Personnel position matching method and system and related equipment

Abstract

The invention provides a personnel position matching method and system and related equipment thereof. The method comprises the following steps: acquiring position analysis data; query index construction is conducted on the position analysis data through a multi-modal query model, and a mixed query index is obtained and comprises a keyword index and a vector index; based on the keyword index and the vector index, matching position contents corresponding to the position analysis data in the target position set to obtain a matching result set; based on the matching result set, semantic analysis is conducted on the position requirements and the candidate objects through a preset language model, the matching result set is reordered according to analysis results, and matching data are obtained; and based on the matching data, generating a structured matching result, and performing pushing and record management on the structured matching result. Through the steps of the method, multi-modal fusion retrieval, accurate matching and structured output of the position requirements and the candidate objects can be achieved, and the availability and traceability of the matching result are improved.

Description

Technical Field

[0001] This invention relates to the field of natural language processing, and more particularly to a method, apparatus, system, electronic device, and storage medium for matching personnel to job positions. Background Technology

[0002] With the rapid development of online recruitment platforms, enterprise talent management systems, and multi-role organizational collaboration platforms, job matching is gradually evolving from manual screening to intelligent, data-driven methods. Job descriptions typically include skill requirements, qualifications, industry background, and other semantic elements, requiring parsing and feature representation through multimodal or natural language models. Simultaneously, candidate information is becoming increasingly diverse and unstructured, making it difficult to achieve a precise match between jobs and candidates through simple keyword comparison. Therefore, building efficient, accurate, and interpretable job matching technology has become a crucial technical requirement for talent management systems.

[0003] Existing job matching solutions generally use keyword matching, rule filtering, or single-vector similarity calculation for retrieval and comparison. Keyword matching lacks semantic awareness and cannot distinguish between semantic similarities and contextual relationships between skills. Single-vector semantic retrieval is prone to failure when faced with missing fields, noisy text, or incomplete job descriptions, resulting in unreliable retrieval results. Existing technologies typically lack a complete link that integrates retrieval, sorting, truncation, and structured output. Matching results are difficult to use directly in business systems, and there is a lack of push and status recording mechanisms, which is not conducive to closed-loop management and subsequent review.

[0004] Among existing technologies, such as the online recruitment generative recommendation system proposed in "CN116452169B," user input is transformed into text features, and a large language model directly generates complete recommendation descriptions. Simultaneously, multi-stage training and reward models are used to constrain the quality of the generated content, thereby improving the interpretability and controllability of the recommendation results. This technology primarily focuses on "generative description quality and user interaction control," essentially representing an optimization of recommendation text generation strategies.

[0005] However, the aforementioned technologies tend to generate and rewrite candidate description text, and do not perform "retrieval-based matching" on massive job data: their output results cannot provide a candidate list, nor can they be sorted and compared based on multiple job entities; although the generated text is interpretable, it is difficult to form a reusable "structured field output", and it lacks data stability when used downstream for comparison, filtering, and batch calling.

[0006] Therefore, existing personnel and job matching methods have the problem of being unable to perform multimodal fusion retrieval, accurate matching, and structured traceable output of job parsing data. Summary of the Invention

[0007] This invention provides a personnel job matching method to solve the problem that existing personnel job matching methods cannot achieve multimodal fusion retrieval, accurate matching, and structured traceable output of job parsing data.

[0008] In a first aspect, the present invention provides a method for matching personnel to job positions, the method comprising the following steps: Obtain job analysis data; By using a multimodal query model, a query index is constructed on the job parsing data to obtain a hybrid query index, which includes a keyword index and a vector index. Based on the keyword index and vector index, the job content corresponding to the job parsing data is matched in the target job set to obtain matching data; Based on the matching data, structured matching results are generated, and push and record management are performed on the structured matching results.

