Recommendation method and device based on two-way implicit demand in recruitment and electronic equipment

Abstract

This application discloses a recommendation method, apparatus, and electronic device based on two-way implicit needs in recruitment. The method includes: acquiring a pre-trained first language model; the first language model is capable of converting the company's competency requirements information in the job seeker's resume information into implicit job seeker requirement features describing the job seeker's deep preferences based on the input job seeker resume information and corresponding first prompt words, and generating a job seeker user profile including the implicit job seeker requirement features; and the first language model is capable of converting the talent competency requirements information in the company's recruitment information into implicit company requirement features describing the company's deep preferences based on the input company recruitment information and corresponding second prompt words, and generating a company user profile including the company's implicit requirement features. Utilizing embodiments of this application can improve the accuracy of job or talent recommendations.

Description

Technical Field

[0001] This application relates to the field of Internet technology, as well as artificial intelligence and data processing technology, and in particular to a recommendation method, apparatus, device, program product and storage medium based on two-way implicit needs in recruitment. Background Technology

[0002] In today's digital recruitment field, talent recommendation systems have become an important bridge connecting job seekers and companies. Mainstream recommendation methods are typically based on structured data matching, where the system calculates the degree of fit by analyzing explicit skill keywords, educational background, years of work experience, and job responsibilities and hard requirements listed in the job description.

[0003] However, this matching model, which relies on structured fields, often only reflects the degree to which job seekers and positions meet the hard requirements, failing to address the deeper factors that influence the long-term sustainability of an employment relationship. For example, key information such as company culture, team work style, and the job seeker's career values ​​are often scattered throughout unstructured text descriptions and cannot be effectively captured by traditional keyword matching. This leads to many candidates who appear highly compatible on paper leaving quickly after joining the company because the actual work environment does not meet their expectations, ultimately resulting in wasted recruitment costs for companies and hindered career development for talented individuals. Summary of the Invention

[0004] In view of this, embodiments of this application provide a recommendation method, apparatus, electronic device, computer-readable storage medium, and computer program product based on two-way implicit needs in recruitment, solving at least one technical problem.

[0005] This application provides a recommendation method based on two-way implicit needs in recruitment, including: obtaining a pre-trained first language model, wherein the first language model can, based on input job seeker resume information and corresponding first prompt words, transform the company's quality requirements information in the job seeker resume information into implicit job seeker needs features describing the job seeker's deep preferences, and generate a job seeker user profile including the job seeker's implicit needs features; and wherein the first language model can, based on input company recruitment information and corresponding second prompt words, transform the talent quality requirements information in the company recruitment information into implicit company needs features describing the company's deep preferences, and generate a company user profile including the company's implicit needs features; inputting the job seeker resume information and corresponding first prompt words into the first language model, so that the first language model generates a user profile including the job seeker's implicit needs features. An initial job seeker profile is generated based on demand characteristics. The first prompt word includes differentiated semantic understanding rules for the same job seeker's employer competency requirements in different job-seeking scenarios or industry contexts. Company recruitment information and corresponding second prompt words are input into a first large language model to generate an initial company profile including implicit company requirements. The second prompt word includes differentiated semantic understanding rules for talent competency requirements associated with the company's attributes. A two-way matching degree between the job seeker and the company is calculated based on the initial job seeker profile, company recruitment information, job seeker resume information, and the initial company profile. A recommendation list is generated and displayed to the user based on the two-way matching degree, the job seeker resume information, and the explicit structured features in the company recruitment information. The explicit structured features are structured fields recording the user's hard requirements.

[0006] Optionally, the method according to the embodiments of this application further includes: analyzing the deviation between the user's actual implicit needs and their initial profile based on the interaction behavior data of job seeker users and / or enterprise users on the recommendation results, wherein the interaction behavior data includes at least one of browsing history, click behavior, job application, interview invitation, hiring results, onboarding feedback and performance; and updating the semantic understanding rules in the first prompt word and / or the second prompt word according to the deviation analysis results, so that the subsequently obtained implicit need features are more consistent with the user's true preferences.

[0007] Optionally, according to the method of this application embodiment, calculating the bidirectional matching degree between job seekers and companies based on the initial job seeker profile, company recruitment information, job seeker resume information, and initial company profile includes: calculating the matching degree between the initial job seeker profile and company recruitment information to obtain a first matching degree; calculating the matching degree between the initial company profile and job seeker resume information to obtain a second matching degree; and generating a bidirectional matching degree based on the first matching degree and the second matching degree.

[0008] Optionally, according to the method of this application embodiment, the enterprise attributes include at least one of industry type, enterprise size, development stage, job category, and enterprise culture label.

[0009] Optionally, the method according to the embodiments of this application further includes: obtaining a second language model, the second language model being able to output successfully matched competency requirement information combinations and / or unmatched competency requirement information combinations based on input feedback data and the recruitment information of enterprises and the resume information of job seekers in the feedback data, wherein the feedback data includes positive feedback data and negative feedback data, the positive feedback data including at least one of talent onboarding records, long-term stable employment records, records of talent receiving good performance evaluations, records of enterprises issuing offer letters, and records of enterprises repeatedly hiring talents with similar implicit characteristics, the negative feedback data including at least one of talent refusing interviews or offer letters, records of enterprises refusing talents, and records of talent short-term departures; inputting the feedback data and the recruitment information of enterprises and the resume information of job seekers in the feedback data into the second language model, so as to The second language model is configured to output successfully matched competency requirement information combinations and / or unmatched competency requirement information combinations. Based on the successfully matched and / or unmatched competency requirement information combinations, the semantic understanding rules in the first and / or second prompt words are adjusted. The adjusted first prompt word and job seeker resume information are input into the first language model, causing the first language model to output an adjusted job seeker profile. Similarly, the adjusted second prompt word and company recruitment information are input into the first language model, causing the first language model to output an adjusted company profile. The adjusted company profile, company recruitment information, job seeker profile, and job seeker resume information are used to calculate the updated bidirectional matching degree. Based on the updated bidirectional matching degree, the explicit structured features in the job seeker resume information, and the company recruitment information, an updated recommendation list is generated and displayed to the user.

[0010] Optionally, the method according to the embodiments of this application further includes: obtaining historical interaction behavior data of the enterprise on the recruitment platform, wherein the historical interaction behavior data includes at least one of viewing records, collection records, communication records, and hiring records of job seekers' resumes; inputting the historical interaction behavior data and the recruitment information into the first large language model to extract supplementary features representing the enterprise's implicit preferences, and using them to update the initial enterprise profile.

