label set includes multiple function labels and label values ​​under each function label;

[0039] A function estimation module, which is used to estimate the functions according to the recruitment demand information to obtain the first function corresponding to the position to be recruited;

[0040] A first construction module, which is used to construct a set of candidate job description samples according to the first function and enterprise information set;

[0041] A selection module, which is used to obtain a target sample of the job description based on a set of candidate sample descriptions and a selection strategy;

[0042] The second construction module is used to construct a job description prompt based on a preset prompt template, recruitment demand information, job description target sample, function label set and enterprise information set;

[0043] The generation module is used to input the constructed job description prompt into the LLM model to generate the job description of the position to be recruited.

[0044] An embodiment of the present application provides an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the method described above are implemented.

[0045] An embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the steps of the method described above are implemented.

[0046] An embodiment of the present application provides a computer program product, wherein the computer program product includes computer program instructions, and when the computer program instructions are executed by a processor, the steps of the method described above are implemented.

[0047] The above technical solution of the present application has at least the following beneficial technical effects:

[0048] 1. In the embodiment of the present application, even if the enterprise HR inputs relatively simple recruitment demand information for the recruitment position, by obtaining a function label set and a company information set, and selecting high-quality JD samples based on the first function and the industry in which the enterprise is located, a position description prompt is constructed, and the LLM model is input to generate a position description for the position to be recruited. In this way, the information such as the enterprise to which the position to be recruited belongs and the industry to which it is located is fully considered, as well as the core function labels that job seekers in the current market are concerned about for the position to be recruited. Therefore, the generated JD is more in line with the needs of HR, so that HR can use it without major changes, which can make up for the small amount of recruitment demand information input by the enterprise HR, and can improve the overall quality of the generated position description, so that the content of the generated position description is more detailed. At the same time, the diversity of high-quality JD samples and the differences in high-quality JD samples constructed under different functions are also considered, so that the generated JD can be differentiated and more personalized, which can improve recruitment efficiency.

[0049] 2. In the embodiment of the present application, the selection of high-quality JD samples takes into account the actual online effect. By setting the selection strategy, each sample can be selected, and samples with better display and recruitment effects are selected more times, so that the generated JD can be diversified and more personalized, and can improve recruitment efficiency; at the same time, appropriate consideration will be given to adding new JD samples so that the entire JD sample can be continuously iterated. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0051] Figure 1 It is a schematic diagram of the system architecture of an embodiment of the present application.

[0052] Figure 2 It is a flowchart of a method for generating job descriptions based on LLM in an embodiment of the present application.

[0053] Figure 3 It is a schematic diagram of the processing process of a method for generating a job description based on LLM in an embodiment of the present application.

[0054] Figure 4 It is a schematic diagram of the processing process of another LLM-based job description generation method in an embodiment of the present application.

[0055] Figure 5 It is a structural block diagram of a job description generating device based on LLM in an embodiment of the present application.

[0056] Figure 6 It is a schematic diagram of an electronic device used to implement a method for generating a job description based on LLM according to an embodiment of the present application. DETAILED DESCRIPTION

[0057] The principles and spirit of the present application will be described below with reference to several exemplary embodiments. It should be understood that the purpose of providing these embodiments is to make the principles and spirit of the present application clearer and more thorough, so that those skilled in the art can better understand and implement the principles and spirit of the present application. The exemplary embodiments provided herein are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments herein, all other embodiments obtained by ordinary technicians of the art without creative work are within the scope of protection of this application.

[0058] Embodiments of the present application relate to terminal devices and / or servers. Those skilled in the art will appreciate that the embodiments of the present application may be implemented as a system, apparatus, device, method, computer-readable storage medium, or computer program product. Therefore, the present disclosure may be specifically implemented in at least one of the following forms: complete hardware, complete software, or a combination of hardware and software. According to the embodiments of the present application, the present application claims protection for a method, apparatus, electronic device, computer-readable storage medium, and computer program product for generating job descriptions based on LLM. Figure 1 A schematic diagram of a system architecture of an embodiment of the present application is shown. Figure 1 As shown, the system includes a terminal device 102 and a server 104. Among them, the terminal device 102 may include at least one of the following: a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, various wearable devices, an augmented reality AR device, a virtual reality VR device, etc. A client can be installed on the terminal device 102. For example, the client can be a client that specifically performs a specific function (such as an application app), or a client with multiple application applets (different functions) embedded in it, or a client logged in through a browser. The user can operate on the terminal device 102. For example, the user can open the client installed on the terminal device 102 and input instructions through the client operation, or the user can open the browser installed on the terminal device 102 and input instructions through the browser operation. After the terminal device 102 receives the instruction input by the user, the request information containing the instruction is sent to the server 104. After receiving the request information, the server 104 performs corresponding processing and then returns the processing result information to the terminal device 102. The user instruction is completed through a series of data processing and information interaction.

