Candidate matching method and device and storage medium

Abstract

The invention provides a candidate matching method and device and a storage medium. The candidate matching method comprises the following steps: obtaining personal information of a plurality of first candidates and post requirements of at least one recruitment post; the personal information comprises numeric information of at least one dimension and text information of at least one dimension. The post requirements correspondingly comprise numeric requirements of at least one dimension and text requirements of at least one dimension; through a pre-trained matching model, performing condition matching on each numeric type requirement of the recruitment post and each numeric type information of each first candidate, and performing semantic matching on each text type requirement of the recruitment post and each text type information of each first candidate, and screening and determining at least one second candidate matched with the recruitment post.

Description

Technical Field

[0001] This invention relates to auxiliary technologies for talent recruitment, and more particularly to a candidate matching method, a candidate matching device, and a computer-readable storage medium. Background Technology

[0002] In the field of intelligent recruitment, accurate matching between candidates and job postings is key to improving recruitment efficiency. However, as companies raise their requirements for talent quality, a single matching model is no longer sufficient to meet actual needs. Existing technologies primarily employ two methods for matching candidates to positions. One is numerical condition matching based on rule models, which uses pre-set structured indicators such as years of work experience, education, and skill certificates for screening. While this method is computationally efficient and has low computational cost, it only covers hard criteria and cannot capture semantic connections and implicit abilities within the text. It is prone to overlooking high-quality candidates due to differences in keyword expression, limiting the matching dimensions. The other method is semantic matching based on language models, which uses natural language processing technology to understand the deeper meaning of resumes and job descriptions, improving the accuracy of matching soft skills. However, this type of model has high computational complexity, especially when processing massive amounts of resumes, leading to a surge in computational demands and a decrease in matching efficiency, making it difficult to balance accuracy and cost. Furthermore, existing language models are trained for general semantic understanding and can only directly calculate the semantic similarity between entire text segments, not for multi-dimensional job matching optimization, thus failing to achieve alignment optimization between job dimensions and candidate dimensions.

[0003] In order to overcome the above-mentioned shortcomings of the existing technology, there is an urgent need in the field for an improved candidate matching method to determine whether a candidate matches the job posting from more dimensions and reduce the computational power requirements of the language model, thereby simultaneously improving matching efficiency and matching accuracy. Summary of the Invention

[0004] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed descriptions that follow.

[0005] To overcome the aforementioned deficiencies in the existing technology, the present invention provides a candidate matching method, a candidate matching device, and a computer-readable storage medium. It can determine whether a candidate matches the job posting from more dimensions through rule-based numerical conditional matching and language-based semantic matching, while reducing the computational power requirement of the language model, thereby simultaneously improving matching efficiency and matching accuracy.

[0006] Specifically, the candidate matching method according to the first aspect of the present invention includes the following steps: obtaining personal information of a plurality of first candidates and job requirements for at least one job posting. The personal information includes at least one dimension of numerical information and at least one dimension of textual information. The job requirements correspondingly include at least one dimension of numerical requirements and at least one dimension of textual requirements; using a pre-trained matching model, performing conditional matching between each of the numerical requirements of the job posting and each of the numerical information of the first candidates, and performing semantic matching between each of the textual requirements of the job posting and each of the textual information of the first candidates, to filter and determine at least one second candidate matching the job posting.

[0007] Furthermore, in some embodiments of the present invention, the step of obtaining the personal information of multiple first candidates includes: determining multiple preset domain name lists and at least one identity information of the first candidate; retrieving web pages that match each domain name in the domain name list based on the identity information to obtain relevant information of the first candidate; and extracting at least one dimension of personal information of the first candidate from the relevant information based on preset personal information dimensions.

[0008] Furthermore, in some embodiments of the present invention, the step of retrieving web pages matching each domain name in the domain name list based on the identity information to obtain relevant information of the first candidate includes: retrieving the primary interface of multiple web pages matching each domain name in the domain name list to obtain thumbnail information in the primary interface; determining whether the thumbnail information matches the identity information of each of the first candidates; and in response to the thumbnail information in any web page matching the identity information of any of the first candidates, opening a link in the primary interface to retrieve its sub-interface and obtaining detailed information in the sub-interface.

[0009] Furthermore, in some embodiments of the present invention, the job requirements include hard first numerical requirements. The step of conditionally matching each of the numerical requirements of the job posting with each of the numerical information of the first candidate includes: conditionally matching the first numerical information of each first candidate in a corresponding dimension according to each of the first numerical requirements; and in response to any first candidate's first numerical information in any dimension not matching the first numerical requirement in the corresponding dimension, filtering out the first candidate and skipping the semantic matching step.

[0010] Furthermore, in some embodiments of the present invention, the job requirements also include hard first textual requirements. The step of semantically matching each of the textual requirements of the job posting with each of the textual information of the first candidates to filter and determine at least one second candidate matching the job posting includes: determining a first candidate whose first numerical information satisfies the first numerical requirements of the corresponding dimension as a third candidate; calculating the second semantic similarity between each of the first textual requirements and the first textual information of each of the third candidates in the corresponding dimension; and filtering and determining at least one second candidate matching the job posting based on the second semantic similarity of each of the third candidates in each of the corresponding dimensions and the corresponding second similarity threshold.