[0011] This application provides a recommendation device based on two-way implicit needs in recruitment, comprising: an acquisition module, configured to acquire a pre-trained first large language model, wherein the first large language model can, based on input job seeker resume information and corresponding first prompt words, transform the company quality requirement information in the job seeker resume information into job seeker implicit requirement features describing the job seeker's deep preferences, and generate a job seeker user profile including the job seeker implicit requirement features; and the first large language model can, based on input company recruitment information and corresponding second prompt words, transform the talent quality requirement information in the company recruitment information into company implicit requirement features describing the company's deep preferences, and generate a company user profile including the company implicit requirement features; and a job seeker profile generation module, configured to input job seeker resume information and corresponding first prompt words into the first large language model, so that the first large language model generates an initial profile including the job seeker implicit requirement features. The job seeker profile includes a first prompt word comprising differentiated semantic understanding rules for the same job seeker's corporate competency requirements in different job-seeking scenarios or industry contexts; a corporate profile generation module, used to input corporate recruitment information and corresponding second prompt words into the first large language model, so that the first large language model generates an initial corporate profile including implicit corporate requirements features, wherein the second prompt word includes differentiated semantic understanding rules for talent competency requirements information associated with the attributes of the company; a two-way matching degree calculation module, used to calculate the two-way matching degree between the job seeker and the company based on the initial job seeker profile, corporate recruitment information, job seeker resume information, and the initial corporate profile; and a recommendation list generation module, used to generate a recommendation list and display it to the user based on the two-way matching degree, the job seeker resume information, and explicit structured features in the corporate recruitment information, wherein the explicit structured features are structured fields recording the user's hard requirements.

[0012] This application provides an electronic device, which includes a processor and a memory storing computer program instructions; the electronic device executes the computer program instructions to implement the method described above.

[0013] This application provides a computer program product, which includes computer program instructions that, when executed, implement the method described above.

[0014] This application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the method described above.

[0015] This application, by introducing a large language model and combining it with a differentiated prompting word strategy, can deeply analyze the preference information hidden in resumes and job postings, effectively capturing deep-seated factors that traditional structured matching struggles to reach, such as corporate culture, team style, and career values. This allows for the construction of a more comprehensive user profile, helping to reduce the risk of rapid turnover due to discrepancies between the actual work environment and expectations after onboarding, reducing wasteful recruitment costs for companies, and improving job satisfaction for talent. This achieves a dual optimization of recommendation accuracy and user experience. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly described below.

[0017] Figure 1 This is a schematic diagram of the system architecture of an embodiment of this application.

[0018] Figure 2 This is a flowchart of a recommendation method based on bidirectional implicit requirements according to an embodiment of this application.

[0019] Figure 3 This is a schematic diagram illustrating the processing of the recommendation method based on bidirectional implicit requirements in an embodiment of this application.

[0020] Figure 4 This is a schematic structural block diagram of a recommendation device based on bidirectional implicit requirements according to an embodiment of this application.

[0021] Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0022] The principles and spirit of this application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided to make the principles and spirit of this application clearer and more thorough. The exemplary embodiments provided herein are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0023] Embodiments of this application relate to terminal devices and / or servers. Implementations of this application can be a system, terminal, device, method, computer-readable storage medium, or computer program product, and can be specifically implemented as entirely hardware, entirely software, or a combination of hardware and software. Figure 1This diagram illustrates a system architecture according to an embodiment of the present application, including a terminal device 102 and a server 104. The terminal device 102 may include at least one of the following: a smartphone, tablet computer, laptop computer, desktop computer, smart TV, various wearable devices, augmented reality (AR) devices, virtual reality (VR) devices, etc. A client application, such as an app, mini-program, or browser-based client, can be installed on the terminal device 102. Users can input commands through the client, and the terminal device 102 can send request information containing the commands to the server 104. Upon receiving the request information, the server 104 performs corresponding processing and returns the processing result information to the terminal device 102. The server 104 may be a local server or a cloud server, and may be a single server or a server cluster, etc.

[0024] In this document, the terms "first," "second," "third," etc., are used only to distinguish one entity (or operation) from another in textual description, and do not require or imply any sequential order between these entities (or operations).

[0025] Figure 2 The diagram illustrates a flowchart of a recommendation method based on two-way implicit needs in recruitment, according to an embodiment of this application. The method includes the following steps: S101: Obtain the pre-trained first language model. The first language model can transform the enterprise quality requirements information in the job seeker's resume information into implicit needs features describing the job seeker's deep preferences based on the input job seeker resume information and the corresponding first prompt words, and generate a job seeker user profile including the implicit needs features. Also, the first language model can transform the talent quality requirements information in the enterprise recruitment information into implicit needs features describing the enterprise's deep preferences based on the input enterprise recruitment information and the corresponding second prompt words, and generate an enterprise user profile including the implicit needs features. S102: Input the job seeker's resume information and the corresponding first prompt words into the first large language model so that the first large language model generates an initial job seeker profile including the job seeker's implicit needs features, wherein the first prompt words include differentiated semantic understanding rules for the same job seeker's corporate quality needs information in different job search scenarios or industry backgrounds. S103: Input the company's recruitment information and the corresponding second prompt words into the first language model so that the first language model can generate an initial company profile including the company's implicit needs features. The second prompt words include differentiated semantic understanding rules for talent quality information related to the company's attributes. S104: Calculate the two-way matching degree between job seekers and companies based on the initial job seeker profile, company recruitment information, job seeker resume information and initial company profile; S105: Based on the bidirectional matching degree, the explicit structured features in the job seeker's resume information and the company's recruitment information, a recommendation list is generated and displayed to the user. The explicit structured features are structured fields that record the user's hard requirements.

[0026] This application provides a recommendation method based on the implicit needs of both parties in recruitment. The purpose is to deeply understand the implicit information in the job seeker's and company's needs through a first language model, thereby constructing a profile that reflects the user's deep preferences and achieving more accurate two-way matching based on it.