[0059] In this document, terms such as first, second, third, etc. are only used to distinguish one entity (or operation) from another entity (or operation), but not to require or imply any order or relationship between these entities (or operations).

[0060] The following is a brief description of the concepts and technical terms that may be involved in the embodiments of the present application.

[0061] HR: The abbreviation of Human Resource, which means human resources, for example, it refers to the recruitment specialist in a company.

[0062] JD: A "Job Description" or "Job Description Document" details the specific requirements, scope of responsibilities, required skills, education, work experience, salary range, and other relevant information for a particular position.

[0063] LLM: Large Language Model (“Large Language Model”) is a natural language processing model based on deep learning technology, especially trained using neural network technology.

[0064] Function: In the recruitment field, function is the classification of positions, such as "sales representative", "sales director", etc.

[0065] ES: ES (Elasticsearch) is a distributed, scalable, real-time search and analysis engine.

[0066] Redis: Redis (Remote Dictionary Server) is an open source, ANSI C-language, network-supported, memory-based, and persistent log-based Key-Value database.

[0067] FastText: FastText is a word vector calculation and text classification tool open sourced by Facebook AI Research in 2016.

[0068] Prompt: refers to an input method or instruction used to guide large-scale language models to perform specific outputs or tasks.

[0069] Few-shot Prompting: Exploiting the generalization ability of the model, by adding a small number of high-quality examples (i.e., input-output pairs) in the prompt to help the model understand the task and generate outputs that meet expectations.

[0070] In corporate recruitment, for each position to be recruited, the company HR first needs to write a JD (job description), detailing key modules such as "job responsibilities", "qualifications", and "company benefits".

[0071] JD generation has undergone significant changes in recent years with the development of digitalization and intelligence. These changes are mainly reflected in the following aspects: 1. Intelligence: With the development of LLM, more and more companies provide JD intelligent generation based on LLM; 2. Standardization: In order to improve recruitment efficiency, many companies have begun to adopt standardized job description templates to ensure the professionalism of JD generation; 3. Diversification: In order to attract a wider range of candidates, companies are paying more and more attention to diversity and inclusiveness in JD, including using neutral language, avoiding gender discrimination, and emphasizing the importance of multicultural backgrounds. In summary, JD intelligent generation based on LLM is a trend in market development, which can help companies focus on solving problems such as efficiency and standardization.

[0072] However, many recruitment platforms currently still require HR to manually fill in job descriptions. However, HR is not the initiator of job recruitment needs and cannot deeply understand recruitment needs. Even if HR uses large models to generate job descriptions, due to the small amount of input information and the impact of response time on online calls, models with large parameters cannot be used, so the overall quality of the generated job descriptions is poor. In addition, because the information filled in by HR is limited, for example, most users' descriptions are "I have a sales position that requires a college degree or above and more than 5 years of work experience." This description is relatively simple and single, so the job descriptions generated by various companies using large models are not much different.

[0073] Based on this, the embodiment of the present application provides a method for generating a job description based on LLM, such as Figure 2 As shown, the method comprises the following specific steps:

[0074] S110: Obtaining recruitment demand information input by the recruiter regarding the recruitment position.

[0075] Specifically, in the embodiment of the present application, the recruiter is, for example, a recruitment specialist HR in a company. The HR can input recruitment demand information in the terminal device 102, which may include, for example, core information such as job requirements for the position and job responsibilities.

[0076] S120: Obtain the corporate profile attributes of the recruiter to obtain a corporate information set and a functional label set; wherein the corporate information set includes multiple corporate information labels and label values ​​under each corporate information label; and the functional label set includes multiple functional labels and label values ​​under each functional label.

[0077] Specifically, server 104 can query the redis KV database for the account login information received from the recruiter, and obtain the corporate portrait attributes of the recruiter's company; the corporate information set can come from the company registration information of the company on the recruitment platform, and the functional label set can be obtained by statistically analyzing the historical job descriptions distributed by the company. The functional label set may include the company's work system and welfare benefits and other corresponding requirements, such as the core functional labels of the positions to be recruited that job seekers in the market are concerned about. The corporate information set may include basic information such as the company name, company nature, company scale, company industry, company establishment time and company location city.