[0009] Optionally, obtaining job parsing data includes: Extract at least one job description from the target resume set; Preprocess at least one of the job description data to obtain clean job description data; Semantic features are extracted from the clean job description data using a natural language model to determine the corresponding feature description elements, which include extracted skill elements, qualification elements, regional elements, and industry context elements. Based on the extracted skill elements, qualification elements, regional elements, and industry context elements, job analysis data for the corresponding resume positions is generated.

[0010] Optionally, the step of constructing a query index for the job parsing data using a multimodal query model to obtain a hybrid query index includes: The multimodal query model is used to perform feature representation calculations on the job parsing data to generate corresponding job query vectors. Keyword extraction is performed on the job parsing data to obtain keyword information, and a keyword query path is constructed based on the keyword information to generate a corresponding keyword index; The job query vector is validated, and a corresponding vector index is generated based on the validation result. Based on the keyword index and the vector index, a hybrid query index is determined.

[0011] Optionally, the step of validating the job query vector and generating a corresponding vector index based on the validation result includes: The verification result of the job query vector is determined by performing dimension consistency and non-zero checks on the job query vector. When the verification result is valid, the job query vector is written into a preset vector index structure to generate a vector index that can be used for subsequent retrieval processing; When the verification result is invalid, the generation of the vector index is prevented, causing the hybrid query index to degenerate into query and matching processing based solely on keywords.

[0012] Optionally, based on the keyword index and vector index, the matching of job content corresponding to the job parsing data in the target job set to obtain matching data includes: Based on the vector index, a semantic similarity retrieval is performed to obtain the first candidate set corresponding to the job query vector; Based on the keyword index, perform keyword matching retrieval to obtain a second candidate set corresponding to the keyword information; The first candidate set and the second candidate set are subjected to uniqueness processing to obtain a unique candidate set; Based on the vector similarity score, keyword matching score, and skill coverage score corresponding to each candidate object in the unique candidate set, a fusion score calculation is performed to generate the matching data.

[0013] Optionally, the step of performing a fusion score calculation based on the vector similarity score, keyword matching score, and skill coverage score corresponding to each candidate object in the unique candidate set to generate the matching data includes: Based on the preset fusion weights, the vector similarity score, keyword matching score, and skill coverage score are weighted and calculated to obtain the fusion score. Based on the fusion score, the candidate objects in the unique candidate set are sorted to obtain sorted data; Based on the output control threshold, the sorted data is truncated to obtain the target candidate set; Matching data is generated based on the fusion score and corresponding candidate object data of the target candidate set.

[0014] Optionally, the step of generating structured matching results based on the matching data, and performing push and record management on the structured matching results, includes: According to a preset format, the matching data is processed in a structured manner to obtain the target structured matching result; The target structured matching results are pre-formatted and packaged, and sent to the target terminal for display or invocation through a message push channel, and the corresponding record information is updated so that the target terminal can track and manage the target structured matching results.

[0015] Secondly, the present invention also provides a personnel job matching device, the personnel job matching device comprising: The first acquisition module is used to acquire job parsing data; The first construction module is used to construct a query index for the job parsing data through a multimodal query model, resulting in a hybrid query index, which includes a keyword index and a vector index. The first matching module is used to match the job content corresponding to the job parsing data in the target job set based on the keyword index and vector index to obtain matching data; The first execution module is used to generate structured matching results based on the matching data, and to perform push and record management on the structured matching results.

[0016] Thirdly, the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the personnel job matching method provided by the present invention.

[0017] Fourthly, the present invention provides a personnel job matching system, wherein the personnel job matching system stores a computer program, and when the computer program is executed by a processor, it implements the steps in the personnel job matching method provided by the invention.

[0018] This invention acquires job parsing data; it constructs a query index for the job parsing data using a multimodal query model, resulting in a hybrid query index, which includes a keyword index and a vector index; based on the keyword index and vector index, it matches the job content corresponding to the job parsing data within a target job set, obtaining matching data; based on the matching data, it generates structured matching results and performs push and record management on the structured matching results. Through the above method steps, it is possible to achieve multimodal fusion retrieval, accurate matching, and structured output of job requirements and candidate objects, improving the usability and traceability of matching results. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0020] Figure 1This is a flowchart of a personnel job matching method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of another personnel job matching device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, Figure 1 This is a flowchart of a personnel job matching method provided by an embodiment of the present invention. The personnel job matching method includes the following steps: 101. Obtain job analysis data.