[0027] In this embodiment, for job seekers, the corporate competency requirements information refers to the information recorded in the job seeker's resume describing the characteristics of the company the job seeker expects, including descriptions of at least one of corporate culture, team atmosphere, work intensity, career development opportunities, and value alignment. The job seeker's implicit needs are the deep preferences analyzed by the first language model based on this information, including potential inclinations towards corporate culture (e.g., flat management, bureaucracy), work pace (e.g., pursuit of stability), and team atmosphere (e.g., results-oriented, humanistic care). For example, the implicit need characteristic analyzed based on "hoping the company values ​​humanistic care" is: the company respects its employees. For companies, the talent competency requirements information refers to the information in the company's recruitment information describing the professional and psychological qualities the company expects in its talent, such as hoping employees have high stress resistance. The implicit needs of enterprises are the deep preferences of enterprises analyzed by the first language model based on the information on talent quality needs. They refer to the soft requirements for talents in addition to hard skills, such as the degree of alignment of values, personality traits, and reaction patterns to pressure. For example, based on "hoping that employees have high stress resistance", the implicit needs of enterprises can be derived as: being able to overcome technical difficulties.

[0028] Regarding the first and second prompt words, in this embodiment, the first prompt word is a pre-designed piece of text used to guide the model to focus on specific aspects of the resume and interpret the text according to preset rules. A key feature of the first prompt word is that it includes differentiated semantic understanding rules for the same job seeker's employer competency requirements in different job-seeking scenarios or industry contexts. This means that when parsing the same resume description, the model will assign different semantic weights to the same words or expressions based on the job-seeking scenario (e.g., recent graduates seeking employment, experienced professionals changing jobs) or the target industry (e.g., internet, manufacturing). For example, for "hoping for flexible working hours," in the internet industry, it has a high weighted association with the implicit requirement of "results-oriented companies"; in the manufacturing industry, it has a high weighted association with the implicit requirement of "simple job content." The second prompt word also contains differentiated semantic understanding rules, which are usually associated with the company's attributes, such as the industry, size, and development stage of the company. For example, for "stress resistance", smaller companies should associate it with "employees' ability to adapt to high-intensity work" with a higher weight, while larger companies should associate it with "employees' ability to adapt to frequent assessments" with a higher weight.

[0029] In this application, the first language model, after pre-training, can extract the implicit content about the user's deep preferences from the input text information and corresponding prompt words, transform it into implicit demand features, and generate a user profile containing these features.

[0030] To enable the first major language model to extract deep content from unstructured text and transform it into structured features, a series of data preparation, model training, and optimization processes are required. Ultimately, the model learns to understand semantic relationships within a specific domain and can map natural language descriptions into quantifiable feature representations according to preset dimensions.

[0031] For example, in the data preparation phase, a large-scale, high-quality domain corpus needs to be built. Taking the talent recommendation scenario as an example, training data can come from multiple channels: first, publicly available recruitment information, including job descriptions, requirements, and company introductions; second, job seekers' resume data, covering unstructured text such as educational background, work experience, project experience, and self-evaluation; and third, historical recruitment behavior data, such as interview feedback, hiring results, and records of reasons for leaving previous jobs. These data collectively constitute the raw materials for model learning. Based on this, the data needs to be cleaned and preprocessed to remove irrelevant information and perform text normalization. More importantly, the data needs to be labeled, that is, each piece of text needs to be labeled with its implicit features. For example, for a job description: "We need candidates who can independently drive projects in a fast-paced environment," it can be labeled as: Autonomy: High, Stress Resistance: Medium, Collaboration Needs: Low. This labeling work requires establishing a unified feature dimension system, such as dividing implicit features into major categories like work style, values, and motivational tendencies, and further subdividing each major category into several quantifiable sub-dimensions. The quantity and quality of labeled samples directly determine the accuracy of the model's subsequent extraction capabilities.

[0032] After the data is ready, the model training phase begins. A common approach is to first pre-train the base model using a large-scale corpus of general-domain data, giving it general language understanding capabilities. The pre-training phase typically employs self-supervised learning, such as using a masked language model task to teach the model the contextual relationships between words. Building on this, domain adaptation and task fine-tuning are performed. Domain adaptation involves further training the pre-trained model using the constructed recruitment domain corpus, familiarizing the model with the domain's terminology and expressions. Task fine-tuning uses labeled latent feature data to design specific learning tasks. A common practice is to add a classification or regression layer to the primary language model, mapping the model's output text representation to preset feature dimensions. During training, the input text (such as a resume excerpt) and corresponding prompts are concatenated and fed into the model. The model's output is compared with manually labeled feature values ​​to calculate the loss, and the model parameters are updated through backpropagation. Through iterative training with numerous such samples, the model gradually learns to capture semantic cues related to latent features from the text and transform them into structured feature outputs.

[0033] After the above training process, the first language model ultimately possesses the ability to extract semantic information from the text layer by layer through its internal attention mechanism and multi-layer neural network when given text and corresponding prompts. This information is then matched with the feature patterns learned during training, ultimately outputting a set of structured implicit requirement features. For example, for a company recruitment ad stating "must lead a team to launch a new product within three months," the model, based on the prompt "associating the completion of a new product in a short time with high stress tolerance," and after internal processing and calculation, might output structured features such as: "High stress tolerance: 0.85," "High leadership: 0.75," and "Strong innovation ability: 0.60," where the numerical values ​​represent the association weights. In this way, the original talent qualification information is transformed into implicit requirement features for subsequent matching calculations.

[0034] During the initial phase of using the first language model, job seekers' resume information and corresponding first cue words need to be input into the first language model to generate an initial job seeker profile. Similarly, for enterprise users, company recruitment information and corresponding second cue words need to be input into the first language model to generate an initial company profile.

[0035] Furthermore, this application calculates the two-way matching degree between job seekers and companies based on the generated initial job seeker profile, initial company profile, job seeker resume information, and company recruitment information. The calculation of the two-way matching degree in this application needs to consider two aspects: first, whether the job seeker's characteristics meet the company's talent requirements; and second, whether the company's characteristics meet the job seeker's expectations of the company. In some embodiments of this application, optionally, the two-way matching degree is a comprehensive matching degree of the matching degree between the job seeker profile and the company recruitment information and the matching degree between the company profile and the job seeker resume information. The job seeker profile includes the job seeker's implicit requirement characteristics, which are matched with information in the company recruitment information that can meet these requirements to obtain the job seeker's matching degree with the company. Similarly, the company profile includes the company's implicit requirement characteristics, which are matched with information in the job seeker's resume information that can meet these requirements to obtain the company's matching degree with the job seeker. Finally, by fusing these two matching degrees, the two-way matching degree can be obtained.

