Resume preliminary screening method and system, electronic equipment and storage medium

Abstract

The invention relates to a resume preliminary screening method and system, electronic equipment and a storage medium. The method comprises the following steps: constructing a corresponding talent portrait based on position information of a target position; acquiring a position label of the target position and a resume label of each resume; performing primary matching of the resumes and the positions based on the resume labels and the position labels, wherein the resumes meeting the primary matching requirement form a primary resume set; matching the talent portrait with resume information of each resume in a first-level resume set to obtain matching detail information of each resume; comprehensively analyzing the position information of the target position, the resume information of each resume and the matching detail information of each resume to obtain comprehensive matching result information of each resume and the target position; and obtaining feedback information of the recruitment user on the delivered resume, and dynamically optimizing the talent portrait based on the feedback information. According to the invention, the resume screening efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a resume screening method, system, electronic device, and storage medium. Background Technology

[0002] With the rapid development of computer technology and the widespread use of the internet, companies now typically publish job postings through their own online platforms or third-party recruitment platforms and receive resumes online. For some popular positions, once the job posting is published, companies often face the dual challenges of a surge in resume submissions (hundreds to thousands per day) and low efficiency in human resource screening. Existing recruitment platforms' initial resume screening functions mainly rely on keyword tag matching mechanisms (such as years of experience and skill keywords), which are essentially static and rigid Boolean logic judgments, only capable of processing explicit structured fields. With the development of natural language processing, some solutions have emerged that use large language models to perform layered semantic matching between job descriptions and resumes to handle unstructured content.

[0003] Research on resumes, job descriptions, and recruitment results reveals that HR professionals often struggle to accurately express talent needs in written job descriptions, leading to significant homogenization in descriptions for similar roles. However, actual recruitment requirements vary. This inherent gap between job descriptions and company needs results in less than ideal resume screening effectiveness. Summary of the Invention

[0004] In view of the technical problems existing in the prior art, the present invention proposes a resume screening method, system, electronic device and storage medium to improve the efficiency and accuracy of resume screening.

[0005] To address the aforementioned technical problems, according to one aspect of the present invention, a resume initial screening method is provided for selecting resumes that match a target position from an initial resume set, wherein the method includes: Build a corresponding talent profile based on the job information of the target position; Obtain the job tags for the target position and the resume tags for each resume; The initial matching of resumes and jobs is performed based on resume tags and job tags. Resumes that do not meet the initial matching requirements are filtered out from the initial resume set, and the resumes that meet the initial matching requirements constitute the first-level resume set. The talent profile is matched with the resume information of each resume in the primary resume set to obtain the matching details of each resume; The job information for the target position, the resume information for each resume, and the matching details for each resume are comprehensively analyzed to obtain the overall matching result information between each resume and the target position; and Obtain feedback from recruiting users regarding submitted resumes, and dynamically optimize the talent profile based on this feedback.

[0006] Optionally, the steps for building a talent profile based on the job information of the target position include: Tag extraction is performed on the job information of the target job to obtain multiple job tags; Construct a first prompt text, which includes at least a description of the task to extract the core job requirements, one or more core job requirements to be extracted, and the corresponding extraction principles. Input the first prompt text and the job information of the target job into the first language model; The large language model extracts the core requirements of the job information based on the first prompt text and outputs the extraction results. Receive the extraction results from the first major language model and use them as the core requirement information for the target position; and A talent profile is constructed by combining the tag content of multiple job tags and one or more core requirements of the target job. The job tags and core requirements are talent profile elements, and the tag content of the job tags and the core requirements are feature values ​​of the corresponding talent profile elements.

[0007] Optionally, the step of optimizing the talent profile based on feedback information includes: A resume sample set was constructed based on feedback information; The rejection rate for each talent profile element is calculated based on a sample set of resumes. The rejection rate of the corresponding feature values ​​of the talent profile elements is evaluated based on the optimization conditions of the aforementioned talent profile elements; and The rejection rate in response to the feature values ​​of talent profile elements satisfies the optimization conditions for the feature values ​​of talent profile elements of the target position.

[0008] Optionally, the step of matching the talent profile with the resume information of each resume in the primary resume set to obtain matching details for each resume includes: Construct a second prompt text, which includes at least a description of the matching task, matching principles, and output format; The second prompt text, talent profile, and resume information for each resume are input into the second large language model; and The second language model matches the talent profile and the resume information of each resume according to the second prompt text, and outputs the matching details according to the output format.

[0009] Optionally, the method further includes: optimizing the second prompt text based on feedback information, specifically including: A second sample set is constructed based on the feedback information; the second sample includes resume information and matching result information; and Add the second sample to the second prompt text, and add a description of learning according to the sample to the matching task description.

[0010] Optionally, the steps for comprehensively analyzing the job information of the target position, the resume information of each resume, and the matching details of each resume include: Construct a third prompt text, which includes at least a description of the matching analysis task and an output format; The third prompt text, the job information of the target position, the resume information of each resume, and the matching details of each resume are input into the third language model; and The third language model comprehensively analyzes the job information of the target position, the resume information of each resume, and the matching details of each resume according to the third prompt text, and outputs the comprehensive matching result information between the resume and the target position according to the output format.

[0011] Optionally, the method further includes: optimizing the third prompt text based on feedback information, specifically including: A third sample set is constructed based on feedback information; the third sample includes resume information and comprehensive matching result information; and Add the third sample to the third prompt text, and add a description of learning according to the sample to the matching analysis task description.