[0023] In this embodiment of the invention, the above-mentioned personnel job matching method can be applied to a personnel job matching platform. The personnel job matching platform has functions such as job matching data processing, job matching data sending and receiving, and job matching data memory storage. It can be built based on a server or server cluster. The server or server cluster can be an electronic device with job matching data processing capabilities.

[0024] The aforementioned job parsing data can be a structured semantic element set obtained after preprocessing and feature extraction of the original job description text. Specifically, the aforementioned job matching platform can extract key requirements from the job description text, including skill requirements (such as "familiar with Python" and "possessing data analysis capabilities"), qualification requirements (such as "more than 3 years of experience"), industry background (such as "Internet finance"), geographical relevance (such as "Shanghai" and "able to work remotely"), and other semantic contexts.

[0025] The aforementioned original job description text can refer to job description content directly obtained from corporate recruitment systems, public recruitment platforms, user input interfaces, or third-party human resources systems that has not undergone any cleaning, parsing, or structuring processing, including but not limited to natural language text such as job responsibilities, skill requirements, experience conditions, industry information, and regional descriptions.

[0026] 102. By using a multimodal query model, a query index is constructed for the job parsing data to obtain a hybrid query index.

[0027] In this embodiment of the invention, the multimodal query model described above can be used to express semantic features of job parsing data. In this embodiment, the multimodal query model may include, but is not limited to, text embedding models, cross-modal representation models, and natural language models with semantic encoding capabilities (such as BERT, Sentence Transformer, and LLM Encoder). It is understood that the multimodal query model can take job parsing data as input and output a vectorized semantic representation of job requirements, and can simultaneously support data input in different formats such as text, tags, and skill lists.

[0028] In this embodiment, the aforementioned personnel-job matching platform can also generate query information that can be used for retrieval based on job parsing data. Specifically, job query vectors are generated through the aforementioned multimodal model for semantic retrieval; lexical tags (such as "Java", "backend development", "MySQL") in the job are extracted based on the keyword extraction module and organized into keyword query paths for keyword retrieval; the validity of the aforementioned query vectors is verified, and the usable retrieval paths are determined based on the results. The final query index is the query structure used by the job side to retrieve candidate objects.

[0029] The aforementioned hybrid query index can be a query structure composed of both keyword indexes and vector indexes, used to simultaneously perform semantic retrieval and keyword retrieval during the matching phase. It should be noted that when the vector index is valid, the aforementioned personnel and job matching platform can perform semantic similarity matching based on the vector index; when the vector index is invalid or unstable, it can automatically switch to or supplement retrieval using keyword paths, thus forming a dual-channel compatible query engine structure.

[0030] The aforementioned keyword index may be a set of keywords extracted from job parsing data by the job matching platform, and logical query paths are constructed based on these keywords. These keywords are then used to perform term-based retrieval operations in the candidate job set or candidate object set.

[0031] The aforementioned vector index can be a vector query structure constructed by the aforementioned personnel and job matching platform based on the job query vector, used to perform semantic approximate retrieval. It should be noted that the aforementioned vector index can be used to calculate similarity with the pre-built vector library of candidate objects in order to obtain semantically similar candidate objects or job entries.

[0032] 103. Based on keyword indexing and vector indexing, match the job content corresponding to the job parsing data in the target job set to obtain matching data.

[0033] In this embodiment of the invention, the aforementioned target job set may refer to the job database or candidate database used by the personnel job matching platform to perform matching calculations. In this embodiment, the target job set may include, but is not limited to, historical job data, open recruitment job data, or job descriptions managed by the system. It should be noted that each job in the target job set includes job content, structured skills, tags, industry information, etc., which can provide candidate objects for matching calculations.