[0078] S130: Perform function estimation based on the recruitment demand information to obtain a first function corresponding to the position to be recruited.

[0079] Specifically, the server 104 may extract keywords from the received recruitment demand information to classify positions, thereby obtaining the functions corresponding to the positions to be recruited. In the embodiment of the present application, a function estimation method based on FastText may be adopted, or a function estimation method based on a BERT classification model may be adopted, and the function ID and function name corresponding to the position to be recruited may be output. It is understandable that function estimation based on recruitment demand information may also be implemented in other ways, which are not limited too much here.

[0080] S140: Construct a set of candidate job description samples based on the first function and the enterprise information set.

[0081] Specifically, the Redis KV database is queried by the function ID and the industry in which the enterprise is located to obtain high-quality JDs. The generation of high-quality JDs can be constructed manually, and more than 3 high-quality JD samples are constructed for each function in each industry as a set of candidate job description samples. In this way, more high-quality JD samples can be constructed to provide diverse choices, allowing each recruiter to differentiate the job descriptions ultimately generated using the large model.

[0082] S150: Obtaining a target sample of the job description based on the candidate sample set of the job description and the selection strategy.

[0083] Specifically, different high-quality JD samples are selected from the set of candidate job description samples as target job description samples. By setting the selection strategy, the high-quality JD samples selected each time may be different, and more high-quality JD samples can be selected online. The recruitment platform can also continuously add high-quality JD samples to form a virtuous circle, which can differentiate the job descriptions ultimately generated by each recruiter using the large model.

[0084] S160: Based on the preset prompt template, a job description prompt is constructed according to recruitment demand information, job description target sample, function label set and enterprise information set.

[0085] Specifically, prompt refers to an instruction project that interacts with the large-scale language model LLM. The prompt instruction can guide the large-scale language model to generate expected output results, that is, the large-scale language model can generate corresponding outputs according to the prompt instruction; the constructed job description prompt can contain recruitment demand information, job description target samples, function label sets and enterprise information sets, etc., which can make up for the lack of information input by the enterprise HR to improve the overall quality of the generated job description.

[0086] S170: Input the constructed job description prompt into the LLM model to generate a job description for the position to be recruited.

[0087] Among them, the large-scale language model LLM learns the grammar, semantics and contextual information of the language by training a large amount of text data, so that it can model and generate natural language text; based on the preset job description prompt template, the recruitment demand information and the target function attribute information set are input to generate the job description extraction instruction; the large-scale language model LLM can generate the job description of the position to be recruited based on the job description extraction instruction. In the embodiment of the present application, even if the enterprise HR inputs relatively simple recruitment demand information for the recruitment position, it obtains a function label set and a company information set, and selects high-quality JD samples based on the first function and the industry in which the enterprise is located, thereby constructing a position description prompt, inputting the LLM model and generating a job description for the position to be recruited; in this way, full consideration is given to information such as the company to which the position to be recruited belongs and the industry in which it is located, as well as the core function labels that job seekers in the current position to be recruited are concerned about in the market, so the generated JD is more in line with the needs of HR, so that HR can use it without major changes, which can make up for the small amount of recruitment demand information input by the enterprise HR, and can improve the overall quality of the generated position description, so that the generated position description is more detailed; at the same time, the diversity of high-quality JD samples and the differences in high-quality JD samples constructed under different functions are also considered, so that the generated JD can be differentiated and more personalized, which can improve recruitment efficiency.

[0088] In some embodiments, step S120: obtaining the corporate profile attributes of the recruiter to obtain a corporate information set and a functional label set may include the following specific steps:

[0089] S121: Get historical job descriptions posted by recruiters;

[0090] S122: Screening the acquired historical job descriptions to obtain core historical job descriptions with functional tags;

[0091] S123: Count the core historical job descriptions to obtain multiple candidate label values ​​under each functional label;

[0092] S124: Calculate the first probability of each candidate tag value under each functional tag; wherein the first probability is the number of historical job descriptions containing the current candidate tag value under the corresponding functional tag divided by the total number of historical job descriptions published by the recruiter;

[0093] S125: Sort the first probabilities corresponding to the multiple candidate label values ​​to obtain a target label value; wherein the target label value is the candidate label value corresponding to the maximum first probability.