[0012] Optionally, the comprehensive matching result information includes matching level and its basis details, wherein the basis details of the matching level include matching items and / or non-matching items, wherein the matching items and non-matching items respectively include matching dimensions, matching details under the matching dimensions, and corresponding job partial information.

[0013] Optionally, the method further includes: classifying the resumes in the first-level resume set based on the matching level to obtain resume sets with different matching levels.

[0014] According to another aspect of the present invention, the present invention also provides a resume screening system for selecting resumes that match a target position from an initial resume set, wherein the system comprises: The talent profiling module is configured to build a corresponding talent profile based on the job information of the target position, and optimize the talent profile based on the feedback information from the initial resume screening. The tag acquisition module is configured to obtain the job tags for the target job and the resume tags for each resume. The initial screening module is configured to perform a preliminary matching of resumes and positions based on resume tags and job tags. It filters out resumes that do not meet the preliminary matching requirements from the initial resume set, and the resumes that meet the preliminary matching requirements form a first-level resume set. The core matching module is configured to match the talent profile with the resume information of each resume in the primary resume set to obtain matching details for each resume; and The comprehensive matching module is configured to perform comprehensive analysis on the job information of the target position, the resume information of each resume, and the matching details of each resume to obtain the comprehensive matching result information between each resume and the target position.

[0015] According to another aspect of the present invention, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a set of computer program instructions, which implements the aforementioned resume screening method when the processor executes the set of computer program instructions in the memory.

[0016] According to another aspect of the present invention, the present invention also provides a computer-readable storage medium, wherein a computer program instruction set is stored on the computer-readable storage medium, and the computer program instruction set, when executed by a processor, implements the aforementioned resume screening method.

[0017] According to another aspect of the present invention, the present invention also provides a computer program product comprising a computer program instruction set, which, when executed by a processor, implements the aforementioned resume screening method.

[0018] This invention breaks through the limitations of traditional recruitment systems that rely on explicit tags (such as education level and years of experience) for matching. By dynamically and adaptively expressing the changes in the talent needs of recruiting users through talent profiles, it improves the accuracy of person-job matching and recruitment efficiency, and achieves faster and more accurate talent recruitment and onboarding. Attached Figure Description

[0019] The preferred embodiments of the present invention will now be described in further detail with reference to the accompanying drawings, wherein: Figure 1 This is a system device schematic diagram of a recruitment platform according to an embodiment of the present invention; Figure 2 This is a flowchart of a resume screening method according to an embodiment of the present invention; Figure 3 This is a flowchart of a method for constructing a talent profile corresponding to a target position according to an embodiment of the present invention; Figure 4This is a flowchart of a method for generating matching details of a resume using a large language model according to an embodiment of the present invention; Figure 5 This is a flowchart of a method for comprehensive analysis using a large language model according to an embodiment of the present invention; Figure 6 This is a flowchart of a method for optimizing talent profiles based on feedback information according to an embodiment of the present invention; Figure 7 This is a schematic diagram of a resume screening system according to an embodiment of the present invention; and Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the following detailed description, reference can be made to the accompanying drawings, which form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to implement the technical solutions of the present application. It should be understood that other embodiments can also be used, or structural, logical, or electrical changes can be made to the embodiments of the present application. Furthermore, the terms "first," "second," etc., in this invention are merely for distinguishing different technical features with the same name and do not represent sequential numbers.

[0022] See Figure 1 , Figure 1This is a system device schematic diagram of a recruitment platform according to an embodiment of the present invention. In this embodiment, the recruitment platform includes a user terminal and a server terminal 3. The user terminal is installed as a client on a user terminal device and is divided into a recruitment terminal 1 and a job seeker terminal 2, as shown in the figure. The recruitment terminal 1 is installed on different recruitment user devices, such as the first recruitment user device 1a, the second recruitment user device 1b, etc., as shown in the figure. The job seeker terminal 2 is installed on job seeker user devices, such as the first job seeker user device 2a, the second job seeker user device 2b, the third job seeker user device 2c, etc., as shown in the figure. The server terminal 3 includes multiple servers, such as the first server 3b, the second server 3c, and the service management terminal device 3a, etc. The user devices and terminal devices here are, for example, desktop computers, smartphones, laptops, tablets, etc. Recruiting users publish job postings through the recruitment terminal, and job seekers can search for published jobs through the job seeker terminal and submit job applications to the recruiters who published job postings. Usually, when submitting a job application, a resume is also submitted to the recruiter. For recruiters, after publishing a job posting, they will receive resumes from job seekers. For some popular positions, a large number of resumes will be received, which poses a challenge for recruiters to screen talent. This invention provides a method and system for initial resume screening, enabling recruiters to initially select resumes that meet their recruitment needs.

[0023] See Figure 2 , Figure 2 This is a flowchart of a resume screening method according to an embodiment of the present invention. In this embodiment, when one or more resumes for a published job are received, these resumes are defined as an initial resume set. The initial resume set can contain one or more resumes. The resume screening method described in this embodiment can be executed each time a resume is received, or at a set frequency, such as once a day, or when a sufficient number of resumes are collected. Therefore, in the following description, the timing of the execution of the method described in this invention will not be repeated, and the number of resumes in the initial resume set is not limited. The resume screening method in this embodiment includes the following steps: Step S11: Construct a corresponding talent profile based on the job information of the target position.

[0024] Step S12: Obtain the job tag for the target job and the resume tag for each resume.

[0025] Step S13: Perform initial matching of resumes and positions based on resume tags and job tags. Filter out resumes that do not meet the initial matching requirements from the initial resume set. The resumes that meet the initial matching requirements constitute the first-level resume set.