[0011] Furthermore, in some embodiments of the present invention, the job requirements further include flexible second textual requirements. The step of semantically matching each of the textual requirements of the job posting with each of the textual information of the first candidates to filter and determine at least one second candidate matching the job posting includes: determining a first candidate whose first numerical information satisfies the first numerical requirements of the corresponding dimension as a third candidate; calculating the third semantic similarity between each of the second textual requirements and the second textual information of each of the third candidates in the corresponding dimension; and filtering and determining at least one second candidate matching the job posting based on the third semantic similarity of each of the third candidates in multiple corresponding dimensions.

[0012] Furthermore, in some embodiments of the present invention, the job requirements further include flexible second numerical requirements. The step of conditionally matching each of the numerical requirements of the job posting with each of the numerical information of the first candidate further includes: calculating the difference between each of the second numerical requirements and the second numerical information of each first candidate in a corresponding dimension. The step of selecting and determining at least one second candidate matching the job posting based on the third semantic similarity of each third candidate in multiple corresponding dimensions includes: selecting and determining at least one second candidate matching the job posting based on the third semantic similarity of each third candidate in multiple corresponding dimensions and the differences between them in each corresponding dimension.

[0013] Furthermore, in some embodiments of the present invention, the job requirements include corresponding second numerical requirements and second textual requirements. The step of selecting and determining at least one second candidate matching the job posting based on the third semantic similarity of each of the third candidates across multiple corresponding dimensions, and the differences between them across these dimensions, includes: using the differences between the multiple corresponding dimensions as inference priors to determine the semantic path strength of the second differences in their corresponding dimensions; and selecting and determining at least one second candidate matching the job posting based on the third semantic similarity of each of the third candidates across multiple corresponding dimensions, and the semantic path strength between them.

[0014] Furthermore, in some embodiments of the present invention, the step of semantically matching each of the textual requirements of the job posting with each of the textual information of the first candidate to screen and determine at least one second candidate matching the job posting further includes: determining whether the personal information of each third candidate fully covers all dimensions involved in the job requirements; in response to any third candidate's personal information not fully covering all dimensions involved in the job requirements, performing semantic reasoning on the missing dimensions based on the third candidate's personal information to generate indirect information of the third candidate in the missing dimensions; calculating the fourth semantic similarity between the indirect information and the job requirements of the corresponding dimensions; and screening and determining at least one second candidate matching the job posting based on the second difference between the third semantic similarity of each third candidate in multiple corresponding dimensions and the corresponding third similarity threshold, and the third difference between the fourth semantic similarity of each third candidate in each corresponding dimension and the corresponding fourth similarity threshold.

[0015] Furthermore, in some embodiments of the present invention, the matching model includes an alignment module and a matching module. The alignment module is used to extract candidate embedding vectors from the personal information of each first candidate according to multiple preset dimensions, and to extract aligned job embedding vectors from the job requirements of the advertised position according to the multiple dimensions. The matching module includes a conditional matching unit, a language analysis unit, and a fusion unit. The conditional matching unit is used to perform conditional matching on the numerical dimensions of the candidate embedding vector and the job embedding vector to output corresponding conditional matching results. The language analysis unit is used to perform semantic matching on the textual dimensions of the candidate embedding vector and the job embedding vector to output corresponding semantic matching results. The fusion unit, located after the conditional matching unit and the language analysis unit, is used to filter and determine at least one second candidate matching the advertised position based on each conditional matching result and each semantic matching result.

[0016] Furthermore, in some embodiments of the present invention, the input end of the language analysis unit is also connected to the output end of the condition matching unit, for using the condition matching result of at least one numerical dimension as the reasoning prior for semantic matching of the corresponding textual dimension, weakening the semantic path that does not satisfy the numerical constraints, so as to output the semantic matching result.

[0017] Furthermore, in some embodiments of the present invention, the matching method further includes the following steps: in response to screening and determining at least one second candidate matching the job posting, generating a description of the matching reason based on the personal information of the second candidate matching the job requirements.

[0018] Furthermore, the candidate matching apparatus provided according to the second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and configured to execute the computer instructions stored thereon to implement the candidate matching method as provided in the first aspect of the present invention.

[0019] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions thereon. When the computer instructions are executed by a controller, the candidate matching method as provided in the first aspect of the present invention is implemented.

[0020] In summary, the candidate matching method, candidate matching device, and computer-readable storage medium provided by this invention can all determine whether a candidate matches the job posting from more dimensions through rule-based numerical conditional matching and language-based semantic matching, while reducing the computational power requirements of the language model, thereby simultaneously improving matching efficiency and accuracy. Attached Figure Description

[0021] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related characteristics or features may have the same or similar reference numerals.

[0022] Figure 1 A schematic diagram of the structure of a matching model provided according to some embodiments of the present invention is shown.

[0023] Figure 2 A flowchart illustrating the online matching phase of a candidate matching method provided according to some embodiments of the present invention is shown.

[0024] Figure label: 11. Alignment module; 12. Matching module; 121 Conditional matching unit; 122 Language Analysis Unit; 123 Fusion Unit. Detailed Implementation

[0025] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Although the description of the present invention is presented in conjunction with preferred embodiments, this does not mean that the features of the invention are limited to these embodiments. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of the present invention. To provide a thorough understanding of the invention, many specific details will be included in the following description. The invention may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of the invention, some specific details will be omitted in the description.

