An information matching method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application disclose an information matching method and device, electronic equipment and storage medium, comprising: performing controlled first channel semantic processing on the to-be-matched structured information in the recruitment and job-seeking scene to obtain a to-be-matched language description vector, and performing second channel structure feature processing on the to-be-matched structured information to obtain a to-be-matched constraint condition vector; for each candidate information vector group corresponding to the scene of the to-be-matched structured information, calculating a first similarity between the candidate language description vector in the candidate information vector group and the to-be-matched language description vector, and calculating a second similarity between the candidate constraint condition vector in the candidate information vector group and the to-be-matched constraint condition vector; determining the comprehensive similarity of the candidate information vector group and the to-be-matched structured information according to the position feature requirement of the to-be-matched structured information, the first similarity and the second similarity; and taking the ranking result of the comprehensive similarity of all candidate information vector groups and the to-be-matched structured information as an information matching result for display.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of recruitment and job search technology, and in particular to an information matching method, apparatus, electronic device and storage medium. Background Technology

[0002] Currently, in recruitment and job-seeking scenarios, methods such as calculating structured field similarity, text keywords, or semantic similarity are commonly used to match job information with resume information.

[0003] However, existing natural language processing (NLP) methods generate various expressions from text, leading to situations where the original information and the generated text are similar but semantically different. Therefore, the aforementioned matching methods for job information and resume information struggle to effectively address the semantic generalization demands arising from diverse expressions while maintaining accuracy under strict constraints, resulting in low accuracy in recruitment and job-seeking scenarios. Summary of the Invention

[0004] This invention provides an information matching method, apparatus, electronic device, and storage medium, which realizes the dynamic adjustment of the similarity obtained from two directions after fully considering the user's needs for the job, thereby obtaining a more accurate and comprehensive similarity that is closer to the user's recruitment and job search needs, thus improving the user experience from the perspective of meeting user needs.

[0005] In a first aspect, embodiments of the present invention provide an information matching method, the method comprising: The structured information to be matched in the recruitment and job search scenario is subjected to controlled first-channel semantic processing to obtain the language description vector to be matched, and the structured information to be matched is subjected to second-channel structural feature processing to obtain the constraint vector to be matched. For each candidate information vector group corresponding to the scenario of the structured information to be matched, calculate the first similarity between the candidate language description vector and the language description vector to be matched in the candidate information vector group, and calculate the second similarity between the candidate constraint vector and the constraint vector to be matched in the candidate information vector group. Based on the job feature requirements of the structured information to be matched, the first similarity and the second similarity, the comprehensive similarity between the candidate information vector group and the structured information to be matched is determined, and the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched is displayed as the information matching result.

[0006] The information matching method provided in this invention generates semantically consistent and stable natural language description text vectors. This solves the problem that current natural language description methods, when applied to recruitment and job-seeking scenarios, cannot handle the semantic generalization requirements arising from diverse expressions, leading to low accuracy in such scenarios. The method achieves accurate generation of natural language semantics. It dynamically adjusts the similarity obtained from two directions after fully considering the user's job requirements, resulting in a more accurate and relevant comprehensive similarity score. The ranking of this comprehensive similarity score is then presented to the user as the information matching result, improving the user experience by better meeting user needs.

[0007] Secondly, embodiments of the present invention also provide an information matching device, the device comprising: The processing module is used to perform controlled first-channel semantic processing on the structured information to be matched in the recruitment and job search scenario to obtain the language description vector to be matched, and to perform second-channel structural feature processing on the structured information to be matched to obtain the constraint vector to be matched.

[0008] The calculation module is used to calculate the first similarity between the candidate language description vector in the candidate information vector group and the language description vector to be matched for each candidate information vector group corresponding to the scenario of the structured information to be matched, and to calculate the second similarity between the candidate constraint vector in the candidate information vector group and the constraint vector to be matched.

[0009] The matching module is used to determine the comprehensive similarity between candidate information vector groups and the structured information to be matched based on the job feature requirements, first similarity, and second similarity of the structured information to be matched, and to display the ranking results of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching results.

[0010] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising: At least one processor; and A memory that is communicatively connected to at least one processor; wherein, The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the information matching method of any embodiment of the present invention.

[0011] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that are used to cause a processor to execute the information matching method of any embodiment of the present invention.

[0012] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the information matching method of any embodiment of the present invention.

[0013] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the processor of the information matching device, or it may be packaged separately from the processor of the information matching device; this application does not impose any limitations on this.

[0014] The descriptions of the second, third, fourth, and fifth aspects in this application can be referred to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second, third, fourth, and fifth aspects can be referred to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0015] In this application, the names of the aforementioned information matching devices do not limit the devices or functional modules themselves. In actual implementation, these devices or functional modules may appear under other names. As long as the functions of each device or functional module are similar to those in this application, they fall within the scope of the claims of this application and their equivalents.

[0016] These or other aspects of this application will become more readily apparent in the following description. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an information matching method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another information matching method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an information matching device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0020] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first" and "second," etc., used in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0022] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0023] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc. Moreover, embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0024] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0025] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0026] Figure 1This is a flowchart illustrating an information matching method provided in an embodiment of the present invention. This embodiment is applicable to recruitment and job-seeking scenarios, providing users with more suitable recruitment and job-seeking information. The method can be executed by an information matching device, which can be implemented in hardware and / or software and can be configured in an electronic device. In this embodiment, the electronic device can be a device used for recruitment and job-seeking. (Continue to refer to...) Figure 1 This embodiment specifically includes: S101. Perform controlled first-channel semantic processing on the structured information to be matched in the recruitment and job search scenario to obtain the language description vector to be matched, and perform second-channel structural feature processing on the structured information to be matched to obtain the constraint vector to be matched.

[0027] The recruitment and job-seeking scenario refers to the entire business scenario involving the recruiter (company / employer) and the job seeker (individual / candidate), encompassing job matching, information exchange, interactive communication, and ultimately reaching a hiring agreement. In this embodiment, the structured information to be matched can be structured field information of job requirements provided by the recruiter or structured field information of job seeker information. The language description vector to be matched represents the information vector of the structured information to be matched, processed through natural language description, primarily driven by the needs of the user (recruiter or job seeker). The constraint vector to be matched represents the vector representing the correlation between factual constraints and hard condition features in the structured information to be matched; factual constraints represent the constraints imposed on a particular position due to actual job requirements in the recruitment and job-seeking scenario; hard conditions represent the conditions that the recruiter or job seeker must meet due to their own needs in the recruitment and job-seeking scenario. The semantic processing of the controlled first channel is mainly used to generate control constraints for the semantic processing of the structured information to be matched.