[0026] Step S14: Match the talent profile with the resume information of each resume in the first-level resume set to obtain matching details for each resume.

[0027] Step S15: Perform a comprehensive analysis to obtain the overall matching result information between each resume and the target position. Specifically, perform a comprehensive analysis on the job information of the target position, the resume information of each resume, and the matching details of each resume to obtain the overall matching result information between each resume and the target position.

[0028] Step S16: Obtain feedback information from recruiting users on the resume, and dynamically optimize the talent profile based on the feedback information.

[0029] To improve the alignment of screened resumes with recruiters' talent needs, this invention, in step S11, constructs a corresponding talent profile based on the job information of the target position. Specifically, see [link to relevant documentation]. Figure 3 , Figure 3 This is a flowchart illustrating a method for constructing a talent profile for a target position according to an embodiment of the present invention. Specifically, it includes the following: Step S111: Extract job tags from the job information of the target job.

[0030] Step S112: Construct a first prompt text, which includes at least a description of the task to extract the core job requirements, one or more core job requirements to be extracted, and the corresponding extraction principles.

[0031] Step S113: Input the first prompt text and the job information of the target job into the first large language model.

[0032] Step S114: The large language model extracts the core job requirements information from the job information according to the first prompt text.

[0033] Step S115: Combine job tags and core requirement information of the target job to form a talent profile, wherein the job tags and core requirements are talent profile elements, and the tag content of the job tags and the core requirement information are the feature values ​​of the corresponding talent profile elements.

[0034] In step S111, various extraction techniques can be employed during label extraction, such as rule-based, statistical extraction methods, or models suitable for different information categories. For example, for key information with relatively stable structures or obvious features, such as age, education level, years of experience, and skills, rule-based methods can be used for extraction. These rules are typically manually defined, such as regular expressions, keyword dictionaries, template matching, and dependency parsing patterns. Alternatively, traditional sequence models such as CRF and BiLSTM-CRF, or pre-trained models such as BERT-CRF, RoBERTa-CRF, and Chinese-BERT-NER, can be used.

[0035] When extracting keywords representing job titles or skill terms as labels, statistical methods such as TF-IDF and TextRank can be used, as well as pre-trained models such as BERT, SimCSE, and Sentence-BERT, and supervised deep learning models such as KeyBERT and BERT+Classifier. For dependencies between job titles and skills, or between responsibilities and abilities, which fall under relation extraction, dependency parsing and pattern matching methods can be used, or neural network relation classification models such as BERT+Softmax and BiLSTM+Attention can be employed. Alternatively, Large Language Models (LLMs) can be used to accomplish the above extraction tasks.

[0036] In one embodiment, the recruitment platform provides recruiters with a preliminary screening field setting page on the recruitment end, offering various preliminary screening tag fields, such as industry, place of residence, language skills, etc. Recruiters can select the corresponding preliminary screening tag fields according to their interests and set the corresponding field values. Therefore, after extracting the tags, the system also obtains the recruiter's settings on the preliminary screening field setting page and marks the currently extracted tags with attribute information. For example, if the tag is a field set by the recruiter on the preliminary screening field setting page, it is marked as "required" in the attribute information; otherwise, it is marked as "not required," thus highlighting the tags that recruiters particularly care about.

[0037] In this invention, to preserve the contextual semantics, competency dimensions, and implicit conditions within job postings, a large language model is employed to extract core requirements from the job postings. This allows for in-depth analysis of the job's requirements and parsing of unstructured job descriptions written by recruiters (e.g., HR professionals). For example, the extracted core requirement is "must have over 3 years of experience in internet e-commerce operations and have led user growth projects," rather than being converted into discrete labels like "years of work experience = 3 years" or "industry = internet." This preserves the original semantic expression and resolves the information bias caused by information distortion in traditional methods.

[0038] To extract core requirement information using a large language model, a first prompt text is constructed in step S112. This prompt text includes at least a description of the task to be extracted from the core job requirements, one or more core job requirements to be extracted, and the corresponding extraction principles. For example, the task description specifies the role of the large language model, the specific extraction task, input content requirements, and output content requirements. To ensure the large model can clearly extract the core requirement information, the prompt text also includes one or more specific core job requirements and their corresponding extraction principles. For example, the aforementioned core requirements could be "professional requirements," "skill requirements," "experience requirements," "language requirements," etc.

[0039] Extraction principles include those applicable to all core requirements, as well as different extraction principles adapted to specific core requirements. Principles applicable to all core requirements include, for example, preserving complete semantics rather than discrete information. For instance, for the core requirement "language requirement," the extraction principle requires identifying the smallest complete phrase containing the language name and ability description; for the core requirement "experience requirement," the extraction principle requires extracting the smallest clause containing complete semantics, such as "possessing 5 years of Java development experience." This invention preserves contextual information by specifying in the prompt text that complete semantics should be preserved during extraction. When used as input text for subsequent matching in a large language model, this can guide the large language model to generate more accurate, reliable, and traceable matching results.

[0040] To ensure the accuracy and completeness of the content extracted by the large language model, different extraction principles are included in the prompt text for different core requirements. For example, for the core requirement of "experience requirement," the prompt text includes keywords such as "experience," "background," and "experience." As another example, for the core requirement of "skill requirement," the extraction principles include necessity verification and relevance verification. For instance, the necessity verification states that the large language model must consider the core functions of the job and retain only the technical skills necessary to perform the work. The relevance verification states that the extracted skill descriptions must have a direct supporting relationship with the job requirements explicitly listed in the job information description, thus ensuring that the content extracted by the large language model matches the actual job information.