[0026] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0027] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood as the orientations shown in the relevant paragraphs and accompanying drawings. These relative terms are for illustrative purposes only and do not imply that the described apparatus must be manufactured or operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0028] It is understood that although terms such as "first," "second," and "third" may be used herein to describe various components, regions, layers, and / or parts, these components, regions, layers, and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers, and / or parts. Therefore, the first components, regions, layers, and / or parts discussed below may be referred to as second components, regions, layers, and / or parts without departing from some embodiments of the present invention.

[0029] As mentioned above, with companies raising their requirements for talent quality, a single matching model is no longer sufficient to meet actual needs. Existing technologies for person-job matching mainly fall into two categories. One is numerical condition matching based on rule models, which uses pre-set structured indicators such as years of work experience, education, and skill certificates for screening. While this method has low computational consumption and high execution efficiency, it only covers hard conditions and cannot capture semantic connections and implicit abilities within the text. It is prone to overlooking high-quality candidates due to differences in keyword expression, thus limiting the matching dimensions. The other is semantic matching based on language models, which uses natural language processing technology to understand the deeper meaning of resumes and job descriptions, improving the accuracy of soft skills matching. However, this type of model has high computational complexity, especially when processing massive amounts of resumes, leading to a surge in computational demands and a decrease in matching efficiency, making it difficult to balance accuracy and cost. Furthermore, existing language models are trained for general semantic understanding and can only directly calculate the semantic similarity between entire text segments, not for multi-dimensional job matching optimization, thus failing to achieve alignment optimization between job dimensions and candidate dimensions.

[0030] To overcome the aforementioned deficiencies in the existing technology, the present invention provides a candidate matching method, a candidate matching device, and a computer-readable storage medium. It can determine whether a candidate matches the job posting from more dimensions through rule-based numerical conditional matching and language-based semantic matching, while reducing the computational power requirement of the language model, thereby simultaneously improving matching efficiency and matching accuracy.

[0031] In some non-limiting embodiments, the candidate matching apparatus provided in the second aspect of the present invention can be implemented based on the candidate matching method provided in the first aspect of the present invention.

[0032] Specifically, the candidate matching apparatus provided in the second aspect of the present invention includes a memory and a controller. Here, the memory includes, but is not limited to, the computer-readable storage medium provided in the third aspect, on which computer instructions are stored. The controller is connected to the memory and configured to execute the computer instructions stored in the memory to implement the candidate matching method as provided in the first aspect of the present invention.

[0033] The working principle of the matching device for the aforementioned candidates will be described below with reference to embodiments of some candidate matching methods. Those skilled in the art will understand that these embodiments of matching methods are merely non-limiting implementations provided by the present invention, intended to clearly demonstrate the main concepts of the invention and provide specific solutions convenient for public implementation, rather than limiting all functions or operating methods of the candidate matching device. Similarly, the candidate matching device is also only a non-limiting implementation provided by the present invention, and does not constitute a limitation on the executing entity and execution order of each step in the candidate matching methods.

[0034] In some non-limiting embodiments, the candidate matching method provided in the first aspect of the present invention can be implemented in two stages: offline training and online matching. A technician can first train a matching model in the offline training stage, and then, in the online matching stage, use the matching model to match the personal information of multiple first candidates with the job requirements of at least one recruitment position.

[0035] Please refer to Figure 1 . Figure 1 A schematic diagram of the structure of a matching model provided according to some embodiments of the present invention is shown.

[0036] exist Figure 1 In the embodiment shown, the matching model includes an alignment module 11 and a matching module 12.

[0037] Here, the alignment module 11 is used to extract candidate embedding vectors from the personal information of each first candidate according to multiple preset dimensions, and to extract aligned job embedding vectors from the job requirements of the job posting according to multiple dimensions.

[0038] Furthermore, in Figure 1 In the embodiment shown, the matching module 12 includes a condition matching unit 121 (e.g., a rule model), a language analysis unit 122 (e.g., an LLM (Large Language Model)), and a fusion unit 123.

[0039] Here, the condition matching unit 121 is used to perform condition matching on the numerical dimensions of the candidate embedding vector and the job embedding vector to output the corresponding condition matching results.

[0040] Specifically, during the training of the aforementioned condition matching unit 121, the matching device can first acquire historical recruitment data as training samples, wherein the historical recruitment data includes candidate samples and job posting samples and their matching result labels.

[0041] Next, the matching device can preprocess the historical recruitment data. Specifically, the matching device can extract numerical dimensions (e.g., years of work experience, quantitative values ​​of education level, number of professional skills certificates, project experience duration, salary expectation range, etc.) from the candidate sample, and extract corresponding numerical dimensions (e.g., minimum years of work experience required for the position, target education level, number of required certificates, etc.) from the job posting sample, and then standardize them respectively.

[0042] Then, the matching device can input the preprocessed candidate samples and job posting samples into the conditional matching unit 121 to be trained to obtain the output matching results. Based on the matching results and the matching result labels in the historical recruitment data, the learning parameters (e.g., matching threshold, weight parameters) of the conditional matching unit 121 are adjusted until the evaluation indicators (e.g., accuracy, recall) of the conditional matching unit 121 reach the preset evaluation indicator threshold, thus completing the training of the conditional matching unit 121.

[0043] The language analysis unit 122 is used to perform semantic matching on the textual dimension of the candidate embedding vector and the job embedding vector to output the corresponding semantic matching results.