[0028] Specifically, after obtaining recruitment information (posted by the recruiter) or job application information (uploaded by the job seeker) in a recruitment and job application scenario, this information can be used as the information to be matched. Then, the information to be matched undergoes structured field processing to obtain the structured information to be matched. After obtaining the structured information to be matched, it can be subjected to controlled first-channel semantic processing and second-channel structural feature processing. The controlled first-channel semantic processing is primarily user-driven (e.g., what qualifications the recruiter values ​​more in the job seeker or what qualifications the job seeker values ​​more in the recruiter). The structured information to be matched is processed according to a controlled expression order, semantic weights (determined based on user needs, the recruitment and job application scenario, and scenario requirements), and missing fields are processed. Each field is then filled into a predefined template with high scenario matching degree to generate a natural language description text vector that is semantically consistent with the original information and semantically stable—that is, the language description vector to be matched. The second-channel structural feature processing mainly involves feature engineering and vectorizing different types of structured fields in the structured information to be matched, generating a constraint vector that highlights the factual constraints or hard conditions of the information to be matched.

[0029] For example, the information to be matched is processed into structured fields to obtain structured information to be matched. The selection of structured fields for the information to be matched includes, but is not limited to: For recruitment information: structured fields include job category, job responsibilities, skill requirements, education requirements, years of work experience, city of work, and industry information, etc.; For resume information: structured fields include educational background, work experience, project experience, skill tags, job intention, and city preference, etc.

[0030] In this embodiment, the information to be matched is structured to obtain structured information to be matched. This structured information is then processed through controlled semantic processing via a first channel and structural feature processing via a second channel to obtain a language description vector reflecting the overall natural language semantics of the information to be matched, and a constraint vector reflecting the job profile constraints. This achieves dual structured processing of the information to be matched, providing a two-dimensional matching consideration for subsequent information matching. Furthermore, driven by user needs, the structured information to be matched is processed according to a controlled expression order, semantic weights (determined according to user needs), and predefined templates to generate a natural language description text vector that is semantically consistent with and stable compared to the original information. This solves the problem that current natural language description methods cannot handle the semantic generalization requirements arising from diverse expressions when applied to recruitment and job-seeking scenarios, resulting in low accuracy in such scenarios. It achieves accurate generation of natural language semantics, providing an accurate foundation for subsequent recruitment and job-seeking information matching.

[0031] S102. For each candidate information vector group corresponding to the scenario of the structured information to be matched, calculate the first similarity between the candidate language description vector and the language description vector to be matched in the candidate information vector group, and calculate the second similarity between the candidate constraint vector and the constraint vector to be matched in the candidate information vector group.

[0032] In this embodiment, the "scenario with structured information to be matched" refers to either the first scenario where a recruiter posts job information and matches job seekers' job information, or the second scenario where a job seeker uploads job information and matches recruiter's job information; hereinafter, both scenarios will be referred to as the first and second scenarios. A candidate information vector group includes a candidate language description vector and a candidate constraint vector. In this embodiment, if the scenario with structured information to be matched is the first scenario (i.e., the information to be matched is job posting), then the candidate information vector group is the vector group corresponding to each job posting obtained after processing the job posting information according to the above steps in the second scenario; that is, the candidate information is job posting information, and the candidate information vector group is each job posting information vector group. If the scenario with structured information to be matched is the second scenario (i.e., the information to be matched is job posting), then the candidate information vector group is the vector group corresponding to each job posting obtained after processing the job posting information according to the above steps in the first scenario; that is, the candidate information is job posting information, and the candidate information vector group is each job posting information vector group.

[0033] Specifically, after obtaining the language description vector to be matched and the constraint vector to be matched of the structured information to be matched, the similarity between the structured information to be matched and the information corresponding to its scene can be calculated from two aspects. That is, for each candidate information vector group, a first similarity is calculated based on the candidate language description vector and the language description vector to be matched of that candidate information vector group; a second similarity is calculated based on the candidate constraint vector and the constraint vector to be matched of that candidate information vector group. Finally, the first and second similarities of each candidate information vector group for the structured information to be matched can be obtained.

[0034] In this embodiment, for each candidate information vector group corresponding to the scenario of the structured information to be matched, the first similarity is calculated using its candidate language description vector and the language description vector to be matched, and the second similarity is calculated using its candidate constraint vector and the constraint vector to be matched. This enables the similarity between the structured information to be matched and the candidate information to be matched to be calculated from both semantic description and constraint perspectives, providing a data foundation for determining candidate information that is more closely matched to the information to be matched.

[0035] S103. Based on the job feature requirements, first similarity and second similarity of the structured information to be matched, determine the comprehensive similarity between the candidate information vector group and the structured information to be matched, and display the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching result.

[0036] Among them, job feature requirements are used to characterize whether the requirements for different types of jobs are more inclined towards hard constraints or natural language semantic conditions. Information matching results are used to characterize the matching similarity results between the structured information to be matched and the candidate information corresponding to each candidate information vector group.

[0037] Specifically, the job feature requirements can be determined first based on the job type in the structured information to be matched. For example, the job feature requirements can be determined based on whether the job type leans towards a specialized or general position. Then, dynamic weights are assigned to the first and second similarities based on the job feature requirements. If the job feature requirement is specialized, the dynamic weight corresponding to the second similarity is higher than that corresponding to the first similarity; if the job feature requirement is general, the dynamic weight corresponding to the first similarity is higher than that corresponding to the second similarity. Afterward, the comprehensive similarity between each candidate information vector group and the structured information to be matched can be calculated based on the similarity and its corresponding dynamic weight. Finally, after determining the comprehensive similarity between each candidate information vector group and the structured information to be matched, the candidate information corresponding to the candidate information vector group can be sorted according to the comprehensive similarity, and the sorting result can be displayed as the information matching result to the user to whom the structured information to be matched belongs.

[0038] In this embodiment, job feature requirements are determined based on job type, which can transform users' hard requirements for jobs into dynamic weights of similarity. Then, based on the dynamic weights determined for the first similarity and the second similarity, the comprehensive similarity between each candidate information vector group and the structured information to be matched is obtained. This achieves a more accurate and comprehensive similarity that is closer to the user's recruitment and job search needs by dynamically adjusting the similarity obtained in the two directions after fully considering the user's job requirements. The ranking result of the comprehensive similarity is then presented to the user as the information matching result, which improves the user experience from the perspective of meeting user needs.