[0041] Furthermore, to facilitate the understanding of the output format by the large language model, corresponding examples can be shown in the prompt text. For example, the output format example is: "Professional Requirements": "xxx", "Work Experience Required": ["xxx", "xxx"...] Skill Requirements: ["xxx", "xxx"...] In a further implementation, restrictive instructions may be included in the prompt text. For example, the large language model may be required to extract only the core requirements from the input job information, and all extracted content must be substrings of the original job information, with no characters added, deleted, or modified, including punctuation marks.

[0042] Furthermore, to better describe the talent profile, in one embodiment of the present invention, the extracted content can be categorized, such as mandatory requirements for a given condition, priority items for which advantages exist, etc. Therefore, the first prompt text can instruct the large language model to classify each extracted core requirement and specify the category in the output. The following is an example of the core requirements extracted by the first large language model: { "Experience Required": { "must": [ "At least 3 years of work experience in machine learning projects in manufacturing or other related industries" ], "Priority": [ "There are successful case studies of deploying machine learning models to production environments." ], Skill Requirements: { "must": [ "Proficient in Java language" "Familiar with 3D modeling, possessing a deep understanding of the representation and editing of 3D structures." ], "Priority": [ "Familiar with Alibaba Cloud's PAI and other machine learning platforms" ] } } This invention combines extracted job tags and core requirement information to form a preliminary talent profile. Each job tag and each core requirement is called a talent profile element; therefore, the names of the tags and core elements are called talent profile element names, while the specific content of the job tags and core requirements are called feature values ​​of the talent profile elements. The following is an example of a preliminary talent profile:

[0043] In order to enable the primary language model to perform the task of extracting the core required information well, the primary language model can be trained in advance. The training process includes the following steps: First, training data is collected to construct a training sample set. Each sample includes a job description text and extracted core requirements text.

[0044] Then, a prompt text is constructed. This prompt text includes a learning task description of the core requirements, the core requirements extraction task, one or more core requirements for the position to be extracted, and the corresponding extraction principles. The learning task description requires the primary language model to learn the core requirements based on the sample data.

[0045] The prompt text and sample input are fed into the first language model.

[0046] After the first language model has completed its learning, a test sample set is constructed. The test samples include the first sub-sample used as input and the second sub-sample of the extraction result that should be output corresponding to the first sample.

[0047] When testing the learning effect of the first language model, a first prompt text is constructed, and the first prompt text and the first subsample of the test sample are input into the first language model. The extraction result output by the first language model is received, and the extraction result output by the first language model is compared with the corresponding second subsample.

[0048] Based on the comparison results, adjust the content of the first prompt text, then send the first prompt text and the first subsample of the test sample to the first large language model, and receive the extraction result output by the first large language model. Compare the extraction result output by the first large language model with the corresponding second subsample again. Repeat this process until the extraction result output by the first large language model matches the second subsample. At this point, the training of the first large language model is complete, and the first prompt text is saved to the database. In practical applications, when constructing the first prompt text in step S112, the stored first prompt text content can be read from the database, or the stored first prompt text file can be directly called.

[0049] In step S12, the extraction of job tags and resume tags is similar to the tag extraction when generating the talent profile, and will not be described again here. In one specific embodiment, job tags are extracted only from the fields and their contents set by the recruiting user on the initial screening field setting page. For example, when extracting job tags, the information set by the recruiting user based on the initial screening field setting page is queried to obtain the initial screening fields and field values ​​set by the recruiting user, and the initial screening fields and field values ​​are used as job tags, such as: {Industry: xxx}, {Language Skills: xx}. In another embodiment, the recruitment platform sets multiple job tag fields, such as industry, place of residence, language skills, expected salary, expected working hours, etc., and the job tags are extracted from the job information of the target job based on these preset job tag fields. Alternatively, tags with the attribute information of "required" and their contents can be obtained from the talent profile.

[0050] For resume tags, in one embodiment, the recruitment platform sets multiple resume tag fields, such as: major, industry, years of work experience, address, highest education level, language skills, etc. When receiving a resume submitted by a job seeker for a specific position, the platform extracts the content of the resume tag fields from the resume information according to the preset resume tag field information. The content that matches the resume tag field information constitutes the resume tags of the resume. For example: {Industry: xxx}, {Major: xx}, {Language Skills: xx}, {Years of Work Experience: xx}, etc.

[0051] In step S13, a preliminary matching of resumes and jobs is performed by comparing the resume tags and job tags of each resume. Taking a resume as an example, each job tag is used as a target, and each resume tag is semantically compared with the target. If the semantics are similar, it is considered a match for that job tag; if the semantics differ significantly, it is considered a mismatch. Then, the next job tag is used as the target for semantic matching, until all job tags are matched. In another embodiment, job tags and resume tags are converted into vectors, and the distance between the job tag vector and the resume tag vector is calculated. When the distance between the two vectors is greater than a threshold, the resume tag and job tag do not match; when the distance between the two vectors is less than the threshold, the resume tag and job tag match.

[0052] Then, resumes that do not meet the initial matching requirements are filtered out from the initial resume set. The initial matching requirements may include, for example, that the resume tags match all job tags. Alternatively, the resume tags may match some job tags. Or, the initial matching requirements may be determined based on the number of resumes received. For example, when the number of received resumes is less than a threshold, the initial matching requirement is that the resume tags match some job tags; when the number of received resumes is greater than the threshold, the initial matching requirement is that the resume tags match all job tags. The "some job tags" mentioned here refer to specific job tags. For example, specific job tags may be industry-specific; different industries have different specific job tags, thus allowing resumes that do not meet the relevant requirements of a particular industry to be filtered out during the initial screening.