[0044] Specifically, during the training of the language analysis unit 122, the matching device can first acquire historical recruitment data as training samples. The historical recruitment data includes candidate samples and job posting samples and their matching result labels. Both types of samples contain textual dimension data (e.g., descriptions of candidates' job content, detailed descriptions of project experience, and summaries of skills and abilities; job descriptions, job qualification descriptions, and core competency requirements).

[0045] Afterwards, the matching device can preprocess the historical recruitment data. Specifically, the matching device can extract the aforementioned textual dimension data from the candidate sample and the job posting sample, perform text cleaning such as removing redundant symbols and irrelevant expressions, standardize industry terms, and then transform the textual dimension data into a unified-dimensional semantic embedding vector to ensure effective representation of text semantics.

[0046] Next, the matching device can input the preprocessed candidate semantic embedding vector and job semantic embedding vector into the language analysis unit 122 to be trained, obtain the output semantic matching result by calculating vector similarity (e.g., cosine similarity), and adjust the learning parameters of the language analysis unit 122 (e.g., LLM fine-tuning coefficient, semantic similarity threshold, text feature attention weight) according to the matching result label in the historical recruitment data, until the evaluation index (e.g., accuracy) of the language analysis unit 122 reaches the preset evaluation index threshold, thus completing the training of the language analysis unit 122.

[0047] The fusion unit 123 is located after the condition matching unit 121 and the language analysis unit 122, and is used to screen and determine at least one second candidate that matches the recruitment position based on the condition matching results and the semantic matching results.

[0048] Specifically, during the training of the fusion unit 123, the matching device can first acquire historical recruitment data as training samples. The historical recruitment data includes candidate samples, job posting samples, final matching result labels, historical conditional matching results output by the conditional matching unit 121, and historical semantic matching results output by the language analysis unit 122.

[0049] Afterwards, the matching device can preprocess the training samples. Specifically, the matching device can standardize and normalize the historical conditional matching results and historical semantic matching results to eliminate the dimensional differences in the output results of different units and ensure the effectiveness of the fusion calculation.

[0050] Then, the matching device can input the preprocessed historical condition matching results, historical semantic matching results, and the corresponding final matching result labels into the fusion unit 123 to be trained, so as to output a comprehensive matching result. Based on the deviation between the comprehensive matching result and the final matching result label, the learning parameters of the fusion unit 123 (e.g., the fusion weight ratio of the two types of matching results) are adjusted until the evaluation index of the fusion unit 123 (e.g., accuracy, recall) reaches the preset evaluation index threshold, thus completing the training of the fusion unit 123.

[0051] Then, the matching device can construct a trained matching model based on the trained units mentioned above.

[0052] Furthermore, since existing language models are trained for general semantic understanding, they can only directly calculate the semantic similarity between entire text segments, not for multi-dimensional job matching optimization, and cannot achieve alignment optimization between the job dimension and candidate dimensions. Therefore, the matching method and apparatus provided by this invention introduces a job dimension embedding layer and a candidate dimension embedding layer before the output layer of the matching model, calculates the conditional similarity for each dimension separately, and then aggregates the similarity matching models to predict the final matching result, thus realizing a task-customized alignment mechanism.

[0053] By performing the above steps, those skilled in the art can obtain the matching module 12. They can then store it on an optical disc, hard disk, cloud drive, or various other storage media for access and retrieval by the candidate's matching device during the online matching phase.

[0054] Please refer to Figure 2 . Figure 2 A flowchart illustrating the online matching phase of a candidate matching method provided according to some embodiments of the present invention is shown.

[0055] like Figure 2As shown, the candidate matching device provided in the second aspect of the present invention can first obtain the personal information of multiple first candidates and the job requirements of at least one recruitment position.

[0056] Here, the aforementioned personal information includes at least one dimension of numerical information and at least one dimension of textual information. Specifically, numerical information includes, but is not limited to, information that can be represented numerically, such as age, gender, education level, years of work experience, and skill certificates. Textual information includes, but is not limited to, information that needs to be represented in text, such as professional name, project experience, and personal strengths.

[0057] Correspondingly, the above job requirements include at least one dimension of numerical requirements and at least one dimension of textual requirements. Specifically, numerical information includes requirements that can be expressed numerically, such as age, gender, education, years of work experience, and skill certificates. Textual information includes requirements that need to be expressed in text, such as professional name, project experience, and personal strengths.

[0058] Furthermore, in the process of obtaining the personal information of multiple first-choice candidates, the matching device can first determine a list of multiple preset domain names, as well as the identity information of at least one first-choice candidate (e.g., name, nickname, etc.). Here, the aforementioned domain name list includes, but is not limited to, website domain names containing publicly available personal information, such as various recruitment platforms and microblogs.

[0059] Afterwards, the matching device can use web crawler software to search for web pages matching each domain in the domain list periodically and / or before resume matching, based on the identity information and after obtaining user authorization, in order to obtain relevant information about the first candidate (e.g., the first candidate's public information on the corresponding web page), so as to improve the richness of resume information and thus improve matching accuracy.

[0060] Specifically, during the retrieval process, the matching device can retrieve the first-level interfaces of multiple web pages that match each domain name in the domain name list in order to obtain (e.g., text recognition, image recognition) the abbreviated information (e.g., title, tags, summary) in the first-level interface.