[0039] The information matching method provided in this invention performs structured processing on the information to be matched to obtain structured information to be matched. This structured information is then processed through controlled semantic processing via a first channel and structural feature processing via a second channel to obtain a language description vector reflecting the overall natural language semantics of the information to be matched, and a constraint vector reflecting the job profile constraints. This achieves dual structured processing of the information to be matched, providing a two-dimensional matching consideration for subsequent information matching. Furthermore, driven by user needs, the structured information to be matched is processed according to a certain expression order and semantic weight (determined according to user needs) to generate a natural language description text vector that is semantically consistent with and stable compared to the original information. This solves the problem that current natural language description methods, when applied to recruitment and job-seeking scenarios, cannot handle the semantic generalization requirements brought about by diverse expressions, resulting in low accuracy in recruitment and job-seeking scenarios. This method achieves accurate generation of natural language semantics. For each candidate information vector group corresponding to a scenario of structured information to be matched, a first similarity is calculated using its candidate language description vector and the language description vector to be matched, and a second similarity is calculated using its candidate constraint vector and the constraint vector to be matched. This allows for the calculation of similarity between the structured information to be matched and the candidate information from both semantic and constraint perspectives. By determining job feature requirements based on job type, the user's hard requirements for the job can be transformed into dynamic weights for similarity. Based on the dynamic weights determined for the first and second similarities, the comprehensive similarity between each candidate information vector group and the structured information to be matched is obtained. This achieves a more accurate and job-seeking-oriented comprehensive similarity by dynamically adjusting the similarity obtained from both directions after fully considering the user's job requirements. The ranking result of the comprehensive similarity is then presented to the user as the information matching result, improving the user experience from the perspective of meeting user needs.

[0040] Figure 2 This is a flowchart illustrating another information matching method provided by an embodiment of the present invention. This embodiment is based on the above embodiment and specifies the steps of obtaining the language description vector to be matched, obtaining the constraint vector to be matched, determining the comprehensive similarity, and other optional steps.

[0041] Furthermore, this section details the matching schemes for current recruitment and job-seeking scenarios. First, the current matching schemes are divided into four types: (1) Pure keyword / word frequency-inverse document frequency matching: can only do literal matching, and synonyms cannot be matched if they are expressed differently. For example, "Java development" and "Java programming" will be treated as different things; (2) Pure semantic embedding matching (e.g., directly using Transformer-based bidirectional encoder representation / sentence-level bidirectional encoder representation to do sentence vector similarity): can understand semantics, but the problem is that structured hard conditions (education, city, years of experience, etc.) will be "diluted" in the semantic space, which will cause candidates who do not meet the hard conditions to be recommended because of semantic similarity; (3) Direct matching of structured fields (rule engine / Boolean filtering): too rigid, and cannot be matched if the expression is slightly different. It also cannot handle long-tail positions and non-standard resumes; (4) Current natural language generation methods applied to matching scenarios: generate natural language uncontrollably, and the same input produces different text expressions each time. After being converted into vectors, the position is unstable (i.e., "semantic drift"), which makes the matching results unreproducible and unstable.

[0042] In summary, this embodiment proposes another information matching method, which specifically may include: S201. Obtain recruitment and job search information in the recruitment and job search scenario, and generate structured information to be matched based on the scenario of the recruitment and job search information, the key fields corresponding to the scenario, and the recruitment and job search information.

[0043] In this embodiment, "recruitment and job-seeking information" refers to either recruitment information or job-seeking information; these are just general terms. Furthermore, the scenario for recruitment information is the first scenario, and the scenario for job-seeking information is the second scenario. The key fields in this embodiment are the selectable fields from the aforementioned structured fields.

[0044] Specifically, fields that are consistent with or similar to the key fields can be extracted from the recruitment information posted by the recruiter or the job information uploaded by the job seeker, according to the key field definition as described above. Recruitment information fields or job information fields can be obtained from the recruitment information or job information through rule extraction, entity recognition and other methods. Then, the obtained recruitment information fields or job information fields can be standardized and normalized, and finally the obtained fields can be used as the structured information to be matched.

[0045] For example, the structured information to be matched generated in this embodiment can be as follows: Structured information to be matched (recruitment scenario, posted by the recruiter). |Field|Value| |---|---| Job Type | Backend Development Engineer | |Skill Requirements|Skill 1 Skill 2 Skill 3| |Experience Required|3+ years| |Educational Requirements|Bachelor's Degree or Above| |Work City|City A| Salary Range: 25,000 - 35,000 |Industry|Internet|.

[0046] The above is an example of the final structured information to be matched.

[0047] S202. Determine the semantic weight of the fields based on the scenario and the job characteristics required in the scenario, and determine the order of the fields based on the semantic weight of the fields.

[0048] Here, "scenario" refers to the job search or recruitment scenario to which the structured information to be matched belongs. "Field semantic weight" assigns a semantic weight to each field under different scenario or job feature requirements, representing the degree of importance each field receives under those requirements. "Field order" controls the arrangement of fields in the natural language processing.

[0049] Specifically, even within the same job type, the specific requirements for job features can differ depending on the user context (e.g., a job seeker is looking for a hardware engineer position, while a recruiter is looking for a hardware engineer in a specific field like aviation). In this case, it's necessary to determine the semantic weights of the fields based on the context and the specific job requirements within that context. Then, the order of the fields is determined according to their semantic weights.

[0050] For example, one implementation is that the semantic weights of different fields can be defined by the user according to their own scenario needs, such as recruiters defining or inputting the job characteristics they value more (e.g., paying more attention to job professionalism, job matching with the scenario, skill or experience requirements, etc.); another implementation is that the weights can be determined based on the type and level of professionalism of the job characteristics in the structured information to be matched.