[0053] In step S14, in one embodiment, a large language model can be used to generate matching details for each resume. See also Figure 4 , Figure 4 This is a flowchart of a method for generating matching details of a resume using a large language model according to an embodiment of the present invention. The method includes the following steps: Step S141: Construct a second prompt text, which includes at least a description of the matching task, matching principles, and output format.

[0054] Step S142: Input the second prompt text, talent profile, and resume information of each resume into the second language model.

[0055] In step S143, the second language model matches the talent profile and the resume information of each resume according to the second prompt text, and outputs the matching details according to the output format.

[0056] In one embodiment, the second large language model is a trained large language model. During the training process, prompt text is obtained, similar to the first prompt text mentioned above. After the second large language model is trained, the prompt text is stored in the database. When creating the second prompt text in step S141, the stored prompt text file can be retrieved directly from the database as the second prompt text, or the prompt text content can be read to obtain the second prompt text in step S141.

[0057] The second prompt text in this embodiment includes a matching task description, a matching requirement description, and output requirements. The matching task description defines the role of the second major language model, such as an expert in the recruitment field. The matching requirement description typically includes feature values ​​of multiple talent profile elements, as well as matching principles and standards for determining the degree of matching for each talent profile element. For example, the matching requirement description includes the talent profile element "work experience requirement," and the corresponding matching principle is, for example, "The target job information may have multiple work experience requirements, which need to be judged item by item." The corresponding standard for determining the degree of matching is, for example, "A candidate only needs to meet any one sub-condition of a certain requirement to be considered a match for that requirement." The output requirements define the content and format to be output. In one embodiment, for subsequent data processing, the output format is defined as JSON. The content to be output includes matching details for each talent profile element, including but not limited to matching items, whether a match is made, and the reason for matching. The matching items here are, for example, the aforementioned talent profile elements.

[0058] The following is the matching details of a resume output by the second largest language model: {"matchItems":[ { "matchPoint":"Professional", "matchRes":"Match", "matchDetail":"The candidate has a computer-related major and meets the job requirements." "imageDetail":"Computer-related majors" }, { "matchPoint":"Work Experience 1", "matchRes":"Match", "matchDetail":"The candidate has worked for 5 years in a manufacturing company, primarily responsible for the practical application and development of machine learning." "imageDetail": "At least 3 years of work experience in machine learning projects in manufacturing or other related industries" }, { "matchPoint":"Skill Requirement 1", "matchRes":"No match", "matchDetail":"The candidate primarily works on large-scale model text generation and has no 3D modeling experience." "imageDetail": "Familiar with 3D modeling, possesses in-depth understanding of the representation and editing of 3D structures." } ]} The matchPoint is the matching item, corresponding to the talent profile element; matchRes is the matching result of the matching item; matchDetail is the specific description of the match, that is, the reason or cause for obtaining the matching result; and imageDetail is the feature value of the talent profile element, that is, the specific content.

[0059] In another embodiment, three levels of matching and judgment criteria can be set, each represented by a symbol. For example, if the relevant content in the resume is completely identical to the talent profile elements in semantic alignment, it is judged as a perfect match, represented by the number 2; if some parts are identical, it is judged as a partial match, represented by the number 1; if they are completely different, it is judged as a mismatch, represented by the number 0. Correspondingly, the output requirements specification indicates that matching symbols should be marked in the output, and the meaning of each symbol should be given, as shown in the figure below: { "match": “0 / 1 / 2” # 2: indicates a complete match; 1: indicates a partial match; 0: indicates no match } In step S15, a large language model can be used to perform comprehensive analysis on the job information of the target position, the resume information of each resume, and the matching details of each resume, and a matching report can be output. Specifically, see... Figure 5 , Figure 5 This is a flowchart of a method for comprehensive analysis using a large language model according to an embodiment of the present invention. The method includes the following steps: Step S151: Construct a third prompt text, which includes at least a description of the matching analysis task and an output format.

[0060] In step S152, the third prompt text, the job information of the target job, the resume information of each resume, and the matching details of each resume are input into the third language model.

[0061] Step S153: Receive the comprehensive matching result information output by the third language model. The third language model comprehensively analyzes the job information of the target position, the resume information of each resume, and the matching details of each resume according to the third prompt text, and outputs the comprehensive matching result information between the resume and the target position according to the output format.

[0062] In one embodiment, the matching analysis task description in the third prompt text includes at least a requirement for the large language model to list matching and non-matching items with the talent profile elements, and to list the corresponding job information. Furthermore, additional items not mentioned in the job information but consistent with the talent profile can also be listed. This invention continuously optimizes the talent profile based on feedback from recruiting users on the selected resumes, identifying points not mentioned in the job information but valued by recruiting users, such as soft skills that job seekers should possess. This allows for the selection of resumes that meet the actual needs of recruiting users during the resume screening process.

[0063] Similarly, the third language model can also be trained. The training process is similar to that of the first language model, and will not be repeated here.

[0064] The following is an example of the comprehensive matching result information between a resume output by the third language model and the target position: { "summary": " ", "matchPoints": [ { "matchDetail": "XXX", "jdRequirement": "XXX" } ], "unmatchPoints": [ { "matchDetail": "XXX", "jdRequirement": "XXX" } ], "addPoints": ["XXX",... Among them, matchPoints represents matching items; unmatchPoints represents non-matching items; matchDetail represents a detailed description of whether a match is made or not; jdRequirement is the descriptive text in the job information; addPoints represents additional items that are not mentioned in the job information but match the talent profile.