[0061] Then, the matching device can determine whether the abbreviation information matches the identity information of each first candidate and generate a judgment result of "whether to enter the details page".

[0062] Subsequently, in response to the thumbnail information in any webpage, the identity information of any first candidate is matched, thus generating a judgment result of "need to enter the detail page". The matching device can further parse the structure of the webpage to determine multiple sub-interfaces of the first-level interface in the webpage.

[0063] Next, the matching device can open links in the primary interface to retrieve its sub-interfaces and obtain detailed information from them. Here, the sub-interface can be a second-level interface directly linked by a link, or a further-level interface indirectly linked by a link in the second-level interface or its tertiary interface.

[0064] Specifically, the matching device can determine whether each field in each level of the interface has an explicit field representation, whether it is in image form, or whether it is a mixed display, in order to determine the extraction method for each field.

[0065] Then, the matching device can extract at least one dimension of personal information of the first candidate from the relevant information through the alignment module 11, based on the preset personal information dimensions.

[0066] Those skilled in the art will understand that the above-described embodiments of first obtaining relevant information of the first candidate from the resume and then extracting personal information of the corresponding dimension are merely some non-limiting implementation methods provided by the present invention, intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are easy for the public to implement, rather than intended to limit the scope of protection of the present invention.

[0067] Alternatively, in other embodiments, those skilled in the art can also directly obtain the personal information of each first candidate from the corresponding dimension of their resume via the alignment module 11.

[0068] Thus, the matching method and apparatus provided by the present invention can obtain candidates' personal information across platforms to expand the amount of data and further improve matching accuracy.

[0069] After that, such as Figure 1 As shown, the candidate matching device provided in the second aspect of the present invention can perform conditional matching (e.g., Boolean judgment) on each numerical requirement of the job posting and each numerical information of the first candidate via the conditional matching unit 121 of the matching module 12 in the pre-trained matching model, and perform semantic matching (e.g., LLM analysis) on each textual requirement of the job posting and each textual information of the first candidate via the language analysis unit 122, so as to screen and determine at least one second candidate matching the job posting.

[0070] Specifically, in some embodiments, the job requirements of the above-mentioned recruitment positions include hard first numerical requirements (e.g., zero work experience in the real estate industry). In the process of matching these requirements with the numerical information of each first candidate, the condition matching unit 121 can perform condition matching on the first numerical information of each first candidate in the corresponding dimension according to the first numerical requirements.

[0071] Subsequently, in response to the fact that the first numerical information of any dimension of any first candidate does not match the first numerical requirement of the corresponding dimension, the condition matching unit 121 can directly determine that the first candidate does not match the job posting, filter out the first candidate, and skip the semantic matching step.

[0072] For example, if the first numerical requirement for a job posting includes zero work experience in the real estate industry, then if any first-choice candidate's work experience in the real estate industry is not zero, that candidate can be directly eliminated.

[0073] Correspondingly, during the offline training phase of the condition matching unit 121, the matching device can train and adjust the learning parameters of the condition matching unit 121 to ensure that the condition matching unit 121 adapts to the enterprise's recruitment decision preferences. Here, the learning parameters may include matching thresholds and / or weight parameters. Matching thresholds include thresholds for different job types at various sub-dimensions (e.g., experience duration thresholds). Weight parameters include the priority weights for each dimension.

[0074] Thus, the matching method and apparatus provided by the present invention can skip the semantic matching step by performing only conditional matching with hard numerical requirements, and directly filter the first candidate, thereby greatly saving the computing power required for candidate matching and improving the efficiency of candidate matching.

[0075] Furthermore, in some embodiments, the job requirements for the above-mentioned recruitment positions include hard first textual requirements (e.g., major in computer science). In the process of semantically matching the first textual information with each first candidate, the matching device can first determine the first candidate who meets the first numerical requirements of the corresponding dimension through the condition matching unit 121 as the third candidate.

[0076] Subsequently, the matching device can calculate the second semantic similarity between the first textual requirements and the corresponding dimension of the first textual information of each third candidate via the language analysis unit 122.

[0077] Then, the language analysis unit 122 can filter and determine at least one second candidate that matches the job posting based on the second semantic similarity of each third candidate in each corresponding dimension and the corresponding second similarity threshold.

[0078] Furthermore, in some embodiments, in response to the second semantic similarity between any first textual information of any third candidate and the second semantic similarity of the first textual requirement of the corresponding dimension being less than the corresponding second similarity threshold, the language analysis unit 122 can determine that the third candidate does not match the job posting, thereby filtering out the third candidate.

[0079] Conversely, in response to the fact that the second semantic similarity of any third candidate in all dimensions is greater than or equal to the corresponding second similarity threshold, the language analysis unit 122 can determine that the third candidate matches the job posting, and thus identify the third candidate as the second candidate matching the job posting.

[0080] Correspondingly, during the offline training phase of the language analysis unit 122, the matching device can train and adjust the learning parameters of the language analysis unit 122 to ensure that the language analysis unit 122 adapts to the enterprise's recruitment decision preferences. Here, the learning parameters may include the fine-tuning coefficients of the LLM model, the semantic similarity threshold, and the attention weights for text features. The fine-tuning coefficients are used to fine-tune the general large language model to fit the industry terminology and job-specific expressions required for the recruitment scenario, optimizing the representation accuracy of the semantic embedding vectors. The semantic similarity threshold is used to define the semantic matching boundary. The attention weights are used to determine the semantic calculation weights of each text feature, weakening the interference of irrelevant text information.