[0034] The aforementioned job content can refer to the basic information of each job in the target job set, including but not limited to text descriptions, skill tags, and job requirements. It is understood that the job matching platform will extract corresponding features from this content during the matching phase and perform semantic and keyword comparisons with the job parsing data.

[0035] In this embodiment, the aforementioned personnel job matching platform can perform dual-channel retrieval in the target job set based on a hybrid query index, and use semantic similarity, keyword matching degree, and skill matching as comprehensive indicators to score and rank the results, so as to determine the correspondence between job parsing data and candidate jobs. Specifically, matching can be performed through steps such as semantic measurement, keyword filtering, comprehensive scoring, and candidate truncation to obtain the corresponding matching data.

[0036] The aforementioned matching data can be a set of candidate objects obtained by the aforementioned personnel and job matching platform based on semantic retrieval and keyword retrieval, along with unstructured data such as similarity scores, matching tags, and skill coverage.

[0037] 104. Based on the matching result set, semantic analysis of job requirements and candidate objects is performed using a preset language model, and the matching result set is reordered according to the analysis results to obtain matching data.

[0038] In this embodiment of the invention, after obtaining the target candidate set based on the fusion score and output control threshold, the personnel-job matching platform can further perform secondary screening and re-ranking processing on the target candidate set. Specifically, the information such as the name, skill elements, qualification elements, and industry experience of each candidate in the target candidate set, along with the job responsibilities, required skills, and priority conditions from the job parsing data, are organized into a unified analysis input and input into a preset natural language processing model. The model then outputs analysis results for each candidate, including the comprehensive matching degree, skill matching degree, experience matching degree, advantages, and items to be verified. Based on these analysis results, the personnel-job matching platform re-ranks the candidate objects in the target candidate set to form target matching data for subsequent structured output.

[0039] 105. Based on the matching data, generate structured matching results, and perform push and record management on the structured matching results.

[0040] In this embodiment of the invention, the structured matching result can be the final output content generated by the personnel and job matching platform based on the matching data, including the candidate's name, overall matching degree, skill matching degree, experience matching degree, strengths, items to be verified, and information such as the account manager associated with the candidate and the resume update time. This is used to display and track the result in a table format within the business system. It may also include table data with fixed fields (such as job title, skill requirements, and similarity score) and a TOP N recommendation list. In this embodiment, the platform also performs encapsulation, push, recording, and tracking management of the structured matching result, enabling it to be displayed on the user's terminal and supporting subsequent review operations.

[0041] In one possible embodiment, the aforementioned personnel job matching platform first parses the input job description, extracts skills, qualifications, industry and semantic elements, generates job parsing data, calculates job query vectors through a multimodal query model and extracts keyword information to construct a hybrid query index, performs semantic retrieval and keyword retrieval in the target job set based on the index, generates matching data, formats the matching data into structured matching results, and pushes them to the terminal through a message channel, while recording corresponding status information for tracking and management.

[0042] By employing the above methods and steps, a closed-loop link can be formed between job parsing, index building, hybrid retrieval, rating ranking, and structured output, achieving high-precision matching that combines semantic understanding and keyword filtering; and by improving the usability, traceability, and operational efficiency of the results through structured results and push management.

[0043] In this embodiment of the invention, job parsing data is acquired; a multimodal query model is used to construct a query index for the job parsing data, resulting in a hybrid query index, which includes a keyword index and a vector index; based on the keyword index and the vector index, the job content corresponding to the job parsing data is matched within the target job set to obtain matching data; based on the matching data, structured matching results are generated, and push and record management are performed on the structured matching results. Through the above method steps, multimodal fusion retrieval, accurate matching, and structured output of job requirements and candidate objects can be achieved, improving the usability and traceability of the matching results.