[0051] Optionally, in one implementation, if the recruiter's job requirement is limited in skill matching, that is, they value whether the job seeker has mastered the required skills most, followed by work experience, then they can assign semantic weights to fields for different job requirement characteristics, such as: skill requirements = high weight, experience requirements = medium-high weight, and the rest = standard weight; then, the field order can be determined as: skill requirements > experience requirements > job type > education requirements > work city > salary range > industry. In another implementation, if the scenario is a job search scenario and the structured information to be matched is the structured information generated from the job search information corresponding to the job seeker, then firstly, the professionalism can be determined based on the job feature information, such as whether the applied skills are general or specific skills, whether specific experience is required, whether in-depth professional knowledge is required, or whether qualification certification is required. Furthermore, the professionalism can also be determined based on the specific scenario in the job features, such as whether it is a hardware designer in a specific field, as exemplified above, or based on the specific field in the experience related to the job in the job search information. At the same time, the job seeker's job search needs can also be obtained. Finally, based on the job seeker's job search needs, and the professionalism of the job search feature information is dynamically determined, the semantic weight of the fields is determined. For example, if the job seeker's job search needs to be for a state-owned enterprise, then the final order of the fields determined according to the corresponding semantic weights can be, from first to last, company type, job type, skills possessed, experience possessed, and educational information, etc. If the scenario is a recruitment scenario, and the structured information to be matched is the structured information generated from the recruitment information corresponding to the recruiter, then firstly, its professionalism can be determined based on the required job feature information, in the same way as the above determination method; at the same time, the recruiter's recruitment requirements can be obtained. For example, if the recruitment requirement is a graduate student in a specific field with more than three years of work experience, then the final field arrangement order determined according to the corresponding field semantic weight can be from front to back as follows: specific technical field job type, education information, experience requirements, required skills, etc.

[0052] In a practical example, following the example above, the semantic weights of each field can be obtained as follows: Skill Requirements = 1.0 (High), Experience Requirements = 0.7 (Medium-High), and the rest = 0.3 (Standard). Therefore, based on the semantic weights of each field, the field order can be determined as follows: Skill Requirements → Experience Requirements → Job Type → Education Requirements → Work City → Salary Range → Industry.

[0053] S203. Determine the predefined filling template according to the scenario and the order of the fields, and fill the structured information to be matched into the predefined filling template according to the semantics of each field to obtain the initial description text.

[0054] Among them, the predefined fill template is a descriptive text template set for the scenario requirements of each scenario; optionally, the predefined fill template in this embodiment can be pre-generated or generated during historical application, or it can be generated in real time according to the scenario and the order of field arrangement.

[0055] Specifically, since there may be multiple predefined fill templates with different field arrangement orders in different scenarios, after determining the field arrangement order, the predefined fill template to be used can be determined according to the scenario and the field arrangement order, and the structured information to be matched can be filled into the predefined fill template according to the semantics of each field to obtain the initial description text.

[0056] For example, continuing the above example, the predefined fill template can be formatted as: "{Skill Requirements Description}; {Experience Requirements Description}; This position is {Job Type}, Education Requirements {Education Requirements}, Work Location {Work City}, Salary Range {Salary Range}, Industry {Industry}". Furthermore, the structured information to be matched can be filled into the predefined fill template according to the semantics of each field, ultimately resulting in an initial description text such as: "Requires mastery of skills 1, 2, and 3. Requires at least 3 years of relevant work experience. This position is a backend development engineer, with a bachelor's degree or above, work location in City A, salary range 25K-35K, and industry: Internet."

[0057] S204. For each field in the initial description text, determine the strengthening keywords according to the semantic weight of the corresponding field, and inject the strengthening keywords into the fields of the initial description text to obtain the target description text.

[0058] Among them, the reinforcement keywords are used to strengthen the semantics of their corresponding fields; in this embodiment, the reinforcement keywords can be such as "repeated reinforcement keywords" and "enhanced keywords"; "repeated reinforcement keywords" are keywords that express the keywords of the field multiple times from different angles, and "enhanced keywords" can be keywords injected to emphasize the field.

[0059] Specifically, for each field in the initial description text, we can determine whether each field is a key field based on its corresponding semantic weight. Furthermore, for fields with different semantic weights, we can identify suitable reinforcement keywords and inject these keywords into the fields of the initial description text. Finally, after adding reinforcement keywords to all fields that need them, we obtain the target description text.

[0060] For example, continuing the above case, applying "repeated reinforcement keywords" and "enhanced keywords" to high-weight fields (such as skill requirements, weight = 1.0) yields a result such as: "Core skill requirements are Skill 1, Skill 2, and Skill 3. Must master Skill 1 for backend development and be proficient in Skills 2 and 3." Applying "enhanced keywords" (e.g., at least, or above) to medium-to-high-weight fields (such as experience requirements, weight = 0.7) yields a result such as: "Requires at least 3 years of relevant work experience." The final generated target description text is: "Core skill requirements are Skill 1, Skill 2, and Skill 3. Must master Skill 1 for backend development and be proficient in Skills 2 and 3. Requires at least 3 years of relevant work experience. This position is for a backend development engineer, with a bachelor's degree or above, located in City A, with a salary range of 25K-35K, and belonging to the internet industry."

[0061] It is worth noting that in this embodiment, the "target description text" generated by the semantic matching method in current recruitment and job-seeking scenarios is briefly described to facilitate comparison with the solution provided in this embodiment: The structured information to be matched is directly input into the large language model without any control, allowing the model to freely generate natural language descriptions. The results of the three generation steps may be as follows: (1) First generation: "We are an internet company located in City A and are looking for a backend development engineer. We hope you have experience in using skills 1 and 2 and are familiar with skill 3. The salary and benefits are generous. We welcome candidates with a bachelor's degree or above and more than 3 years of work experience to submit their resumes!"; (2) Second generation: "We are sincerely recruiting backend developers. Location: City A, Internet industry, monthly salary 25K-35K. Job requirements: Bachelor's degree, 3 years of relevant experience, technology mainly based on skill 1."; (3) Third generation: "Internet companies in City A are recruiting backend engineers with skill 1. Requirements: more than 3 years of development experience, bachelor's degree or above, proficient in skills 2 and 3, salary negotiable." The target description text generated in this way has the problem of semantic drift with completely different wording, structure and focus. It is also unable to generate semantic description text that meets the key focus of the job seeker based on the job characteristics of the job seeker. In addition, the large language model will modify some of the original content (such as salary) in the form of "brainstorming or summarizing" when generating freely.

[0062] S205. Input the target description text into the first channel for semantic processing to obtain the language description vector to be matched output by the first channel.

[0063] Specifically, in this embodiment, the target description text is input into the first channel for semantic processing to obtain the language description vector to be matched output by the first channel. One optional approach is to vectorize the target description text into a pre-trained model; another optional approach is to generate text from the target description text using a large language model under controlled prompts and then quantize it. Therefore, in this embodiment, after obtaining the target description text, the method for generating the language description vector to be matched can be the vectorization method described above, and is not limited here.