[0094] Specifically, according to the core functional tags of the current positions to be recruited that job seekers are concerned about in the market, such as the company's work system and welfare benefits and other corresponding requirements, the historical job descriptions that have been published by the company where the recruiter is located are screened to obtain the core historical job descriptions with functional tags; the core historical job descriptions are counted to obtain multiple candidate tag values ​​under each functional tag, for example, the company's work system may have two days off a week, one day off a week, and a shift system for different jobs, and whether the welfare benefits include year-end bonuses, annual leave, and performance awards, etc.; by calculating the first probability of each candidate tag value under each functional tag, the candidate tag value corresponding to the maximum value in the first probability can be determined as the target tag value under each functional tag; in this way, the tag values ​​under the functional tags calculated in this way meet the requirements of different jobs, and the generated JD is more in line with the needs of HR.

[0095] In some embodiments, step S150: obtaining a target job description sample according to the job description candidate sample set and the selection strategy may include the following specific steps:

[0096] S151: Obtain the current number of resume submissions and current exposures of job seekers for the job positions published by the job description generated using the current job description candidate sample within a preset time period;

[0097] S152: Obtaining a reward value for a job description generated from a current job description candidate sample under the first function according to the current number of resume submissions and the current number of exposures;

[0098] S153: Obtaining a first total exposure count of the current job description candidate sample under the first function;

[0099] S154: Obtaining a second total exposure number of job descriptions generated by all job description candidate samples under the first function;

[0100] S155: Obtain a score for each candidate job description example according to the reward value, the first total exposure number, and the second total exposure number.

[0101] Specifically, the preset time period is set to 3 months, for example, and the current number of resumes submitted by job seekers for the job positions published by the job description generated by the current job description candidate sample within the latest 3 months from the current time and the current number of exposures of the job description are counted; the ratio of the current number of resumes submitted and the current number of exposures of the job description is calculated to obtain the submission rate, that is, the reward value obtained by the job description. The first total exposure times n of the candidate sample of the current job description under the first function on the terminal device 102 of the enterprise HR can be obtained. j ; Obtain the second total exposure times n of the job descriptions generated by all the job description candidate samples under the first function on the job seeker's terminal device 102; The score Score of each job description candidate sample can be calculated according to the following conditional formula j :

[0102]

[0103] It can be seen from the above formula that job description candidate samples with high submission rates have higher scores, and job description candidate samples that are selected fewer times can obtain higher scores, thereby ensuring that each job description candidate sample can be fully displayed. At the same time, job description candidate samples with high scores can obtain better display and recruitment effects.

[0104] In some embodiments, step S150: obtaining a target job description sample according to the job description candidate sample set and the selection strategy may further include the following specific steps:

[0105] S156: normalizing the score of each job description candidate sample to obtain a normalized score of each job description candidate sample; wherein the sum of the normalized scores of each job description candidate sample is set to 100;

[0106] S157: constructing each numerical range under each job description candidate sample according to the normalized score of each job description candidate sample; wherein each numerical range is connected to each other and is between 0 and 100;

[0107] S158: Obtain a random number between 0 and 100. If the random number is within a corresponding numerical range, select the corresponding job description candidate sample as a high-quality job description sample.

[0108] Specifically, for example, a job description candidate sample set may include 3 samples, and the sum of the normalized scores of the 3 samples is set to 100. A corresponding numerical interval is constructed for each sample, and a random number of 0-100 is set. If the obtained random number falls within the corresponding numerical interval, the corresponding job description candidate sample is selected as the high-quality job description sample. It can be concluded that the larger the score of the sample, the higher the probability of being selected, thereby ensuring that samples with good display and recruitment effects are selected more times; at the same time, it can also be achieved that each sample can be selected, so that the generated JD can be diversified and more personalized, which can improve recruitment efficiency.

[0109] In some embodiments, step S130: performing function estimation according to the recruitment demand information to obtain the first function corresponding to the position to be recruited may include the following specific steps:

[0110] S131: classify the recruitment demand information by function based on a pre-built multi-classification model to obtain estimated scores of multiple candidate functions;

[0111] S132: Sort the estimated scores of multiple candidate functions, and select the candidate function corresponding to the maximum estimated score as the first function corresponding to the position to be recruited.

[0112] Specifically, the pre-built multi-classification model can be built based on the BERT model. The BERT model (Bidirectional Encoder Representations from Transformers) is a pre-trained language representation model based on the Transformer structure. It learns rich language feature representations through unsupervised pre-training on large-scale text data. These representations can be used for various downstream natural language processing (NLP) tasks, such as text classification, named entity recognition, question-answering systems, etc.; the recruitment demand information is classified by function through the pre-built multi-classification model to obtain estimated scores of multiple candidate functions; the estimated scores of multiple candidate functions are sorted, and the candidate function corresponding to the maximum estimated score is selected as the first function corresponding to the position to be recruited.