[0081] Thus, the matching method and apparatus provided by the present invention can significantly reduce the number of times the language model is activated by first performing conditional matching with hard numerical requirements and then performing semantic matching, thereby greatly saving the computing power required for candidate matching and improving the efficiency of candidate matching.

[0082] Those skilled in the art will understand that the above-described embodiments, in which the first candidate is screened by the condition matching unit 121 based on numerical requirements, and the third candidate is further screened by the language analysis unit 122 based on textual requirements, are merely some non-limiting implementations provided by the present invention. They are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are easy for the public to implement, rather than to limit the scope of protection of the present invention.

[0083] Optionally, in other embodiments, the language analysis unit 122 may also operate synchronously with the condition matching unit 121 to calculate the second semantic similarity between each first textual requirement and the first textual information of each first candidate in the corresponding dimension, and directly filter and determine at least one second candidate matching the recruitment position based on the second semantic similarity of each third candidate in each corresponding dimension and the corresponding second similarity threshold.

[0084] In addition, in some embodiments, the above job requirements also include flexible second textual requirements. In the process of semantically matching the textual information of each first candidate, the matching device can first determine the first candidate whose first numerical information meets the first numerical requirements of the corresponding dimension as the third candidate through the condition matching unit 121.

[0085] Subsequently, the matching device can calculate the third semantic similarity between the second textual information of each second textual requirement and the corresponding dimension of the third candidate via the language analysis unit 122.

[0086] Then, the language analysis unit 122 can filter and determine at least one second candidate that matches the job posting based on the third semantic similarity of each third candidate in multiple corresponding dimensions.

[0087] Furthermore, in some embodiments, the present invention can determine the weights of the third semantic similarity of each corresponding dimension through training, and calculate the total similarity score of each third candidate by weighting, thereby determining the third candidate whose total similarity score is greater than or equal to a preset first total score threshold as the second candidate matching the recruitment position.

[0088] Correspondingly, during the offline training phase of the fusion unit 123, the matching device can train and adjust the learning parameters of the fusion unit 123 to ensure that the fusion unit 123 adapts to the enterprise's recruitment decision preferences. Here, the learning parameters may include the fusion weight parameters of the two types of matching results and / or the comprehensive matching score threshold.

[0089] Those skilled in the art will understand that the above-described embodiments, in which the first candidate is screened by the condition matching unit 121 based on numerical requirements, and the third candidate is further screened by the language analysis unit 122 based on textual requirements, are merely some non-limiting implementations provided by the present invention. They are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are easy for the public to implement, rather than to limit the scope of protection of the present invention.

[0090] Optionally, in other embodiments, the language analysis unit 122 may also operate synchronously with the condition matching unit 121 to calculate the third semantic similarity between each second textual requirement and the second textual information of each first candidate in the corresponding dimension, and to screen and determine at least one second candidate that matches the recruitment position based on the third semantic similarity of each third candidate in multiple corresponding dimensions.

[0091] In addition, Figure 1 In the embodiment shown, the input of the language analysis unit 122 is also connected to the output of the condition matching unit 121, which is used to take the condition matching result of at least one numerical dimension as the reasoning prior of the semantic matching of the corresponding text dimension, weaken the semantic path that does not meet the numerical constraints, and output the semantic matching result.

[0092] Specifically, the job requirements for the above-mentioned recruitment positions also include flexible second numerical requirements (e.g., IELTS score greater than 7). In the process of matching these requirements with the numerical information of each first candidate, the matching device can also calculate the difference between each second numerical requirement and the second numerical information of each first candidate in the corresponding dimension via the condition matching unit 121.

[0093] Subsequently, during the screening of candidates based on flexible textual requirements, the matching model can also use fusion unit 123 to screen and determine at least one second candidate that matches the job posting based on the third semantic similarity of each third candidate in multiple corresponding dimensions and the difference between them in each corresponding dimension.

[0094] Correspondingly, during the offline training phase of the fusion unit 123, the matching device can train and adjust the learning parameters of the fusion unit 123 to ensure that the fusion unit 123 adapts to the enterprise's recruitment decision preferences. Here, the learning parameters may include the fusion weight parameters of the two types of matching results and / or the comprehensive matching score threshold.

[0095] Furthermore, in some embodiments, the job requirements for the aforementioned recruitment positions include corresponding second numerical requirements and second textual requirements. During the process of screening third candidates by combining the third semantic similarity determined based on the textual requirements and the difference determined based on the numerical requirements, the fusion unit 123 can use the differences in multiple corresponding dimensions as reasoning priors to determine the semantic path strength of the second difference in the corresponding dimension.

[0096] Here, the semantic path strength is recorded in the form of an attention mask or a constraint signaling token.

[0097] Thus, when calculating semantic similarity, this invention can automatically weaken semantic paths that do not meet numerical constraints and strengthen highly feasible job-candidate combinations that meet numerical constraints, thereby using the rule matching results as prior conditions for semantic reasoning within the model to further improve matching accuracy.

[0098] Subsequently, the fusion unit 123 can select and determine at least one second candidate that matches the recruitment position based on the third semantic similarity of each third candidate in multiple corresponding dimensions and the semantic path strength of its corresponding dimensions.