[0064] For example, in one alternative approach, multiple standard generation examples of similar scenarios are provided to the pre-trained model as references when generating text. The model imitates the structure, word style, and expression patterns of these examples to generate new text, thus maintaining output consistency. For instance, if three standard descriptions of the job title "backend engineer" are provided as examples, the model will maintain stylistic consistency when generating a fourth backend engineer description. This essentially constrains the generation boundaries through "demonstrations." In another alternative approach, when the first channel is implemented based on a large language model, the generation results can be controlled by constructing structured prompts. These prompts include: field priority instructions, output format constraints, style consistency requirements, and restrictions prohibiting improvisation. For example, the prompt could be: "Please describe the job strictly according to the following format, without adding any extra embellishments or speculative descriptions. Fields are arranged from highest to lowest importance, and high-weight fields need to be repeatedly emphasized. Input field: {structured data}. Output format: {template structure}." Furthermore, during the decoding stage of the large model's generated text, hard constraints can be imposed on the generation probability of candidate tokens, such as: (1) Lexicon constraint: restricting the large model to select tokens only from a predefined lexicon, excluding words unrelated to structured fields; (2) Syntax constraint: restricting the output to conform to a predefined syntactic structure through finite state automata or context-free grammars; (3) Bundle search constraint: during the bundle search process, forcibly retaining candidate sequences containing specified keywords / fields, and discarding paths that do not contain necessary fields.

[0065] In this embodiment, the semantics of the final generated target description text using a multi-controlled approach are controlled. First, the semantic weights of the fields are determined based on the scenario and job characteristics. Then, the field order is determined based on these semantic weights. This allows for the determination of the importance of different fields for information matching in job search scenarios based on the specific job type requirements of different scenarios. Furthermore, the order of each field is determined based on its importance (semantic weight), achieving stable control over the semantic weights. Next, a predefined filling template is determined based on the scenario and field order, and each field is filled into its corresponding position in a controlled manner according to the predefined template, achieving a second level of semantic and sequential control over the final generated initial description text. Finally, corresponding reinforcing keywords are added to each field based on its semantic weight, further emphasizing the semantics of important field content. This ensures that the final generated target description text clearly distinguishes key job requirements semantically, preventing semantic deviations in important features due to semantic transformation diversity during semantic generation. This increases the semantic accuracy of important information features while maintaining semantic diversity.

[0066] S206. Enumerate the category information in the structured information to be matched by fields, and vectorize the symbol information after field enumeration to obtain a semantic vector.

[0067] In this embodiment, category information is also known as enumeration field; field enumeration is the enumeration mapping process for enumeration fields; in this embodiment, enumeration field is a field whose value is in a limited, predefined set, such as education information, which only includes doctoral, master's, bachelor's, junior college, etc.; enumeration mapping process is the digitization of these fields.

[0068] Specifically, for category information in the structured information to be matched, such as education level, job category, and work location, which are essentially discrete symbolic information, symbolic digitization can be performed, i.e., field enumeration, mapping text labels to unique numeric indices. For example, for education level, undergraduate, master's, and doctoral degrees can be mapped to numeric indices 0, 1, and 2, respectively. Furthermore, after symbolically digitizing the category information in the structured information to be matched, each numeric index can be semantically embedded, such as mapping each numeric index to a vector in a high-dimensional continuous space. This can be achieved using a trained semantic embedding model to obtain the semantic vector corresponding to each category. The role of this model is to bring semantically similar categories closer together in the vector space.

[0069] S207. Perform dimensionless processing on the data information in the structured information to be matched to obtain a scaled vector.

[0070] In this embodiment, the data information mainly refers to continuous measurement information, such as expected salary based on work experience / years of service and age. Dimensionless processing involves converting measurement information with units into a unified standard scale or discrete interval without units. In this embodiment, dimensionless processing can be performed by intervalization or normalization of the data.

[0071] Specifically, the data in the structured information to be matched can be processed to be dimensionless, that is, to be range-based or normalized. The appropriate processing method can be selected for each type of data. For example, the expected salary can be mapped to the range of 0-1; or different years of work experience can be mapped to several discrete ranges, such as 0-2 years of work experience as junior level, 3-5 years as intermediate level, and so on.

[0072] S208. Encode the multi-value field information in the structured information to be matched to obtain a multi-value encoded vector.

[0073] In this embodiment, the multiple multi-valued fields are essentially variable-length sets of information, such as skill tags, language abilities, and welfare preferences. The encoding process in this embodiment can be one-hot encoding or rule vector processing.

[0074] Specifically, in real-world recruitment scenarios, skill tags in different resumes or job postings may have different values. For example, resume A might have 3 skill tags, while resume B might have 5. Therefore, to make subsequent vector processing faster and more accurate, this multi-valued field information can be encoded into a "fixed-length numerical vector." One implementation involves sorting and encoding the tags that may appear in each multi-valued field. In practice, if a tag appears in the actual multi-valued field, its corresponding encoded value is 1; if it doesn't appear, its encoded value is 0. Furthermore, the sorting order of the tags corresponding to each value in the final multi-valued encoded vector is the same as the pre-set sorting order. Another implementation utilizes methods such as counting vectors, term frequency-inverse document frequency, and embedding vector averaging to obtain the multi-valued encoded vector.

[0075] S209. Input the semantic vector, the scaling vector, and the multi-value encoding vector into the second channel to combine structural features and obtain the constraint vector to be matched.

[0076] Specifically, the obtained semantic vector, scaling vector, and multi-valued encoding vector are input into the second channel for structural feature combination to obtain the constraint vector to be matched.

[0077] Optionally, steps S206-S208 above can be performed one or multiple times, that is, steps S206-S208 are parallel steps; if not all of the semantic vector, scaling vector and multi-valued encoding vector exist, then only the existing vectors can be merged to obtain the constraint vector to be matched.

[0078] In this embodiment, by performing different forms of structural feature vectorization processing on the structured information to be matched, it is possible not only to select the appropriate structural feature processing method according to the different types (such as enumerated fields, numerical fields, and multi-valued fields) of the structural fields in the structured information to be matched, ensuring the accuracy of the processing, but also to a certain extent to reduce the time and error added by processing the same field with different processing methods. This ensures that the obtained constraint vector to be matched can accurately reflect the factual constraints and hard condition features of the information to be matched, providing the most critical information points for subsequent matching. Furthermore, regarding the above steps, the method of this embodiment can still achieve stable semantic matching based on natural language semantic processing even when the structured fields are missing or non-standard, and can achieve adaptive dynamic semantic understanding and dynamic job weight adjustment for non-standard resumes or long-tail positions.