[0036] Finally, based on the calculated two-way matching degree and combined with the explicit structured features clearly stated in the resume information or company recruitment information, a recommendation list is generated and displayed to the user (job seeker or company). In this embodiment, explicit structured features refer to structured field information directly filled in by the user, such as educational requirements, years of work experience, skill certificates, salary range, etc.—content with clear meaning. This information usually exists in the form of standardized forms and can be directly used for filtering. When generating the recommendation list, preliminary filtering can be performed based on explicit structured features to eliminate obviously mismatched candidates. Then, the remaining candidates are sorted according to the two-way matching degree, and results with high matching degrees are displayed to the user first. Alternatively, the degree of conformity between the two-way matching degree and explicit structured features can be comprehensively scored, and the list can be sorted according to the comprehensive score. The final recommendation list can simultaneously consider both hard criteria and soft fit, thereby improving the accuracy of recommendations and user satisfaction.

[0037] This application, by introducing a large language model and combining it with a differentiated prompting word strategy, can deeply analyze the preference information hidden in resumes and job postings, effectively capturing deep-seated factors that traditional structured matching struggles to reach, such as corporate culture, team style, and career values. This allows for the construction of a more comprehensive user profile, helping to reduce the risk of rapid turnover due to discrepancies between the actual work environment and expectations after onboarding, reducing wasteful recruitment costs for companies, and improving job satisfaction for talent. This achieves a dual optimization of recommendation accuracy and user experience.

[0038] Optionally, the method according to the embodiments of this application further includes: Based on the interaction behavior data of job seeker users and / or enterprise users on the recommendation results, the deviation between the user's actual implicit needs and their initial profile is analyzed. The interaction behavior data includes at least one of browsing history, click behavior, job application, interview invitation, recruitment results, onboarding feedback and performance. Based on the deviation analysis results, update the semantic understanding rules in the first prompt word and / or the second prompt word so that the subsequently obtained implicit demand features are more in line with the user's true preferences.

[0039] In this application embodiment, historical interaction data covers the entire process from a user's initial contact with recommendation results to their final decision, encompassing a very broad range. For example, for job seekers, their interaction behavior might include which types of positions they browsed, which positions they submitted their resumes for, whether they accepted interview invitations, whether they ultimately accepted job offers, and even their work performance and tenure after joining the company. For enterprise users, interaction behavior might be reflected in which candidates' resumes HR clicked and viewed, which candidates they sent interview invitations to, who ultimately received job offers, and the performance evaluation results and departure records of new employees after joining the company. This behavioral data represents the result of users voting with their actual choices, and compared to the words they write in their resumes or job postings, it more accurately reflects their deep-seated preferences, sometimes even those they themselves are not consciously aware of.

[0040] Next, the system compares this real-world interaction data with the initial profiles of either the company or the job seeker to determine whether the implicit needs generated by the primary language model match the user's actual needs. For example, the system might extract implicit needs from a job seeker's resume, including a preference for flat management structures and a desire for work-life balance, and recommend positions in relatively relaxed small and medium-sized enterprises. However, by observing the user's interaction data, the system discovers that the user frequently clicks on and applies for positions at leading internet companies known for their high salaries and demanding work schedules. This discrepancy between actual choices and the initial profile constitutes a bias. The system needs to identify statistically significant and inductive patterns from a large number of individual biases. For instance, the system might find that all job seekers graduating from certain universities with specific skill sets, despite generally expressing a desire for stability in their resumes, tend to apply to rapidly growing startups. This forms a bias pattern that can be calibrated later.

[0041] After analyzing the deviation patterns, the system does not directly modify existing user profiles. Instead, it adopts a more fundamental and long-term optimization approach: updating the semantic understanding rules in the first and / or second prompt words. In this application, prompt words are the analysis framework provided for the first major language model, which includes rules on how to map textual descriptions to implicit demand features. The update here involves adjusting these rules. For example, in the initial version of the prompt words, there might be a rule that when a resume mentions seeking stable career development, it should be highly weighted towards implicit features such as a preference for mature companies and risk aversion. However, through the aforementioned deviation analysis, the system finds that this rule does not hold true for a specific group (such as top technical talents), whose stable career development might be interpreted as long-term dedication to platforms with high growth potential. Therefore, the system generates a calibration rule to supplement or correct the original prompt words. For example, for technical talents who graduated from top universities and have contributed to open-source projects, seeking stable career development should be more highly weighted towards implicit demand features such as pursuing long-term technical accumulation and valuing the company's technical atmosphere. This updated rule will be applied to all new users with similar backgrounds who join the system in the future.

[0042] In some embodiments of this application, the enterprise attributes may optionally include at least one of industry type, enterprise size, development stage, job category, and corporate culture label.

[0043] When the primary language model processes corporate recruitment information and generates initial corporate profiles, it needs to incorporate corporate attributes to activate specific semantic understanding rules. For example, regarding the description of strong resilience, if the corporate attributes indicate that its industry is internet and its development stage is startup, the model might associate resilience with implicit requirements such as adapting to rapid iteration and accepting long working hours with higher weight. However, if the corporate attributes indicate that its industry is public utilities and its size is large, the model might associate it with characteristics such as handling complex interpersonal relationships and coping with procedural assessments. The job category attribute is equally important. For instance, the requirement for resilience in corporate recruitment information should be interpreted differently for sales positions and R&D positions. The former might be associated with performance pressure, while the latter focuses more on technical challenge pressure. Corporate culture tags may directly provide a set of predefined value keywords, such as results-oriented and humanistic care, guiding the model to find matching semantic cues in the text.

[0044] By associating semantic understanding rules with enterprise attributes, large models can generate user profiles that better reflect the real needs of enterprises, thereby improving the accuracy of matching job seekers with enterprises.

[0045] In some embodiments of this application, optionally, the bidirectional matching degree between job seekers and companies is calculated based on the initial job seeker profile, company recruitment information, job seeker resume information, and initial company profile, including: Calculate the matching degree between the initial job seeker profile and the company's recruitment information to obtain the first matching degree; Calculate the matching degree between the initial company profile and the job seeker's resume information to obtain the second matching degree; Based on the first and second matching degrees, a bidirectional matching degree is generated.