[0044] Optionally, in the step of obtaining job parsing data, at least one job description data can be obtained from the target resume set; the at least one job description data can be preprocessed to obtain clean job description data; semantic features can be extracted from the clean job description data using a natural language model to determine the corresponding feature description elements, including extracted skill elements, qualification elements, regional elements, and industry context elements; based on the extracted skill elements, qualification elements, regional elements, and industry context elements, job parsing data for the corresponding resume job can be generated.

[0045] In this embodiment of the invention, the aforementioned target resume set may refer to the resume data set used by the personnel job matching platform in the current matching process to analyze or extract job description data. It may be the user resume database that has been stored in the past by the personnel job matching platform, the candidate data source synchronized from the outside, or the enterprise talent pool that has been authorized to access, including but not limited to structured and unstructured fields such as personal information, skill tags, past work experience and candidate self-description content.

[0046] The aforementioned job description data can refer to the job intentions or job descriptions filled in by candidates and recorded in the target resume collection. For example, a candidate might specify in their resume that they "expect to work in data analysis, are familiar with Python and SQL, and hope to develop a career in the Shanghai internet industry," in order to construct a characteristic expression related to the job requirements.

[0047] In this embodiment, the aforementioned personnel job matching platform can perform standardization operations on the original job description data, including removing redundant punctuation marks, unifying capitalization, cleaning up meaningless stop words, extracting keyword fragments, sentence segmentation, encoding format correction, and deleting advertising or noisy text, in order to achieve the purpose of preprocessing and thus obtain clean job description data.

[0048] The aforementioned clean job description data can refer to descriptive text data that does not contain noisy symbols, incomplete fragments, irrelevant fields, or duplicate strings, and can accurately express the candidate's job intentions, skill background, and industry preferences described in their resume. For example, "Familiar with Python data analysis, has 3 years of experience in the financial industry, and expects to work in Shanghai" is considered clean job description data.

[0049] The aforementioned natural language models can also refer to text encoding models with semantic understanding capabilities, such as BERT, ERNIE, Transformer encoders, LLM embedding models, etc., which are used to transform clean job description data into measurable and comparable semantic vectors or feature representations for subsequent extraction of semantic elements such as skills, qualifications, and industry.

[0050] In this embodiment, the aforementioned personnel job matching platform can utilize natural language models to perform semantic analysis on the input text, identifying and extracting key semantic information contained within the text, including skill vocabulary, time-related descriptions, location information, industry vocabulary, and contextual semantic relationships. For example, it can structurally identify the corresponding skills, qualifications, and industry semantic content from the description "familiar with Java backend development, more than five years of development experience, familiar with e-commerce business."

[0051] The aforementioned feature description elements refer to structured semantic units extracted through semantic feature extraction, used to represent the meaning of different dimensions in job description data. This embodiment includes: skill elements such as "Python," "data analysis," and "Java development"; qualification elements such as "3 years of experience" and "more than 5 years of development experience"; geographical elements such as "Shanghai" and "willing to travel"; and industry context elements such as "internet finance" and "e-commerce." It can be understood that these elements constitute the core expressive dimensions in the job description data.

[0052] The aforementioned skill elements can be used to identify a candidate's specific technical abilities or tool experience; the aforementioned qualification elements can be used to express experience level, educational background, or stage of development; the geographical element can be used to identify candidate preferences or work location restrictions; and the industry context element can be used to depict a candidate's industry experience background or habitual business scenarios. After structuring this content, the personnel and job matching platform can perform weighted integration and precise comparison based on different dimensions during the matching stage.

[0053] The aforementioned job matching platform can also perform structured integration based on the multiple feature description elements mentioned above to form job parsing data for the corresponding resumes. It is understood that the generated job parsing data is typically in a structured or semi-structured format, including but not limited to fielded skill lists, standardized qualification levels, regional indication information, and industry contextual tags, which can be directly used for index building and matching calculations in multimodal query models.