[0079] S210. For each candidate information vector group corresponding to the scenario to be matched with the structured information, calculate the first similarity between the candidate language description vector and the language description vector to be matched in the candidate information vector group, and calculate the second similarity between the candidate constraint vector and the constraint vector to be matched in the candidate information vector group.

[0080] Specifically, for each candidate information vector group corresponding to a scenario with structured information to be matched, a first similarity can be calculated between the candidate language description vectors obtained in advance according to the above steps and the language description vector to be matched; and a second similarity can be calculated between the candidate constraint vectors obtained in advance according to the above steps and the constraint vector to be matched. Optionally, the similarity calculation method in this embodiment can be such as Jaccard similarity, dot product similarity, and cosine similarity, etc., and is not limited here.

[0081] S211. Based on the job characteristics and requirements for the matching language description vector and the matching constraint vector, determine the first dynamic weight assigned to the first similarity and the second dynamic weight assigned to the second similarity.

[0082] Specifically, different job types have varying degrees of demand for hard requirements (corresponding to the constraint vector to be matched) or soft requirements (corresponding to the language description vector to be matched) in the recruitment and job application process. Therefore, the first dynamic weight assigned to the first similarity and the second dynamic weight assigned to the second similarity can be determined based on the job feature requirements. For example, when the job feature requirements are more inclined towards hard requirements, such as a hardware engineer in the aerospace field, a higher second dynamic weight can be assigned to the second similarity corresponding to the constraint vector to be matched, and a lower first dynamic weight can be assigned to the first similarity. Similarly, when the job feature requirements are more inclined towards soft requirements, such as a sales or customer service position, a higher first dynamic weight can be assigned to the first similarity corresponding to the language description vector to be matched, and a lower second dynamic weight can be assigned to the second similarity. Furthermore, if the job is a hardware engineer with no particular restrictions on the specific work scenario, the same dynamic weight can be assigned to the first and second similarities, i.e., the first dynamic weight equals the second dynamic weight. The specific dynamic weight can be determined based on other requirements in the job feature requirements, such as the degree of user demand for the position.

[0083] Optionally, the determination of dynamic weights in this embodiment can be achieved using a pre-trained weight allocation model. For example, during training, the sample matching language description vector and sample constraint vector, obtained after processing the structured information of each sample, are used as inputs to the weight allocation model. This allows the model to determine the job feature requirements and the user's potential needs from the two aspects of the structured information of each sample, and then output dynamic weights corresponding to different vectors. Another feasible approach is to pre-define a rule engine for different job features, such as A scores for non-technical positions, B scores for technical positions, and different C scores for different work scenarios. x The score (where x is the subscript corresponding to different work scenarios) has different D values ​​for different job requirements. y The score (where y is the subscript corresponding to different work scenarios),...; In practical applications, based on the job feature requirements corresponding to the structured information to be matched, a total score is calculated by the rule engine as described above (e.g., if a technical position corresponds to scenario x1 and the job requirement is y1, then the total score is B+C). x1 +D y1 Then, based on the total score and different score thresholds determined in advance in the experiment or test (the score thresholds are used to characterize the possible values ​​of the first dynamic weight and the second dynamic weight when different score ranges are met), the first dynamic weight and the second dynamic weight are determined.

[0084] S212. Calculate the comprehensive similarity between the candidate information vector group and the structured information to be matched based on the first dynamic weight, the first similarity, the second dynamic weight, and the second similarity.

[0085] Specifically, the comprehensive similarity between the candidate information vector group and the structured information to be matched can be calculated by weighted summation. For example, the first dynamic weight is multiplied by the first similarity, the second dynamic weight is multiplied by the second similarity, and the products of the two are added together to obtain the comprehensive similarity between the candidate information vector group and the structured information to be matched.

[0086] In this embodiment, the system determines whether the structured information to be matched prioritizes linguistic description features or constraint features in the candidate information based on the job feature requirements corresponding to the structured information to be matched. Different weights are then assigned to the similarity of these two aspects, resulting in a first dynamic weight and a second dynamic weight. The comprehensive similarity between the candidate information vector group and the structured information to be matched is then calculated. This achieves dynamic allocation of weights corresponding to different similarities based on the features emphasized by the structured information to be matched, making the determined comprehensive similarity more accurately reflect the degree of similarity between the candidate information that the structured information to be matched truly needs to match. This provides a more accurate comprehensive similarity result for matching more similar candidate information to the structured information to be matched. Furthermore, this embodiment can simultaneously satisfy the comparison of semantic similarity and structured constraint similarity, significantly reducing rule maintenance costs, improving the scalability of information matching, and increasing the matching coverage of long-tail positions and non-standard resumes.

[0087] S213. Based on the comprehensive similarity between all candidate information vector groups and the structured information to be matched, sort the candidate information vector groups corresponding to the comprehensive similarity in descending order to obtain the vector group sorting result.

[0088] Specifically, after calculating the comprehensive similarity between each candidate information vector group and the structured information to be matched, the candidate information vector groups corresponding to all comprehensive similarities can be sorted in descending order of comprehensive similarity to obtain the vector group sorting result.

[0089] S214. Replace each candidate information vector group in the sorting result of the vector group with the recruitment and job-seeking information corresponding to the candidate information vector group to obtain the sorting result of the recruitment and job-seeking information.

[0090] Specifically, since each candidate information vector group corresponds to a job posting / job seeking information, for example, if the information to be matched this time is a job posting, then all candidate information is job seeking information. Therefore, the candidate information vector group is the vector group of all job seeking information obtained after pre-processing all job seeking information. Thus, each candidate information vector group in the vector group sorting result can be replaced with its corresponding job posting / job seeking information to obtain the sorting result of the job posting / job seeking information. Optionally, in the sorting result of this embodiment, the job posting / job seeking information with the lowest overall similarity can be deleted in advance to ensure that the job posting / job seeking information displayed to the user has a certain degree of matching relevance, and the matching information first displayed to the user has a higher matching relevance.

[0091] S215. Display the ranking results of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching results.

[0092] Specifically, after obtaining the ranking results of recruitment and job seeking information, the ranking results of the comprehensive similarity between all candidate information vector groups and the structured information to be matched, that is, the final ranking results of recruitment and job seeking information, can be used as the information matching results and displayed to the users corresponding to the information to be matched in rotation according to the ranking results.