[0046] In this embodiment, the company recruitment information includes not only explicit structured features (salary, age) and talent skill requirements, but also some of the company's own characteristics, such as meal allowances and holiday benefits. This information can be matched with the implicit job seeker needs represented in the initial job seeker profile to determine the degree to which the company's conditions meet the job seeker's needs—that is, the matching degree between the initial job seeker profile and the company recruitment information. Similarly, the job seeker's resume information also includes not only explicit structured features (age, graduating school) and company skill requirements, but also some of the job seeker's own characteristics, such as the ability to complete complex tasks and strong stress resistance. This information can be matched with the implicit company needs represented in the initial company profile to determine the degree to which the job seeker's conditions meet the company's needs—that is, the matching degree between the initial company profile and the job seeker's resume information.

[0047] Furthermore, if we are calculating the match between job seeker profiles and job postings to obtain the first match between the job seeker and the company, we can match the implicit requirements in the job seeker profile with similar content in the company's job postings. For example, we can match "good company atmosphere" in the job seeker profile with "the company has a good atmosphere" in the job postings, because both contain "atmosphere" and "good," thus obtaining a high match. If we are calculating the match between company profiles and job seeker resume information, we can match the implicit requirements in the company profile with similar content in the job seeker resume information. For example, we can match "able to complete complex tasks" in the company profile with "possesses the ability to complete complex tasks" in the job postings, because both contain "complete" and "complex tasks," thus obtaining a high match.

[0048] Afterwards, a comprehensive calculation of these two matching scores yields a bidirectional matching score. This application's bidirectional matching score considers needs from different directions, improving the matching accuracy between the two needs. For the same company A and job seeker B, from company A's perspective, job seeker B's score is 60 points, while from job seeker B's perspective, company A's score is 90 points. The method proposed in this application merges these two scores to generate a comprehensive bidirectional matching score, providing a clear indication of the candidate's ranking and priority. For example, for company A, the scores of 60 and 90 can be multiplied by their respective weights and summed to obtain the final bidirectional matching score between job seeker B and company A. In some embodiments of this application, optionally, the bidirectional matching score can be a weighted sum of two matching scores.

[0049] Optionally, the method according to the embodiments of this application further includes: The second language model is obtained, which can output successfully matched and / or unmatched combinations of competency requirements information based on the input feedback data, company recruitment information, and job seeker resume information. The feedback data includes positive and negative feedback data. Positive feedback data includes at least one of the following: talent onboarding records, long-term stable employment records, records of good performance evaluations, records of company offer letters, and records of companies repeatedly hiring talent with similar implicit characteristics. Negative feedback data includes at least one of the following: records of talent refusing interviews or offer letters, records of companies rejecting talent, and records of short-term employee departures. The feedback data, along with the recruitment information of companies and the resumes of job seekers contained in the feedback data, are input into the second language model so that the second language model can output successfully matched competency requirement information combinations and / or unmatched competency requirement information combinations. Adjust the semantic understanding rules in the first and / or second prompt words based on the successfully matched literacy requirement information and the unmatched literacy requirement information; The adjusted first prompt word and job seeker resume information are input into the first language model, so that the first language model outputs the adjusted job seeker profile; and the adjusted second prompt word and company recruitment information are input into the first language model, so that the first language model outputs the adjusted company profile. The updated two-way matching degree is calculated using the adjusted enterprise profile, enterprise recruitment information, job seeker profile, and job seeker resume information. Based on the updated two-way matching degree, job seeker resume information, and explicit structured features in company recruitment information, an updated recommendation list is generated and displayed to the user.

[0050] This application proposes a second major language model specifically responsible for analysis and learning, thereby elevating the system's self-optimization capabilities to a new level of specialization. The entire solution can be viewed as a dual-engine system with an "execution unit" and a "learning unit" working collaboratively. The execution unit (first major language model) is responsible for understanding text and generating profiles, while the learning unit (second major language model) is responsible for extracting patterns from historical experience and, in turn, optimizing the working method of the execution unit.

[0051] In this embodiment, the second language model and the first language model used to generate the initial user profile have clearly defined roles. The initial user profile includes an initial job seeker profile and an initial company profile. The input information for the second language model consists of feedback data, job seeker resume information, and company recruitment information. The feedback information includes decision results indicating successful or unsuccessful matching, such as records of successful onboarding and stable employment, records of job offers issued by the company, records of job seekers declining offers, or records of job resignations within a short period. Based on this feedback data, cases of successful matching between companies and job seekers (job seekers) and cases of unsuccessful matching can be determined. The job seeker resume information and company recruitment information refer to the resume information of the job seeker and the recruitment information of the company involved in the feedback data. For example, for the feedback data of successful onboarding and stable employment, the input information for the second language model also includes the resume information of the successfully onboarded and stable employee and the recruitment information of the company for the position applied for by that employee. Based on this information, the second language model can perform in-depth analysis and output two types of results: one type is the successfully matched combination of competency requirements, which includes the combinations of job seekers' implicit competency characteristics and the corresponding company competency conditions in the company's recruitment information that satisfy the job seekers' implicit competency characteristics in cases where a stable employment relationship is ultimately reached, as well as the combinations of company competency characteristics and the corresponding job seekers' competency conditions in the job seekers' resumes that satisfy the company's implicit competency characteristics in cases where an employment relationship is terminated early or no employment relationship is established, which includes the combinations of job seekers' implicit competency characteristics and the corresponding company competency conditions in the company's recruitment information that satisfy the job seekers' implicit competency characteristics in cases where an employment relationship is terminated early or no employment relationship is established in cases, as well as the combinations of job seekers' implicit competency characteristics and the corresponding job seekers' sampling conditions in the job seekers' resumes that satisfy the company's implicit competency characteristics.