[0113] In some embodiments, after step S170: inputting the constructed job description prompt into the LLM model to generate the job description of the position to be recruited, the job description generation method may further include the following specific steps:

[0114] S180: receiving the recruitment demand information after adjusting the recruitment position input by the recruiter;

[0115] S190: Based on the constructed job description prompt, the adjusted recruitment demand information is input into the LLM model to generate an optimized job description for the position to be recruited.

[0116] Specifically, there is no need for HR to restate their recruitment needs or manually optimize the returned results, which can improve the user experience of HR.

[0117] In an exemplary embodiment, the enterprise information set includes one or more of the following tags and corresponding tag values: enterprise name, enterprise nature, enterprise scale, enterprise industry, enterprise establishment time or enterprise city; the functional tag set includes one or more of the following tags and corresponding tag values: work system or welfare benefits.

[0118] The above describes the implementation methods and advantages of the embodiments of the present application through multiple embodiments. Figure 3-Figure 4 , describes in detail the specific processing process of the embodiment of the present application.

[0119] A method for generating a job description based on LLM in an embodiment of the present application may include the following specific steps:

[0120] S1. HR inputs recruitment demand information: HR uses a paragraph of text to express its recruitment demand, which mainly includes job requirements, job responsibilities and other core information. An example is as follows: "Recruiting sales representatives, responsible for product sales in East China, those with experience in medical devices are preferred, under 35 years old, need to travel frequently, and the company has good benefits."

[0121] S2. Obtain the attributes of the enterprise portrait where the HR is located: the enterprise is, for example, a corporate entity. The enterprise portrait label mainly comes from two parts. One is based on the information provided by the company when registering on the platform; the other is based on the statistics of the JD published by the company. The attributes of the enterprise portrait are shown in Table 1:

[0122] Table 1 Enterprise portrait attributes

[0123] Label value source Company Name XX Network Technology (Shanghai) Co., Ltd. Company registration information Nature of Company Foreign companies Company registration information Company size More than 2,000 people Company registration information industry human Resources Company registration information City Shanghai Company registration information Establishment 1998 Company registration information … … … Work system Two-day weekend JD Statistical Analysis Benefits Year-end bonus, annual leave, performance bonus JD Statistical Analysis … … …

[0124] JD statistical analysis: Based on the existing JD tag extraction system of the recruitment platform, relevant functional tags such as benefits and work system can be selected according to the historical job descriptions published by the company, and the number of occurrences of the tag values ​​can be counted. By comparing the probability of occurrence of the tag values, the tag value with the highest probability of occurrence can be screened out, thereby determining the company's core functional tags and corresponding tag values;

[0125] The probability of a tag value appearing = the number of JDs containing the tag value / the total number of JDs issued by the company.

[0126] S3. Function estimation: For example, based on the BERT multi-classification model, the recruitment demand information input by the recruiter is classified by function. The recruitment demand information in the above step S1 is classified by function, and the following functions can be obtained: sales representative, regional sales representative, sales manager, sales director and CEO, etc. The BERT multi-classification model can give a score to each function classification, and the function ID with the highest score can be selected as the final estimation result; for example, the regional sales representative has the highest estimated score, so the regional sales representative is selected as the function corresponding to the position to be recruited.

[0127] S4. Determine high-quality JD samples: Based on the job ID and company industry, the real-time search and analysis engine ES can be used to query Redis to obtain high-quality JDs stored in the database as candidate JD samples. The generation of high-quality JDs can be constructed manually, with more than 3 samples constructed for each job in each industry. More samples can also be selected online, which also supports the recruitment platform to continuously add new samples, forming a virtuous cycle.

[0128] According to the following conditional formula, the score of the j-th candidate JD sample under this function can be calculated: j :

[0129]

[0130] in is the reward value obtained by the jth JD candidate sample under this function at the current moment; the reward value can be set as the delivery rate of the JD generated by the JD candidate sample when the job seeker sees it:

[0131] Delivery rate = number of resumes submitted by job seekers / current exposure of the JD

[0132] The number of resume submissions and exposures of job seekers refer to the number of resume submissions of job seekers on the online client terminal 102 and the number of JD exposures generated by the JD candidate sample for job seekers after the current JD sample is released. The statistical period can be set to the latest 3 months;

[0133] n represents the total exposure times of JD generated by all JD candidate samples under this function after release; n j It represents the total exposure times of JD generated by the jth JD candidate sample after release under this function.