[0093] For example, the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched is displayed as the information matching result, including: (a) Obtain the pre-set number of extractions per cycle.

[0094] The number of periodic extractions is the number of job postings that can be extracted in each pre-set extraction period.

[0095] Specifically, a fixed number of extractions per period can be preset, or a dynamic number of extractions per period can be assigned based on the size of the display area of ​​the device interface of the user corresponding to the information to be matched.

[0096] (ii) For the current extraction period, extract recruitment and job information from the sorting results of recruitment and job information in descending order according to the number of extractions per period, and display the recruitment and job information as the information matching results corresponding to the current extraction period.

[0097] The current extraction period refers to the period during which recruitment and job-seeking information is currently being extracted. It is worth noting that in this embodiment, the sorting result of the recruitment and job-seeking information is the directly obtained sorting result in the first "current extraction period," while in subsequent "current extraction periods," it is the sorting result of previously extracted recruitment and job-seeking information that has been deleted.

[0098] Specifically, for each current extraction period, recruitment and job information can be extracted from the sorted results of recruitment and job information in descending order according to the number of extractions per period. The number of recruitment and job information extracted per period is used as the information matching result corresponding to the current extraction period, and these information matching results are displayed.

[0099] (iii) Remove the recruitment and job search information corresponding to the current extraction period that has been displayed from the sorting results to obtain the updated sorting results of recruitment and job search information.

[0100] Specifically, after extraction or display, the recruitment and job-seeking information extracted (displayed) in the current extraction period can be deleted from the sorting results to update the sorting results of recruitment and job-seeking information. Optionally, in this embodiment, if a user is satisfied with a certain recruitment and job-seeking information or wants to include it as a candidate during the display process, they can add it to their favorites, but it will still be deleted from the current sorting results and will only be displayed again after all sorting results have been displayed.

[0101] (iv) After receiving the extraction instruction to enter the next extraction cycle, the next extraction cycle is taken as the current extraction cycle, and the process of extracting recruitment and job information from the sorting results of recruitment and job information in descending order according to the number of extractions per cycle is returned to the execution until it is determined that all recruitment and job information has been displayed.

[0102] Specifically, after the user selects from the recruitment and job information extracted in the current extraction cycle, if the selection is completed, the user can send an extraction instruction to the information matching device through the user terminal. At this time, the information matching device can determine to enter the next extraction cycle after receiving the extraction instruction, and then take the next extraction cycle as the current extraction cycle, and return to execute step (II) until it is determined that all recruitment and job information has been displayed.

[0103] In this embodiment, recruitment and job search information is extracted and displayed to the user in different extraction cycles according to the ranking results of comprehensive similarity. This can provide the user with the most similar corresponding information to the information to be matched provided by the user from the beginning, and achieve full coverage matching of the information to be matched.

[0104] Figure 3 This is a schematic diagram of the structure of an information matching device provided in an embodiment of the present invention, as shown below. Figure 3 The device shown includes: The processing module 301 is used to perform controlled first-channel semantic processing on the structured information to be matched in the recruitment and job search scenario to obtain the language description vector to be matched, and to perform second-channel structural feature processing on the structured information to be matched to obtain the constraint vector to be matched.

[0105] The calculation module 302 is used to calculate the first similarity between the candidate language description vector in the candidate information vector group and the language description vector to be matched for each candidate information vector group corresponding to the scenario of the structured information to be matched, and to calculate the second similarity between the candidate constraint vector in the candidate information vector group and the constraint vector to be matched.

[0106] The matching module 303 is used to determine the comprehensive similarity between the candidate information vector group and the structured information to be matched based on the job feature requirements, the first similarity, and the second similarity of the structured information to be matched, and to display the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching result.

[0107] Optionally, before performing controlled first-channel semantic processing on the structured information to be matched in the recruitment and job-seeking scenario to obtain the language description vector to be matched, the processing module 301 is further used for: It acquires recruitment and job-seeking information in recruitment and job-seeking scenarios, and generates structured information to be matched based on the scenario, the key fields corresponding to the scenario, and the recruitment and job-seeking information.

[0108] Optionally, the structured information to be matched in the recruitment and job-seeking scenario undergoes controlled first-channel semantic processing to obtain the language description vector to be matched. Specifically, processing module 301 is used for: Based on the scenario and the job characteristics required within that scenario, determine the semantic weights of the fields, and then determine the order of the fields according to these semantic weights. Based on the scenario and the order of the fields, determine a predefined filling template, and fill the structured information to be matched into the predefined filling template according to the semantics of each field to obtain the initial description text. For each field in the initial description text, determine the strengthening keywords based on the corresponding semantic weights of the fields, and inject the strengthening keywords into the fields of the initial description text to obtain the target description text. Input the target description text into the first channel for semantic processing to obtain the language description vector to be matched output by the first channel.

[0109] Optionally, the structured information to be matched is processed using the second channel's structural features to obtain the constraint vector to be matched. Specifically, processing module 301 is used for: The category information in the structured information to be matched is enumerated into fields, and the symbol information after field enumeration is vectorized to obtain a semantic vector; the data information in the structured information to be matched is dimensionless to obtain a scaled vector; the multi-valued field information in the structured information to be matched is encoded to obtain a multi-valued encoded vector; the semantic vector, scaled vector, and multi-valued encoded vector are input into the second channel for structural feature combination to obtain the constraint vector to be matched.

[0110] Optionally, based on the job feature requirements of the structured information to be matched, the first similarity, and the second similarity, the comprehensive similarity between the candidate information vector group and the structured information to be matched is determined. The matching module 303 is specifically used for: Based on the job characteristics and the degree of need for matching the language description vector and the constraint vector, a first dynamic weight is assigned to the first similarity and a second dynamic weight is assigned to the second similarity. The comprehensive similarity between the candidate information vector group and the structured information to be matched is calculated based on the first dynamic weight, the first similarity, the second dynamic weight, and the second similarity.

[0111] Optionally, before displaying the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching result, the matching module 303 is also used to: Based on the comprehensive similarity between all candidate information vector groups and the structured information to be matched, the candidate information vector groups corresponding to the comprehensive similarity are sorted in descending order to obtain the vector group sorting result; each candidate information vector group in the vector group sorting result is replaced with the recruitment and job application information corresponding to the candidate information vector group to obtain the recruitment and job application information sorting result.