[0052] Next, the second language model is used to output successful and unsuccessful skill requirement combinations to adjust the semantic understanding rules in the first and / or second prompt words. For example, in the initial version of the first prompt word, one rule is: "When a job seeker's resume contains the description 'hopes for challenging tasks,' it should be highly weighted in association with the implicit requirement feature 'hopes the company has a wide range of business operations.'" Then, analysis of multiple successfully matched company recruitment information reveals that the company's skill requirement of "broad business scope" accounts for a high proportion of the recruitment information that satisfies this implicit requirement, so the first prompt word is not adjusted. If, in unsuccessfully matched company recruitment information, the company's skill requirement of "broad business scope" accounts for a high proportion of the recruitment information that satisfies this implicit requirement, the first prompt word is adjusted to: "When a job seeker's resume contains the description 'challenging tasks,' its association weight with the implicit requirement feature 'hopes the company has a wide range of business operations' is reduced." If adjustments to the second prompt are needed, the process is similar to that of the first prompt. That is, if the company and job seeker have a long-term employment relationship, the company's implicit needs are considered to match the job seeker's qualifications in their resume. This indicates that the implicit needs output by the first language model based on the company's talent requirements and the second prompt are correct and require no adjustment. Conversely, if the implicit needs derived from combinations of premature termination of the employment relationship do not match the job seeker's qualifications in their resume, this indicates that the implicit needs output by the first language model based on the company's talent requirements and the second prompt are inaccurate and require adjustment.

[0053] After updating the prompt word rules, the process moves to the profile regeneration stage. At this point, the optimized first prompt word, incorporating the new semantic understanding rules, is input again into the first language model along with the job seeker's original resume information to generate an adjusted job seeker profile. Similarly, the optimized second prompt word, along with company recruitment information, generates an adjusted company profile. This process essentially allows the first language model to learn from historical experience the user's true preferences and, based on the user's original text, produce a user profile that better reflects those preferences.

[0054] Finally, using these two optimized profiles, along with job seeker resumes and company recruitment information, the two-way matching degree between job seekers and companies is recalculated. This updated matching degree, because the profiles of both parties have been calibrated based on historical experience, is theoretically more accurate in predicting the degree of compatibility in reality. Ultimately, based on this updated two-way matching degree and combined with the original explicit structured needs of both parties, a new recommendation list is generated and displayed to the user. This completes a full "learn-optimize-execute" cycle.

[0055] In addition, in this embodiment of the application, in order to train the second language model to automatically extract and combine the competency requirements of both parties from historical matching cases with success or failure labels, a supervised learning-based fine-tuning process needs to be constructed. The entire training process revolves around sample construction and model optimization. First, a large number of input-output pairs with clear result labels need to be prepared as training data. Each input is composed of job seeker resume information, company recruitment information, and matching result labels (such as successful onboarding or stable employment, rejection of offer or short-term resignation, etc.). The output is the combination of key competency requirements manually annotated from the corresponding text. For successful cases, the company's competency requirements for talent and the competency conditions in the talent's resume information, as well as the talent's competency requirements for the company and the competency conditions in the company's recruitment information, need to be extracted. For failed cases, the competency requirements and competency conditions of both parties are also extracted, but they are classified as failures. The annotation format usually adopts a structured form such as JSON or a fixed template to ensure consistency. Next, sequence-to-sequence fine-tuning is performed based on pre-trained language models (such as T5 or GPT series). Input text sequences are fed into the model, and the cross-entropy loss function is used to gradually approximate the manually labeled target output. During training, parameters are adjusted using mini-batch gradient descent, while data augmentation techniques such as synonym replacement and sentence transformation are employed to increase sample diversity and improve robustness. The model needs to learn to distinguish output types based on the result labels in the input and accurately locate key phrases related to the requirements from unstructured text, then organize them into successful or failed combinations according to a specified template. Finally, on the validation set, the accuracy and completeness of the extraction are verified using automatic metrics such as ROUGE and BLEU combined with human evaluation. Error analysis is performed to address issues such as omissions, mis-extractions, or formatting errors. If necessary, specific domain data is supplemented or labeling specifications are adjusted. After multiple iterations until the model performance stabilizes, the trained second language model can automatically generate the required literacy requirement combinations for new input feedback data, providing a basis for subsequent profile calibration or matching analysis.

[0056] In some embodiments of this application, the explicit structured features may optionally include hard requirements set by the talent or enterprise, which may include at least one of education level, years of work experience, salary range, and work location.

[0057] The explicit structured features in this application are information directly set by the user, with clear numerical or conditional boundaries, and which can be confirmed without deep semantic analysis, such as "Master's degree or above", "more than five years of Java experience", "monthly salary of 20k to 30k", and "work location in Beijing". These features complement the implicit needs of enterprises or job seekers (such as corporate culture preferences, expectations of team atmosphere, and understanding of stress resistance), together forming a complete user profile.

[0058] The implementation methods and advantages of the embodiments of this application have been described above through multiple examples. The specific processing procedures of the embodiments of this application are described in detail below with reference to specific examples.

[0059] Figure 3 This is a schematic diagram illustrating the processing flow of a recommendation method based on two-way implicit needs in recruitment, as exemplified in this application. Combined with... Figure 3 As shown, the recommendation method based on bidirectional implicit requirements in this application includes the following steps: Step 1: Initial Bidirectional Latent Feature Extraction. Using the first major language model, extract the first set of latent features (the set of latent needs features of the company) of the target talent from the resume information to generate an initial talent profile; using the first major language model, extract the second set of latent features (the combination of latent needs features of the job seeker) of the target talent from the relevant information of the target company to generate an initial company profile; Step 2: Two-way matching and priority recommendation. Calculate the matching degree between the initial talent profile and the enterprise, and the matching degree between the initial enterprise profile and the talent; combine the two matching degrees, and then combine them with the explicit structured features set by the enterprise and the user to generate a comprehensive two-way matching degree; based on the two-way matching degree, generate and display a recommendation list to the user; Step 3: Collection of Interaction Data and Decision Results. The system continuously monitors and collects user interaction behavior and decision result data triggered by the recommendations in Step 2; the data includes, but is not limited to: positive feedback data: talent onboarding, long-term stable employment of talent, talent receiving good performance evaluations, company issuing offer letters, company repeatedly hiring talent with similar implicit characteristics; negative feedback data: talent refusing interviews or offers, company rejecting talent, talent leaving the company for a short period of time; Step 4: Dynamically Update Feature Extraction Model Strategy. As a supplement to this application, the system also includes a dynamic calibration module for optimizing the accuracy of expected feature extraction based on historical matching decision results. The workflow of this calibration module includes: performing bias analysis based on user decision data generated in Step 3, such as a company hiring or rejecting a candidate, or a candidate accepting or rejecting an offer; comparing the decision results with the initial expected profiles of relevant candidates or companies to identify patterns of systematic bias. For example, it may be found that the characteristics of the candidates actually hired by a certain type of company differ in a generalizable way from the expected features extracted from their text. Finally, calibration parameters are generated. Specifically, based on the above analysis, expected feature extraction calibration prompts are generated for specific industries, job types, or text descriptions. These calibration prompts are used to fine-tune the parsing focus of the first language model when extracting expected features from similar texts in the future, so that the extracted expected profiles are closer to the potential selection criteria of that type of user.