[0099] Furthermore, in some embodiments, the matching method and apparatus provided by the present invention can determine the weights of the first difference and the second difference between the third semantic similarity and the corresponding third similarity threshold by training, and calculate the total difference score of each third candidate by weighting, thereby determining the third candidate whose total difference score is less than or equal to the preset second total score threshold as the second candidate for the job position.

[0100] Thus, the matching method and apparatus provided by this invention upgrade the "rule model and LLM model" to an "interpretable joint structure." The language model reasoning structure, guided by rule constraints, uses "numerical constraints" as the reasoning prior of the language model, encoding the numerical matching results into attention masks or constraint tokens. Furthermore, when calculating semantic similarity, the language model automatically weakens semantic paths that do not meet the numerical constraints and strengthens highly feasible job-candidate combinations. This treats rules as internal reasoning conditions within the model, forming a joint structure that integrates rule constraints and language model reasoning paths. Through a combination of conditional matching and semantic matching, both hardware and software approaches are used to further improve matching accuracy and efficiency.

[0101] In addition, in some embodiments, during the screening of the first candidate based on textual requirements, the matching device can also use the language analysis unit 122 to determine whether the personal information of each third candidate fully covers all dimensions involved in the job requirements.

[0102] Subsequently, in response to the fact that the personal information of any third candidate does not fully cover all dimensions involved in the job requirements, the language analysis unit 122 performs semantic reasoning on the missing dimensions based on the personal information of the third candidate to generate indirect information of the third candidate in the missing dimensions, thereby further improving the matching accuracy.

[0103] Then, the language analysis unit 122 can calculate the fourth semantic similarity between the indirect information and the job requirements of the corresponding dimension.

[0104] Furthermore, in some embodiments, the aforementioned job requirements may be the aforementioned rigid first numerical requirements and / or first textual requirements, or they may be the aforementioned flexible second numerical requirements and / or second textual requirements.

[0105] In this way, by setting aside the hard condition matching and hard semantic matching of missing items, and by using a language model to perform flexible semantic matching of the indirect information of missing items with the corresponding job requirements, the present invention can effectively avoid false screening due to missing personal information, thereby providing interview opportunities for more potential matching candidates.

[0106] Subsequently, the language analysis unit 122 can filter and determine at least one second candidate that matches the job posting based on the second difference between the third semantic similarity of each third candidate in multiple corresponding dimensions and the corresponding third similarity threshold, and the third difference between the fourth semantic similarity of each third candidate in each corresponding dimension and the corresponding fourth similarity threshold.

[0107] Furthermore, in some embodiments, the present invention can determine the weights of the first and second differences for each corresponding dimension through training, and calculate the total difference score of each third candidate by weighting, thereby identifying the third candidates whose total difference score is less than or equal to a preset second total score threshold as the second candidates matching the recruitment position.

[0108] Subsequently, in response to identifying at least one second candidate matching the job posting, the matching device can generate a description of the matching reason based on the second candidate's personal information matching the job requirements. Optionally, this description may include details of the contribution of various metrics to the matching reason.

[0109] Thus, the matching method and apparatus provided by the present invention can reuse the language analysis unit 122 to generate a description of the matching reason, thereby improving the interpretability of the matching result.

[0110] In summary, the candidate matching method, candidate matching device, and computer-readable storage medium provided by this invention can all determine whether a candidate matches the job posting from more dimensions through rule-based numerical conditional matching and language-based semantic matching, while reducing the computational power requirements of the language model, thereby simultaneously improving matching efficiency and accuracy.

[0111] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0112] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different techniques and skills. For example, the data, instructions, commands, information, signals, bits, symbols, and chips described throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, light fields or optical particles, or any combination thereof.

[0113] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0114] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0115] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0116] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for matching candidates, characterized in that, Includes the following steps: Obtain personal information of multiple first-choice candidates and job requirements for at least one job posting, wherein the personal information includes at least one dimension of numerical information and at least one dimension of textual information, and the job requirements correspondingly include at least one dimension of numerical requirements and at least one dimension of textual requirements. Using a pre-trained matching model, conditional matching is performed on each of the numerical requirements of the job posting and on each of the numerical information of the first candidate, and semantic matching is performed on each of the textual requirements of the job posting and on each of the textual information of the first candidate, so as to screen and determine at least one second candidate who matches the job posting.

2. The matching method as described in claim 1, characterized in that, The steps for obtaining the personal information of multiple first candidates include: Determine multiple preset lists of domain names, and at least one identity information of the first candidate; Based on the identity information, web pages matching each domain name in the domain name list are retrieved to obtain relevant information about the first candidate; and Based on preset personal information dimensions, at least one dimension of personal information of the first candidate is extracted from the relevant information.

3. The matching method as described in claim 2, characterized in that, The step of retrieving web pages matching each domain name in the domain name list based on the identity information to obtain relevant information about the first candidate includes: Retrieve the primary interfaces of multiple web pages that match each domain name in the domain name list to obtain the thumbnail information in the primary interfaces; Determine whether the abbreviation information matches the identity information of each of the first candidates; and In response to a thumbnail information in any of the web pages matching the identity information of any of the first candidates, a link in the first-level interface is opened to retrieve its sub-interface and obtain detailed information from the sub-interface.