[0065] In another embodiment, the matching analysis task description in the third prompt text requires the large language model to perform a comprehensive rating based on the matched items and / or non-matches. The comprehensive matching result information indicates the matching level and its basis details. The basis details for the matching level include the matched items and / or non-matches, wherein the matched items and non-matches respectively include the matching dimension, the matching details under the matching dimension, and the corresponding job local information. The following is a format example of a comprehensive matching result information: { "summary": "XXXX ", "matchLevel": "XXX", "report": { "matchPoints": [ { "title": "XXX", "matchDetail": "XXX", "jdRequirement": "XXXX" } ], "unmatchPoints": [ { "title": "", "matchDetail": "XXXX", "jdRequirement": "XXXX" } ], "addPoints":["XXX",... } } Wherein, matchLevel is the matching level, which in one embodiment is divided into three levels: match, partial match, and no match; Report is a detailed report; matchPoints represents matching items; unmatchPoints represents non-matching items; title is a talent profile element; matchDetail represents a detailed description of the match or non-match; jdRequirement is the descriptive text in the job information; addPoints represents additional items not mentioned in the job information but that match the talent profile; and XXX represents specific content.

[0066] Since the overall matching result in this embodiment includes the matching level of the resumes to the target position, the resumes in the first-level resume set can be classified based on the matching level to obtain resume sets with different matching levels. For example, a set of matched resumes, a set of unmatched resumes, and a set of partially matched resumes, thus facilitating use by recruiting users.

[0067] Through the aforementioned steps S11 to S15, a matching details report for each resume for the target position is obtained. Recruiting users review this matching details report to determine the next steps, such as accepting the resume, scheduling an interview, adding it to the talent pool, or rejecting the resume. The recruitment platform provides recruiting users with an operation page for each resume, displaying the aforementioned matching details report and processing operation buttons. After a recruiting user clicks an operation button, the recruitment platform records the operation information, including rejection, communication, adding to the talent pool, interview, and hiring. In this invention, this operation information is stored as feedback information for resume screening. When set optimization conditions are met, the talent profile is optimized, and a few samples are used to optimize the prompt text of the large language model.

[0068] Therefore, after generating the initial talent profile, the talent profile is further improved based on the information of the recruiting user's actions in submitting resumes to the target position, so that the talent profile is closer to the recruiting user's recruitment needs.

[0069] See Figure 6 , Figure 6 This is a flowchart of a method for optimizing talent profiles based on feedback information according to an embodiment of the present invention. The method includes the following steps: Step S161: Construct a resume sample set based on the feedback information.

[0070] Step S162: Calculate the rejection rate of the feature values ​​of each talent profile element based on the resume sample.

[0071] Step S163: Select one of the talent profile elements as the target talent profile element for processing.

[0072] Step S164: Evaluate the rejection rate of the feature values ​​of the corresponding target talent profile elements based on the optimization conditions of the talent profile elements.

[0073] Step S165: Determine whether the rejection rate of the feature values ​​of the target talent profile elements meets the optimization conditions. If the rejection rate of the feature values ​​of the target talent profile elements meets the optimization conditions, then determine that the feature value is the optimized value and proceed to step S166. If the rejection rate of the feature values ​​of the target talent profile elements does not meet the optimization conditions, then proceed to step S167.

[0074] Step S166: Optimize the feature values ​​of the talent profile elements for the target position.

[0075] Step S167: Determine if there are any unprocessed talent profile elements. If so, return to step S163; otherwise, end the process.

[0076] In one embodiment, in step S161, resumes submitted to the target position and receiving feedback from recruiting users within a recent period are collected. These resumes are then grouped together to construct a resume sample set, meaning the optimization conditions meet a preset time period. For example, the duration of the time period is set and timed; when the timer reaches the set duration, feedback information from resumes within that period is collected. Another example is optimizing the talent profile at a preset frequency, such as once a week. Alternatively, the talent profile can be optimized after collecting a preset number of resume feedback messages. For instance, whenever 10 resume feedback messages are collected, the talent profile is optimized based on these 10 resumes.

[0077] In step S162, the statistical method for the rejection rate varies depending on the feature value type of the talent profile elements. For example, for discrete value features of the tag type (such as education level, gender, etc.), the rejection rate represents the proportion of offers (offer letters) that do not meet the requirements of the tag that are marked as suitable (not rejected). The specific definition is as follows: Unrejection rate = Number of offers marked as suitable / Number of offers submitted For example, if the feature value of the talent profile element "Education" is "Master's degree or above", and there are 10 resumes in the resume sample set that have received offer letters, of which 5 resumes marked as suitable have a Bachelor's degree, then according to its definition, the rejection rate is 0.5, which is 50%. Based on this data, it is clear that the current feature value of the talent profile element "Education" does not accurately describe the recruiter's hiring needs. Therefore, for the feature value of this type of talent profile element, the optimization condition is set to a rejection rate greater than or equal to a threshold, such as 50% in this embodiment. When the rejection rate is greater than or equal to the threshold, the optimization condition is determined to be met. In step S26, the feature value of the talent profile element for the target position is optimized. In this embodiment, the feature value of the talent profile element "Education" is adjusted to "Master's degree or above preferred" or "Bachelor's degree or above".