[0060] Step 5: Apply calibration. When performing Step 1 or Step 2 in subsequent steps, the system prioritizes calling calibration prompts that match the current text features, guiding the large model to generate a more accurate desired profile.

[0061] For example, according to the method of this application embodiment, the specific processing procedure may be as follows: During the initial analysis, when the system extracts the expectation of "stress resistance" from the job descriptions of multiple Internet companies, it generally interprets it as "able to overcome technical difficulties." However, the resumes of the talents actually hired by these companies contain numerous descriptions such as "working overtime continuously to overcome projects" and "7×24-hour response." After analyzing this deviation, the dynamic calibration module generates a feature extraction calibration rule for the context of "Internet company" + "stress resistance": "When extracting the expectation related to 'stress resistance' from the text of Internet companies, it should be associated with 'high-intensity, continuous work endurance' with a higher weight." Subsequently, when the system extracts expectations for any new Internet company, it will automatically apply this optimized rule, making its initially extracted 'stress resistance' feature value closer to the actual selection standards of the industry. When performing step 1 extraction for new users or re-extracting after updating the text for old users, the updated model strategy is applied to generate a more accurate initial profile, thereby achieving continuous optimization of the overall system recommendation effect.

[0062] Correspondingly, this application also provides a recommendation device based on the implicit two-way needs in recruitment, with reference to... Figure 4 ,include: The acquisition module 110 is used to acquire a pre-trained first large language model. The first large language model can convert the enterprise quality requirement information in the job seeker's resume information into job seeker implicit requirement features describing the deep preferences of the job seeker based on the input job seeker resume information and the corresponding first prompt words, and generate a job seeker user profile including the job seeker implicit requirement features. The first large language model can also convert the talent quality requirement information in the enterprise recruitment information into enterprise implicit requirement features describing the deep preferences of the enterprise based on the input enterprise recruitment information and the corresponding second prompt words, and generate an enterprise user profile including the enterprise implicit requirement features. The job seeker profile generation module 120 is used to input the job seeker's resume information and the corresponding first prompt words into the first large language model, so that the first large language model generates an initial job seeker profile including the job seeker's implicit needs features. The first prompt words include differentiated semantic understanding rules for the same job seeker's corporate quality needs information in different job search scenarios or industry backgrounds. The enterprise profile generation module 130 is used to input enterprise recruitment information and corresponding second prompt words into the first large language model so that the first large language model generates an initial enterprise profile including the implicit needs of the enterprise, wherein the second prompt words include differentiated semantic understanding rules for talent quality needs information associated with the attributes of the enterprise. The two-way matching degree calculation module 140 is used to calculate the two-way matching degree between job seekers and companies based on the initial job seeker profile, company recruitment information, job seeker resume information and initial company profile. The recommendation list generation module 150 is used to generate a recommendation list and display it to the user based on the explicit structured features in the bidirectional matching degree, job seeker resume information and enterprise recruitment information, wherein the explicit structured features are structured fields that record the user's hard requirements.

[0063] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of various types of information, such as user personal information, device information, log data, etc., all comply with the provisions and requirements of relevant laws and regulations, and the relevant processing methods and scope are open and transparent and do not violate public order and good morals.

[0064] Based on at least one of the above embodiments, the electronic device in the embodiments of this application may be a user terminal device, a server, other computing devices, or a cloud server. Figure 5This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. The electronic device may include a processor 601 and a memory 602 storing computer program instructions. The processor 601 reads and executes the computer program instructions stored in the memory 602 to implement the process or function of any of the methods in the above embodiments.

[0065] Specifically, processor 601 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Memory 602 may include a mass storage device for data or instructions. For example, memory 602 may be at least one of the following: a hard disk drive (HDD), read-only memory (ROM), random access memory (RAM), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, universal serial bus (USB) drive, or other physical / tangible memory storage device. Alternatively, memory 602 may include removable or non-removable (or fixed) media. Furthermore, memory 602 may be internal or external to the integrated gateway disaster recovery device. Memory 602 may be non-volatile solid-state memory. In other words, typically memory 602 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 (e.g., by one or more processors), it can perform the operations described in the methods of the embodiments of this application.

[0066] As an example, Figure 5 The illustrated electronic device may also include a communication interface 603 and a bus 610. The processor 601, memory 602, and communication interface 603 are connected via bus 610 and communicate with each other. Bus 610 may include hardware, software, or both, and may couple components of an online data traffic metering device together. The bus may include at least one of the following: Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), HyperTransport (HT) Interconnect, Industry Standard Architecture (ISA) bus, Infinite Bandwidth Interconnect, Low Pin Count (LPC) bus, memory bus, Microchannel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus, or other suitable bus. Bus 610 may include one or more buses. Although specific buses are described or shown in the embodiments of this application, any suitable bus or interconnection method is contemplated in the embodiments of this application.

[0067] In conjunction with the methods in the above embodiments, this application also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the process or function of any of the methods in the above embodiments.

[0068] This application also provides a computer program product that stores computer program instructions, which, when executed by a processor, implement the process or function of any of the methods described above.

[0069] The flowcharts and / or block diagrams of methods, terminals, systems, and computer program products according to embodiments of this application have been exemplarily described above, and related aspects have been described. It should be understood that each block or combination thereof in the flowcharts and / or block diagrams may be implemented by computer program instructions, by dedicated hardware performing a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine such that these instructions, executed via such processor, enable the implementation of the function / action specified in each block or combination thereof in the flowcharts and / or block diagrams. Such a processor may be a general-purpose processor, a dedicated processor, a special-purpose application processor, or a field-programmable logic circuit.

[0070] The functional blocks shown in the structural block diagrams of this application can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc.; when implemented in software, they are programs or code segments used to perform the required tasks. Programs or code segments can be stored in memory or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. Code segments can be downloaded via computer networks such as the Internet or intranets.

[0071] It should be noted that this application is not limited to the specific configurations and processes described above or shown in the figures. The above descriptions are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the described systems, devices, terminals, modules, or units can be referred to the corresponding processes in the method embodiments, and need not be repeated here. It should be understood that the scope of protection of this application is not limited thereto. Any equivalent modifications or substitutions that can be conceived by those skilled in the art within the scope of the technology disclosed in this application should be covered within the scope of protection of this patent application.