[0112] Optionally, the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched is displayed as the information matching result. The matching module 303 is specifically used for: Obtain the pre-set number of entries to be extracted per period; for the current extraction period, extract job postings and job seeking information from the sorted results in descending order according to the number of entries to be extracted per period, and display the job postings and job seeking information as the information matching result corresponding to the current extraction period; delete the displayed job postings and job seeking information corresponding to the current extraction period from the sorted results, and obtain the updated sorted results of job postings and job seeking information; upon receiving the extraction instruction to enter the next extraction period, take the next extraction period as the current extraction period, and return to execute the step of extracting job postings and job seeking information from the sorted results in descending order according to the number of entries to be extracted per period, until it is determined that all job postings and job seeking information has been displayed.

[0113] The information matching device provided in the embodiments of the present invention can execute the information matching method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0114] It is worth noting that in the embodiments of the above information matching device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0115] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0116] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0117] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0118] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as information matching methods.

[0119] In some embodiments, the information matching method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the information matching method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the information matching method by any other suitable means (e.g., by means of firmware).

[0120] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements, for example, the information matching method provided in this invention.

[0121] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0122] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0123] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0124] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the information matching method provided in any embodiment of this invention.

[0125] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0126] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0127] Furthermore, the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with the relevant provisions of national laws and regulations.

[0128] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. An information matching method, characterized in that, The method includes: The structured information to be matched in the recruitment and job search scenario is subjected to controlled semantic processing in the first channel to obtain the language description vector to be matched, and the structured information to be matched is subjected to structural feature processing in the second channel to obtain the constraint vector to be matched. For each candidate information vector group corresponding to the scenario of the structured information to be matched, calculate the first similarity between the candidate language description vector in the candidate information vector group and the language description vector to be matched, and calculate the second similarity between the candidate constraint vector in the candidate information vector group and the constraint vector to be matched. Based on the job feature requirements of the structured information to be matched, the first similarity and the second similarity, the comprehensive similarity between the candidate information vector group and the structured information to be matched is determined, and the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched is displayed as the information matching result.

2. The method according to claim 1, characterized in that, Before performing controlled first-channel semantic processing on the structured information to be matched in a recruitment and job-seeking scenario to obtain the language description vector to be matched, the process also includes: The system acquires recruitment and job-seeking information in recruitment and job-seeking scenarios, and generates structured information to be matched based on the scenario of the recruitment and job-seeking information, the key fields corresponding to the scenario, and the recruitment and job-seeking information.

3. The method according to claim 2, characterized in that, The controlled first-channel semantic processing of the structured information to be matched in the recruitment and job-seeking scenario yields a language description vector to be matched, including: The semantic weights of the fields are determined based on the scenario and the job characteristics required in the scenario, and the order of the fields is determined based on the semantic weights of the fields. A predefined filling template is determined based on the scenario and the order of the fields, and the structured information to be matched is filled into the predefined filling template according to the semantics of each field to obtain the initial description text; For each field in the initial description text, reinforcement keywords are determined according to the semantic weight of the field, and the reinforcement keywords are injected into the field of the initial description text to obtain the target description text; The target description text is input into the first channel for semantic processing to obtain the language description vector to be matched output by the first channel.

4. The method according to claim 1, characterized in that, The step of performing structural feature processing on the structured information to be matched in the second channel to obtain the constraint vector to be matched includes: The category information in the structured information to be matched is enumerated into fields, and the symbol information after field enumeration is vectorized to obtain a semantic vector; The data information in the structured information to be matched is processed to be dimensionless to obtain a scaled vector; The multi-value field information in the structured information to be matched is encoded to obtain a multi-value encoded vector; The semantic vector, the scaling vector, and the multi-value encoding vector are input into the second channel to combine structural features, thereby obtaining the constraint vector to be matched.

5. The method according to claim 1, characterized in that, The step of determining the comprehensive similarity between the candidate information vector group and the structured information to be matched based on the job feature requirements of the structured information to be matched, the first similarity, and the second similarity includes: Based on the degree of demand for the language description vector to be matched and the constraint vector to be matched according to the job feature requirements, a first dynamic weight is determined for the first similarity and a second dynamic weight is determined for the second similarity. The comprehensive similarity between the candidate information vector group and the structured information to be matched is calculated based on the first dynamic weight, the first similarity, the second dynamic weight, and the second similarity.

6. The method according to claim 1, characterized in that, Before displaying the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching result, the following is also included: Based on the comprehensive similarity between all candidate information vector groups and the structured information to be matched, the candidate information vector groups corresponding to the comprehensive similarity are sorted in descending order to obtain the vector group sorting result; Replace each candidate information vector group in the sorting result of the vector group with the recruitment and job-seeking information corresponding to the candidate information vector group to obtain the sorting result of the recruitment and job-seeking information.

7. The method according to claim 6, characterized in that, The ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched is displayed as the information matching result, including: Get the pre-set number of extractions per cycle; For the current extraction period, extract recruitment and job information from the sorting results of the recruitment and job information in descending order according to the number of extractions in the period, and display the recruitment and job information as the information matching result corresponding to the current extraction period; The displayed job postings and job seeking information corresponding to the current extraction period are removed from the sorting results to obtain an updated sorting result for the job postings and job seeking information. Upon receiving the extraction instruction to enter the next extraction cycle, the next extraction cycle is taken as the current extraction cycle. The process returns to the step of extracting recruitment and job-seeking information from the sorted results of the recruitment and job-seeking information in descending order of sorting order according to the number of extractions in the cycle, until it is determined that all recruitment and job-seeking information has been displayed.

8. An information matching device, characterized in that, The device includes: The processing module is used to perform controlled first-channel semantic processing on the structured information to be matched in the recruitment and job search scenario to obtain the language description vector to be matched, and to perform second-channel structural feature processing on the structured information to be matched to obtain the constraint vector to be matched. The calculation module is used to calculate, for each candidate information vector group corresponding to the scenario of the structured information to be matched, the first similarity between the candidate language description vector in the candidate information vector group and the language description vector to be matched, and to calculate the second similarity between the candidate constraint vector in the candidate information vector group and the constraint vector to be matched. The matching module is used to determine the comprehensive similarity between the candidate information vector group and the structured information to be matched based on the job feature requirements of the structured information to be matched, the first similarity and the second similarity, and to display the ranking result of the comprehensive similarity between all candidate information vector groups and the structured information to be matched as the information matching result.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the information matching method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the information matching method as described in any one of claims 1 to 7.

Images