[0078] For continuous features such as age and work experience, the actual range distribution and the proportion exceeding the continuous range are statistically analyzed. For example, the feature value of the talent profile element "age" is "18-35". When analyzing the sample set, the actual age range of resumes receiving job offers is 18-45 years old, with 3 individuals exceeding the 35-45 range, resulting in a rejection rate of 0.3. Similarly, the optimization condition is set to a rejection rate greater than or equal to a threshold. For continuous features, the threshold can be less than the threshold for discrete features, for example, set to 20%. In this embodiment, the rejection rate of the talent profile element "age" is greater than 20%, therefore, in step S26, the feature value of the talent profile element "age" is optimized. In this case, the actual range, i.e., "18-45", can be used as the feature value.

[0079] For text-based features, such as the talent profile element "experience," the matching results in step S15 determine whether this item in the resume sample matches the talent profile element. During optimization, the number of resumes that do not match but have received job offers from recruiters is counted, and the rejection rate is calculated using the aforementioned formula. When the rejection rate reaches a threshold (e.g., 30%), adjustments are necessary. In step S166, when optimizing the feature values ​​of the talent profile element, the original feature value text can be optimized using a large language model. For example, the original feature value text of the talent profile element, resumes that do not match but have received job offers, and job information are sent to the large language model for optimization.

[0080] As can be seen from the foregoing, this invention dynamically optimizes talent profiles as resumes are screened, making the talent profiles increasingly closer to the actual needs of recruiting users. This not only overcomes the limitations of job information but also effectively improves resume screening efficiency.

[0081] While optimizing talent profiles, the prompt text for the large language model can also be optimized simultaneously. For example, a second sample set can be constructed based on feedback information, including resume information and matching result information. This second sample set is then added to the second prompt text, and an explanation of learning based on the samples is added to the matching task description. Similarly, a third sample set can be constructed based on feedback information, including resume information and comprehensive matching result information; this third sample set is then added to the third prompt text, and an explanation of learning based on the samples is added to the matching analysis task description.

[0082] This invention optimizes the prompt text of a large language model through a few-short mechanism, leveraging feedback from recruiters on the screening results. By incorporating few-short learning, and considering cost and timeliness when providing samples, this invention uses a limited number of recent resumes (e.g., 6), prioritizing an average number of resumes across different matching levels. This approach enhances the large language model's understanding of the matching task in this invention, leading to a better understanding of recruiters' talent needs and effectively improving matching accuracy.

[0083] This invention utilizes a large language model to perform deep semantic parsing and cross-text semantic alignment of resumes and job postings, overcoming the limitations of traditional recruitment systems that rely on explicit tags (such as education level and years of experience) for matching. Because this invention can understand the implicit semantics and contextual relationships within job and resume information, and adapts to changes in recruiters' talent needs based on their feedback, it constructs talent profiles that meet those needs. This talent profile significantly improves the accuracy of person-job matching, enabling faster and more accurate recruitment and onboarding. Furthermore, during the screening process, providing a small number of recruiter feedback samples to the large language model allows it to learn the recruiters' talent needs, further improving the model's matching accuracy.

[0084] On the other hand, the present invention also provides a resume screening system, see [link to relevant documentation]. Figure 7 , Figure 7 This is a schematic diagram of a resume screening system according to an embodiment of the present invention. The system is used to screen resumes matching a target position from an initial resume set. The system includes a talent profiling module 10, a tag acquisition module 20, a preliminary screening module 30, a core matching module 40, and a comprehensive matching module 50. The talent profiling module 10 constructs a corresponding talent profile based on the job information of the target position and optimizes the talent profile based on feedback information from the preliminary resume screening. The tag acquisition module 20 acquires the job tags of the target position and the resume tags of each resume. The preliminary screening module 30 performs a preliminary matching of resumes and positions based on the resume tags and job tags, filtering out resumes that do not meet the preliminary matching requirements from the initial resume set. Resumes that meet the preliminary matching requirements constitute a first-level resume set. The core matching module 40 matches the talent profile with the resume information of each resume in the first-level resume set to obtain matching details for each resume. The comprehensive matching module 50 comprehensively analyzes the job information of the target position, the resume information of each resume, and the matching details of each resume to obtain a comprehensive matching result for each resume and the target position.

[0085] In a further embodiment, the talent profile module 10 constructs a talent profile through a large language model. The construction and optimization processes are described in the preceding description of the method and will not be repeated here.

[0086] In another further embodiment, the core matching module 40 matches the talent profile and each resume using a large language model, and the comprehensive matching module 50 performs a comprehensive analysis of the job information of the target position, the resume information of each resume, and the matching details of each resume using a large language model. For the specific matching process and comprehensive analysis, please refer to the description of the aforementioned method, which will not be repeated here.

[0087] According to another aspect of the invention, the invention also provides an electronic device, see [link to relevant documentation]. Figure 8 , Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. The electronic device can be implemented as a server in the aforementioned application system, including a processor 601 and a memory 602. The memory 602 stores a program instruction set, and the aforementioned resume screening method is implemented when the processor 601 executes the program instruction set in the memory 602.

[0088] Specifically, the processor 601 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0089] Memory 602 may include mass storage for data or instructions. For example, and not limitingly, memory 602 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 602 may include removable or non-removable (or fixed) media. Where appropriate, memory 602 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 602 is non-volatile solid-state memory.

[0090] The memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the CV screening method provided by this invention.

[0091] In one example, the electronic device may also include a communication interface 603 and a bus 604. The processor 601, memory 602, and communication interface 603 are connected via the bus 604 and communicate with each other.

[0092] The communication interface 603 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of the present invention.