Claims

1. A recommendation method based on bidirectional implicit demand in recruitment, characterized by, include: A pre-trained first language model is obtained. The first language model can transform the enterprise quality requirements information in the job seeker's resume information into implicit needs features of the job seeker describing the deep preferences of the job seeker based on the input job seeker resume information and the corresponding first prompt words, and generate a job seeker user profile including the implicit needs features of the job seeker. The first language model can also transform the talent quality requirements information in the enterprise recruitment information into implicit needs features of the enterprise describing the deep preferences of the enterprise based on the input enterprise recruitment information and the corresponding second prompt words, and generate an enterprise user profile including the implicit needs features of the enterprise. The job seeker's resume information and the corresponding first prompt words are input into the first large language model so that the first large language model generates an initial job seeker profile including the job seeker's implicit needs features. The first prompt words include differentiated semantic understanding rules for the same job seeker's corporate quality needs information in different job search scenarios or industry backgrounds. The company's recruitment information and the corresponding second prompt words are input into the first large language model so that the first large language model generates an initial company profile including the company's implicit needs features. The second prompt words include differentiated semantic understanding rules for talent quality needs information associated with the company's attributes. Based on the initial job seeker profile, company recruitment information, job seeker resume information and initial company profile, calculate the two-way matching degree between job seekers and companies; Based on the bidirectional matching degree, the explicit structured features in the job seeker's resume information and the company's recruitment information, a recommendation list is generated and displayed to the user. The explicit structured features are structured fields that record the user's hard requirements.

2. The method of claim 1, wherein, Also includes: Based on the interaction behavior data of job seeker users and / or enterprise users on the recommendation results, the deviation between the user's actual implicit needs and their initial profile is analyzed. The interaction behavior data includes at least one of browsing history, click behavior, job application, interview invitation, recruitment results, onboarding feedback and performance. Based on the deviation analysis results, update the semantic understanding rules in the first prompt word and / or the second prompt word so that the subsequently obtained implicit demand features are more in line with the user's true preferences.

3. The method of claim 1, wherein, Based on the initial job seeker profile, company recruitment information, job seeker resume information, and initial company profile, the two-way matching degree between job seekers and companies is calculated, including: Calculate the matching degree between the initial job seeker profile and the company's recruitment information to obtain the first matching degree; Calculate the matching degree between the initial company profile and the job seeker's resume information to obtain the second matching degree; Based on the first and second matching degrees, a bidirectional matching degree is generated.

4. The method of claim 1, wherein, The enterprise attributes include at least one of the following: industry type, enterprise size, development stage, job category, and corporate culture tags.

5. The method of claim 1, wherein, Also includes: A second language model is obtained, which, based on the input feedback data, company recruitment information, and job seeker resume information, outputs successfully matched combinations of competency requirements and / or unmatched combinations of competency requirements. The feedback data includes positive and negative feedback data. Positive feedback data includes at least one of the following: talent onboarding records, long-term stable employment records, records of good performance evaluations, records of company offer letters, and records of companies repeatedly hiring talent with similar implicit characteristics. Negative feedback data includes at least one of the following: records of talent refusing interviews or offer letters, records of companies rejecting talent, and records of short-term employee departures. The feedback data, along with the recruitment information of the companies and the resumes of the job seekers contained in the feedback data, are input into the second language model so that the second language model can output successfully matched competency requirement information combinations and / or unmatched competency requirement information combinations. Adjust the semantic understanding rules in the first and / or second prompt words based on the successfully matched literacy requirement information and the unmatched literacy requirement information; The adjusted first prompt word and job seeker resume information are input into the first language model, so that the first language model outputs the adjusted job seeker profile; and the adjusted second prompt word and company recruitment information are input into the first language model, so that the first language model outputs the adjusted company profile. The updated two-way matching degree is calculated using the adjusted enterprise profile, enterprise recruitment information, job seeker profile, and job seeker resume information. Based on the updated two-way matching degree, job seeker resume information, and explicit structured features in company recruitment information, an updated recommendation list is generated and displayed to the user.

6. The method of claim 1, wherein, The explicit structured features include hard requirements set by the talent or the company, which include at least one of the following: education level, years of work experience, salary range, and work location.

7. A recommendation device based on two-way implicit demand in recruitment, characterized by, include: The acquisition module is used to acquire a pre-trained first language model. The first language model can convert the enterprise quality requirements information in the job seeker's resume information into implicit needs features of the job seeker describing the deep preferences of the job seeker based on the input job seeker resume information and the corresponding first prompt words, and generate a job seeker user profile including the implicit needs features of the job seeker. The first language model can also convert the talent quality requirements information in the enterprise recruitment information into implicit needs features of the enterprise describing the deep preferences of the enterprise based on the input enterprise recruitment information and the corresponding second prompt words, and generate an enterprise user profile including the implicit needs features of the enterprise. The job seeker profile generation module is used to input the job seeker's resume information and the corresponding first prompt words into the first large language model, so that the first large language model generates an initial job seeker profile including the job seeker's implicit needs features. The first prompt words include differentiated semantic understanding rules for the same job seeker's corporate quality needs information in different job search scenarios or industry backgrounds. The enterprise profile generation module is used to input enterprise recruitment information and corresponding second prompt words into the first large language model, so that the first large language model generates an initial enterprise profile including the implicit needs of the enterprise, wherein the second prompt words include talent quality information differentiation semantic understanding rules associated with the attributes of the enterprise; The two-way matching degree calculation module calculates the two-way matching degree between job seekers and companies based on the initial job seeker profile, company recruitment information, job seeker resume information, and initial company profile. The recommendation list generation module is used to generate a recommendation list and display it to the user based on the bidirectional matching degree, the job seeker's resume information and the explicit structured features in the company's recruitment information. The explicit structured features are structured fields that record the user's hard requirements.

8. An electronic device, comprising: The electronic device is a terminal device or a server. The electronic device includes a processor and a memory storing computer program instructions. When the electronic device executes the computer program instructions, it implements the method as described in any one of claims 1-6.

9. A computer program product, characterised in that, It includes computer program instructions that, when executed, implement the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions that, when executed, implement the method as described in any one of claims 1-6.

Images