[0134] Based on Score j First, the scores of each candidate JD sample are normalized to ensure that the sum of the scores of all JD candidate samples under this function is equal to 100. The score S of each JD candidate sample j The specific calculation method is as follows:

[0135]

[0136] For S j Sort and for each S j Construct a right-closed and left-open numerical interval. The construction method is: traverse each S from small to large j , the left side is the sum of all the numbers before the value, and the right side is the sum of all the numbers before plus its own value. For example, the three JD candidate samples with scores of 20, 30, and 50, the constructed intervals are: [0,20); [20,50); [50,100)

[0137]

[0138] Get a random number between 0 and 100. If the random number falls within the corresponding interval, the corresponding JD candidate sample is selected. It can be seen that the larger the score of the JD candidate sample, the higher the probability of being selected, thereby ensuring that samples with good results can be selected more times.

[0139] The following is a specific example of a high-quality JD:

[0140] [input]

[0141] Company Information:

[0142] {

[0143] Company Name: Shanghai Meiyi Technology Co., Ltd.

[0144] Type: Private enterprise

[0145] Industry:Beauty

[0146] Company size: 50-100 people City: Shanghai

[0147] Benefits to be delivered: one day off per week, quarterly performance bonus, annual outing

[0148] Introduction: Focusing on medical beauty, currently mainly focusing on East China, it is a young startup company

[0149] }

[0150] Recruitment requirements:

[0151] "Recruiting a sales representative, mainly responsible for product sales in Shanghai, with more than three years of sales experience, and relevant work experience in the pharmaceutical industry"

[0152]

output

[0153] Job Responsibilities:

[0154] 1. Responsible for the promotion and sales of medical beauty products of Shanghai regional companies

[0155] 2. Accurately grasp customer needs, provide effective solutions and implement them;

[0156] 3. Develop customers, establish good relationships with customers, conduct customer relationship management, and strive to satisfy customers through interactive communication;

[0157] 4. Report market trends to the company's marketing department in a timely manner, and promptly feedback product quality, service and other issues to other relevant departments, and assist in resolving them.

[0158] Job Requirements:

[0159] 1. College degree or above;

[0160] 2. Strong communication and coordination skills and interpersonal understanding, optimistic and positive;

[0161] 3. Familiar with the medical and beauty industries, have a background in the medical and beauty industries, and have more than three years of sales experience;

[0162] 4. Have a positive working attitude, forge ahead despite difficulties, and carry out work with methods, skills and strategies;

[0163] Benefits: one day off a day, regular company training, quarterly performance bonuses, annual physical examinations, and a group outing every year.

[0164] S5. Build Prompt: Based on the input recruitment demand information, the obtained company information, function labels, and high-quality JD samples, fill them into the Prompt template to build a dynamic Prompt. For example, the constructed Prompt is as follows:

[0165] You are a recruitment expert in the field of human resources. Please complete the following task: Help HR generate JD. The requirements are as follows:

[0166] 1. Optimize the structure to make it clear and include the following parts: "Job responsibilities, job requirements, and benefits";

[0167] 2. Based on the provided company information, optimize the JD according to the company's nature, industry, etc., so that the JD is more suitable for the company's attributes;

[0168] 3. Based on the functional labels provided by the industry, please generate around these labels;

[0169] 4. Do not produce content that does not reflect positive values, and do not generate content that is not related to JD;

[0170] Here is an example:

[0171] ###

[0172] #High-quality JD sample#

[0173] ###

[0174] Please generate JD according to the following input:

[0175]

input

[0176] Company Information:#Company Information#

[0177] Recruitment requirements:#User input#

[0178]

output

[0179] S6. JD generation: Input the constructed Prompt into the large model LLM to generate JD.

[0180] S7, optimized generation: The large model can be called multiple times, and subsequently only the input recruitment demand information needs to be adjusted without repeating steps S2-S6, thereby generating an optimized JD.