[0093] Bus 604 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel 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 Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 604 may include one or more buses. While specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0094] The present invention also provides a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement any of the resume screening methods described in the foregoing embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in connection with an instruction execution system, apparatus, or device. The storage medium can be a transient computer-readable storage medium or a non-transitory computer-readable storage medium. Non-transitory computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical discs based on CD, DVD, or Blu-ray technology, and persistent solid-state storage such as flash memory and solid-state drives.

[0095] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of the invention.

Claims

1. A resume initial screening method for selecting resumes that match a target position from an initial resume set, wherein, The method includes: Build a corresponding talent profile based on the job information of the target position; Obtain the job tags for the target position and the resume tags for each resume; The initial matching of resumes and jobs is performed based on resume tags and job tags. Resumes that do not meet the initial matching requirements are filtered out from the initial resume set, and the resumes that meet the initial matching requirements constitute the first-level resume set. The talent profile is matched with the resume information of each resume in the primary resume set to obtain the matching details of each resume; The job information for the target position, the resume information for each resume, and the matching details for each resume are comprehensively analyzed to obtain the overall matching result information between each resume and the target position; and Obtain feedback from recruiting users regarding submitted resumes, and dynamically optimize the talent profile based on this feedback.

2. The method according to claim 1, wherein, The steps for building a talent profile based on job information for a target position include: Tag extraction is performed on the job information of the target job to obtain multiple job tags; Construct a first prompt text, which includes at least a description of the task to extract the core job requirements, one or more core job requirements to be extracted, and the corresponding extraction principles. Input the first prompt text and the job information of the target job into the first language model; The large language model extracts the core requirements of the job information based on the first prompt text and outputs the extraction results. Receive the extraction results from the first major language model and use them as the core requirement information for the target position; and A talent profile is constructed by combining the tag content of multiple job tags and one or more core requirements of the target job. The job tags and core requirements are talent profile elements, and the tag content of the job tags and the core requirements are feature values ​​of the corresponding talent profile elements.

3. The method according to claim 2, wherein, The steps for optimizing the talent profile based on feedback information include: A resume sample set was constructed based on feedback information; The rejection rate for each talent profile element is calculated based on a sample set of resumes. The rejection rate of the corresponding feature values ​​of the talent profile elements is evaluated based on the optimization conditions of the aforementioned talent profile elements; and The rejection rate in response to the feature values ​​of talent profile elements satisfies the optimization conditions for the feature values ​​of talent profile elements of the target position.

4. The method according to claim 1, wherein, The steps of matching the talent profile with the resume information of each resume in the primary resume set to obtain matching details for each resume include: Construct a second prompt text, which includes at least a description of the matching task, matching principles, and output format; The second prompt text, talent profile, and resume information for each resume are input into the second large language model; and The second language model matches the talent profile and the resume information of each resume according to the second prompt text, and outputs the matching details according to the output format.

5. The method according to claim 4, wherein, The method further includes: optimizing the second prompt text based on feedback information, including: A second sample set is constructed based on the feedback information; the second sample includes resume information and matching result information; and Add the second sample to the second prompt text, and add a description of learning according to the sample to the matching task description.

6. The method according to claim 1, wherein, The steps for comprehensively analyzing the job information for the target position, the resume information for each resume, and the matching details for each resume include: Construct a third prompt text, which includes at least a description of the matching analysis task and an output format; The third prompt text, the job information of the target position, the resume information of each resume, and the matching details of each resume are input into the third language model; and The third language model comprehensively analyzes the job information of the target position, the resume information of each resume, and the matching details of each resume according to the third prompt text, and outputs the comprehensive matching result information between the resume and the target position according to the output format.

7. The method according to claim 6, wherein, The method further includes: optimizing the third prompt text based on feedback information, including: A third sample set is constructed based on feedback information; the third sample includes resume information and comprehensive matching result information; and Add the third sample to the third prompt text, and add a description of learning according to the sample to the matching analysis task description.

8. The method according to claim 1, 6 or 7, wherein, The comprehensive matching result information includes the matching level and its basis details. The basis details of the matching level include matching items and / or non-matching items. The matching items and non-matching items respectively include the matching dimension, the matching details under the matching dimension, and the corresponding job partial information.

9. The method according to claim 8, wherein, The method further includes: classifying the resumes in the primary resume set based on the matching level to obtain resume sets with different matching levels.

10. A resume screening system for selecting resumes that match a target position from an initial resume set, wherein, The system includes: The talent profiling module is configured to build a corresponding talent profile based on the job information of the target position, and optimize the talent profile based on the feedback information from the initial resume screening. The tag acquisition module is configured to obtain the job tags for the target job and the resume tags for each resume. The initial screening module is configured to perform a preliminary matching of resumes and positions based on resume tags and job tags. It filters out resumes that do not meet the preliminary matching requirements from the initial resume set, and the resumes that meet the preliminary matching requirements form a first-level resume set. The core matching module is configured to match the talent profile with the resume information of each resume in the primary resume set to obtain matching details for each resume; and The comprehensive matching module is configured to perform comprehensive analysis on the job information of the target position, the resume information of each resume, and the matching details of each resume to obtain the comprehensive matching result information between each resume and the target position.

11. An electronic device comprising a processor and a memory, wherein the memory stores a set of computer program instructions, characterized in that, The resume screening method according to any one of claims 1-9 is implemented when the processor executes the set of computer program instructions in the memory.

12. A computer-readable storage medium storing a computer program instruction set thereon, characterized in that, When the computer program instruction set is executed by the processor, it implements the resume screening method according to any one of claims 1-9.

13. A computer program product comprising a computer program instruction set, characterized in that, When the computer program instruction set is executed by the processor, it implements the resume screening method according to any one of claims 1-9.

Images