4. The matching method as described in claim 1, characterized in that, The job requirements include hard, first numerical requirements, and the step of matching each of the numerical requirements of the job posting with the numerical information of each of the first candidates includes: Based on the first numerical type requirements, conditional matching is performed on the first numerical type information of each first candidate in the corresponding dimension; and In response to any first candidate's first numerical information in any dimension not matching the first numerical requirement of the corresponding dimension, the first candidate is filtered out, and the semantic matching step is skipped.

5. The matching method as described in claim 4, characterized in that, The job requirements also include hard first textual requirements. The step of semantically matching each of the textual requirements of the job posting with each of the textual information of the first candidate to screen and determine at least one second candidate matching the job posting includes: The first candidate whose first numerical information satisfies the first numerical requirements of the corresponding dimension is determined as the third candidate. Calculate the second semantic similarity between the first textual requirements of each first textual requirement and the corresponding dimension of the first textual information of each of the third candidates; and Based on the second semantic similarity of each third candidate in each corresponding dimension and the corresponding second similarity threshold, at least one second candidate matching the recruitment position is selected.

6. The matching method as described in claim 4 or 5, characterized in that, The job requirements also include flexible second text-based requirements. The step of semantically matching each of the text-based requirements of the job posting with each of the text-based information of the first candidate to screen and determine at least one second candidate matching the job posting includes: The first candidate whose first numerical information satisfies the first numerical requirements of the corresponding dimension is determined as the third candidate. Calculate the third semantic similarity between each of the second textual requirements and the corresponding dimension of the second textual information of each of the third candidates; and Based on the third semantic similarity of each of the third candidates in multiple corresponding dimensions, at least one second candidate matching the recruitment position is selected.

7. The matching method as described in claim 6, characterized in that, The job requirements also include a flexible second numerical requirement, wherein... The step of matching the numerical requirements of the job posting with the numerical information of the first candidate further includes: calculating the difference between the second numerical requirements and the second numerical information of the first candidate in the corresponding dimension. The step of selecting and determining at least one second candidate matching the recruitment position based on the third semantic similarity of each third candidate in multiple corresponding dimensions includes: selecting and determining at least one second candidate matching the recruitment position based on the third semantic similarity of each third candidate in multiple corresponding dimensions and the difference between them in each corresponding dimension.

8. The matching method as described in claim 7, characterized in that, The job requirements include corresponding second numerical requirements and second textual requirements. The step of selecting and determining at least one second candidate matching the job posting based on the third semantic similarity of each third candidate across multiple corresponding dimensions and the differences between them across each corresponding dimension includes: Using the differences between multiple corresponding dimensions as inference priors, the semantic path strength of the second difference in the corresponding dimensions is determined; and Based on the third semantic similarity of each third candidate in multiple corresponding dimensions, and the semantic path strength of its corresponding dimension, at least one second candidate matching the recruitment position is selected.

9. The matching method as described in claim 6, characterized in that, The step of semantically matching each of the textual requirements for the job posting with each of the textual information of the first candidate to filter and determine at least one second candidate matching the job posting further includes: Determine whether the personal information of each of the third candidates fully covers all dimensions involved in the job requirements; In response to the fact that the personal information of any of the third candidates does not fully cover all dimensions involved in the job requirements, semantic reasoning is performed on the missing dimensions based on the personal information of the third candidates to generate indirect information of the third candidates in the missing dimensions; Calculate the fourth semantic similarity between the indirect information and the corresponding dimension of the job requirements; and Based on the second difference between the third semantic similarity of each third candidate in multiple corresponding dimensions and the corresponding third similarity threshold, and the third difference between the fourth semantic similarity of each third candidate in each corresponding dimension and the corresponding fourth similarity threshold, at least one second candidate matching the recruitment position is selected.

10. The matching method as described in claim 1, characterized in that, The matching model includes an alignment module and a matching module, wherein, The alignment module is used to extract candidate embedding vectors from the personal information of each first candidate according to multiple preset dimensions, and to extract aligned job embedding vectors from the job requirements of the advertised job according to the multiple dimensions. The matching module includes a condition matching unit, a language analysis unit, and a fusion unit. The condition matching unit performs conditional matching on the numerical dimensions of the candidate embedding vector and the job embedding vector to output corresponding conditional matching results. The language analysis unit performs semantic matching on the textual dimensions of the candidate embedding vector and the job embedding vector to output corresponding semantic matching results. The fusion unit, located after the condition matching unit and the language analysis unit, is used to filter and determine at least one second candidate matching the job position based on the conditional matching results and the semantic matching results.

11. The matching method as described in claim 10, characterized in that, The input of the language analysis unit is also connected to the output of the condition matching unit, and is used to take the condition matching result of at least one numerical dimension as the reasoning prior for the semantic matching of the corresponding text dimension, weaken the semantic path that does not meet the numerical constraints, and output the semantic matching result.

12. The matching method as described in claim 1, characterized in that, It also includes the following steps: In response to identifying at least one second candidate matching the job posting, a description of the reason for the matching is generated based on the second candidate's personal information matching the job requirements.

13. A candidate matching device, characterized in that, include: Memory, on which computer instructions are stored; and A processor, connected to the memory, and configured to execute computer instructions stored thereon to implement the candidate matching method as described in any one of claims 1 to 12.

14. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the controller, the candidate matching method as described in any one of claims 1 to 12 is implemented.

Images