[0181] Corresponding to the method embodiment of the present application, the present application also provides a job description generation device based on LLM, such as Figure 5 As shown, specifically, it may include:

[0182] The first acquisition module 510 is used to acquire the recruitment demand information input by the recruiter for the recruitment position;

[0183] The second acquisition module 520 is used to acquire the enterprise portrait attributes of the recruiter to obtain an enterprise information set and a function label set; wherein the enterprise information set includes multiple enterprise information labels and label values ​​under each enterprise information label; the function label set includes multiple function labels and label values ​​under each function label;

[0184] A function estimation module 530 is used to perform function estimation according to the recruitment demand information to obtain a first function corresponding to the position to be recruited;

[0185] A first construction module 540 is used to construct a set of candidate job description samples according to the first function and enterprise information set;

[0186] A selection module 550, which is used to obtain a target job description sample according to the job description candidate sample set and the selection strategy;

[0187] The second construction module 560 is used to construct a job description prompt based on a preset prompt template according to recruitment demand information, job description target sample, function label set and enterprise information set;

[0188] The generation module 570 is used to input the constructed job description prompt into the LLM model to generate the job description of the position to be recruited.

[0189] The electronic device in the embodiment of the present application may be a user terminal device, a server, other computing devices, or a cloud server. Figure 6 A schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application is shown. The electronic device may include a processor 601 and a memory 602 storing computer program instructions. When the processor 601 executes the computer program instructions, the process or function of any of the above-mentioned embodiments of the method is implemented.

[0190] 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 embodiments of the present application. The memory 602 may include a large-capacity memory for data or instructions. For example, the memory 602 may be at least one of the following: a hard disk drive (HDD), a read-only memory (ROM), a random access memory (RAM), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, a universal serial bus (USB) drive or other physical / tangible memory storage device. For another example, the memory 602 may include a removable or non-removable (or fixed) medium. For another example, the memory 602 may be inside or outside the integrated gateway disaster recovery device. The memory 602 may be a non-volatile solid-state memory. In other words, the memory 602 usually includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with computer-executable instructions, and when the software is executed (such as executed by one or more processors), the operation described in the method of the embodiment of the present application can be performed. The processor 601 implements the process or function of any method in the above embodiments by reading and executing the computer program instructions stored in the memory 602.

[0191] In one example, Figure 6The electronic device shown may also include a communication interface 603 and a bus 610. Among them, the processor 601, the memory 602, and the communication interface 603 are connected through the bus 610 and complete the communication between each other. The communication interface 603 is mainly used to realize the communication between each module, device, unit and / or device in the embodiment of the present application. The bus 610 includes hardware, software or both, and can couple the components of the online data traffic billing device to each other. For example, the bus may include at least one of the following: an accelerated graphics port (AGP) or other graphics bus, an enhanced industrial standard architecture (EISA) bus, a front-end bus (FSB), a hypertransport (HT) interconnect, an industrial standard architecture (ISA) bus, an infinite bandwidth interconnect, a low pin count (LPC) bus, a memory bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable buses. The bus 610 may include one or more buses. Although the embodiments of the present application describe or illustrate a specific bus, the embodiments of the present application may consider any suitable bus or interconnection method.

[0192] In combination with the method in the above embodiments, an embodiment of the present application also provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any method in the above embodiments is implemented.

[0193] In addition, an embodiment of the present application further provides a computer program product, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any one of the methods in the above embodiments is implemented.

[0194] The above exemplarily describes the flowcharts and / or block diagrams of the methods, devices, systems and computer program products of the embodiments of the present application, and describes the relevant various aspects. It should be understood that each box or combination thereof in the flowchart and / or block diagram can be implemented by computer program instructions, or by dedicated hardware that performs a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine that enables these instructions executed by such a processor to enable the implementation of the functions / actions specified in each box or combination thereof in the flowchart and / or block diagram. Such a processor can be a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.

[0195] The functional blocks shown in the structured block diagram of the embodiment of the present application can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc.; when implemented in software, it is a program or code segment used to perform the required task. The program or code segment can be stored in a memory, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0196] It should be noted that the present application is not limited to the specific configurations and processes described above or shown in the figures. The above is only a specific implementation mode of the present application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the described system, device, module or unit can refer to the corresponding process in the method embodiment without further description. It should be understood that the scope of protection of the present application is not limited to this. Any technician familiar with the technical field can think of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and these modifications or substitutions should be included in the scope of protection of the present application.

Claims

1. A method for generating job description based on LLM, characterized in that: include: Obtain the recruitment demand information input by the recruiter for the recruitment position; Obtain the enterprise portrait attributes of the recruiter to obtain an enterprise information set and a function label set; wherein the enterprise information set includes multiple enterprise information labels and label values ​​under each enterprise information label; the function label set includes multiple function labels and label values ​​under each function label; Estimating functions based on recruitment demand information to obtain the first functions corresponding to the positions to be recruited; According to the first function and the enterprise information set, a candidate sample set of job descriptions is constructed; According to the candidate sample set and selection strategy of job description, the target sample of job description is obtained; Based on the preset prompt template, a job description prompt is constructed according to the recruitment demand information, job description target sample, function label set and enterprise information set; Input the constructed job description prompt into the LLM model to generate the job description for the position to be recruited.

2. The method for generating a job description according to claim 1, characterized in that: The step of obtaining the corporate profile attributes of the recruiter to obtain a corporate information set and a functional label set includes: Get historical job descriptions posted by recruiters; The acquired historical job descriptions are screened to obtain core historical job descriptions with functional tags; Count the core historical job descriptions to obtain multiple candidate label values ​​under each functional label; Calculate the first probability of each candidate label value under each functional label; wherein the first probability is the number of historical job descriptions containing the current candidate label value under the corresponding functional label divided by the total number of historical job descriptions published by the recruiter; The first probabilities corresponding to the multiple candidate tag values ​​are sorted to obtain a target tag value; wherein the target tag value is the candidate tag value corresponding to the maximum first probability.

3. The method for generating a job description according to claim 1, characterized in that: The method of obtaining a target sample of job description based on a set of candidate sample of job description and a selection strategy includes: Obtain the current number of resume submissions and current exposures of job seekers for the job positions published by the job descriptions generated using the current job description candidate samples within a preset time period; According to the current number of resume submissions and the current number of exposures, the reward value obtained by the job description generated by the current job description candidate sample under the first function is obtained; Get the first total exposure times of the candidate samples of the current job description under the first function; Obtain the second total exposure times of the job descriptions generated by all job description candidate samples under the first function; A score for each job description candidate sample is obtained according to the reward value, the first total exposure number, and the second total exposure number.

4. The method for generating a job description according to claim 3, characterized in that: The step of obtaining a target sample of a job description based on a set of candidate sample descriptions and a selection strategy further includes: Normalizing the score of each job description candidate sample to obtain a normalized score of each job description candidate sample; wherein the sum of the normalized scores of each job description candidate sample is set to 100; According to the normalized score of each candidate job description sample, construct each numerical range under each candidate job description sample; wherein each numerical range is connected to each other and is between 0 and 100; Get a random number between 0 and 100. If the random number is within the corresponding numerical range, select the corresponding job description candidate sample as a high-quality job description sample.

5. The method for generating a job description according to claim 1, characterized in that: The function estimation is performed according to the recruitment demand information to obtain the first function corresponding to the position to be recruited, including: Classify the job demand information based on the pre-built multi-classification model to obtain the estimated scores of multiple candidate functions; The estimated scores of multiple candidate functions are sorted, and the candidate function corresponding to the maximum estimated score is selected as the first function corresponding to the position to be recruited.

6. The method for generating a job description according to claim 1, characterized in that: After inputting the constructed job description prompt into the LLM model to generate the job description of the position to be recruited, the following steps are also included: Receive the recruitment demand information after adjusting the recruitment positions input by the recruiter; Based on the constructed job description prompt, the adjusted recruitment demand information is input into the LLM model to generate the optimized job description of the position to be recruited.

7. The method for generating a job description according to claim 1, characterized in that: The enterprise information set includes one or more of the following tags and corresponding tag values: enterprise nature, enterprise scale, enterprise industry, enterprise establishment time or enterprise location city; The functional tag set includes one or more of the following tags and corresponding tag values: work system or welfare benefits.

8. A job description generation device based on LLM, characterized in that: include: The first acquisition module is used to acquire the recruitment demand information input by the recruiter for the recruitment position; The second acquisition module is used to acquire the enterprise portrait attributes of the recruiter to obtain an enterprise information set and a function label set; wherein the enterprise information set includes multiple enterprise information labels and label values ​​under each enterprise information label; the function label set includes multiple function labels and label values ​​under each function label; A function estimation module, which is used to estimate the functions according to the recruitment demand information to obtain the first function corresponding to the position to be recruited; A first construction module, which is used to construct a set of candidate job description samples according to the first function and enterprise information set; A selection module, which is used to obtain a target sample of the job description based on a set of candidate sample descriptions and a selection strategy; The second construction module is used to construct a job description prompt based on a preset prompt template, recruitment demand information, job description target sample, function label set and enterprise information set; The generation module is used to input the constructed job description prompt into the LLM model to generate the job description of the position to be recruited.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the electronic device executes the computer program instructions, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images