[0054] By following the above steps, high-quality job parsing data can be obtained from resume text, enabling standardized and structured expression of job elements. This improves the accuracy and stability of subsequent index construction and matching calculations, avoids recognition biases caused by factors such as noise in the original text and incomplete descriptions, and thus enhances the overall job matching effect.

[0055] Optionally, the step of constructing a query index for job parsing data using a multimodal query model to obtain a hybrid query index further includes: calculating feature representations of the job parsing data using a multimodal query model to generate corresponding job query vectors; extracting keywords from the job parsing data to obtain keyword information, and constructing keyword query paths based on the keyword information to generate corresponding keyword indexes; validating the job query vectors and generating corresponding vector indexes based on the validation results; and determining the hybrid query index based on the keyword index and the vector index.

[0056] In this embodiment of the invention, a multimodal query model can be used to semantically encode job parsing data, mapping skill elements, qualification elements, regional elements, industry context elements, and other content to a high-dimensional feature space in a unified manner, thereby generating a corresponding job query vector. Specifically, the multimodal query model can perform feature calculations such as embedding, encoding, and attention calculation on the input sequence, so that job element information from different sources is converted into a measurable and comparable vectorized representation.

[0057] The aforementioned job query vector can refer to a high-dimensional semantic vector generated from the output of feature representation calculations. It is used to characterize the overall meaning of the job parsing data, including but not limited to skill combinations, experience requirements, industry background, and semantic context. It can serve as a benchmark query vector for semantic retrieval, used to perform vector similarity matching within the candidate object set. It should be noted that the dimensionality, numerical distribution, and semantic quality of the aforementioned job query vector directly affect the recall effect during the retrieval stage.

[0058] In this embodiment, text segmentation, part-of-speech tagging, entity recognition, and skill tag recognition can be performed on the job parsing data to extract keywords that can be directly used for conditional retrieval. For example, keywords such as "Java," "development," "e-commerce," and "Shanghai" can be identified from the phrase "familiar with Java development, has e-commerce industry background, and prefers to work in Shanghai."

[0059] The keyword information mentioned above refers to the set of keywords obtained after keyword extraction, including but not limited to skill terms, industry terms, region terms, role terms, or other job-related tagged information. This keyword information is explicit and searchable, and can be used to perform fast search operations based on inverted index structures or tag filtering mechanisms.

[0060] In one possible embodiment, the aforementioned personnel and job matching platform can generate query logic for performing keyword matching based on keyword information, including constructing term combination relationships, defining matching conditions, and setting term priorities or thresholds. For example, for the three keywords "Python," "data analysis," and "Shanghai," the query path can be set based on the logical rule of "skills must match + region optional match." This can be understood as the aforementioned keyword query path providing explicit execution instructions for subsequent keyword index retrieval.

[0061] In another possible embodiment, the aforementioned personnel and job matching platform can perform quality checks on the generated job query vectors to determine whether the vectors meet the input requirements of semantic retrieval, including but not limited to detection methods such as vector dimension consistency detection, non-zero value detection, and numerical distribution detection. For example, when the job description is too short, semantic information is missing, or the model fails to encode effectively, the vectors may be all zero or do not meet the basic feature distribution, requiring a rollback mechanism.

[0062] The verification results can be used to indicate whether the job query vector can be used for subsequent vector retrieval. When the verification result is valid, the aforementioned personnel and job matching platform includes the vector in the vector index structure; when the verification result is invalid, the aforementioned personnel and job matching platform prevents the generation of the vector index and switches the subsequent retrieval process to a keyword path-based retrieval mode.

[0063] In this embodiment, after the keyword index and vector index are constructed, the personnel job matching platform combines the two to form a query structure that can participate in the retrieval together. It includes two parts: semantic retrieval path and keyword retrieval path. The vector can be dynamically selected or used in combination based on the effectiveness of the vector. It can perform dual-channel retrieval with both semantic matching and keyword filtering in the candidate object set, thereby improving the recall coverage and the robustness of the matching.

[0064] The above methods and steps can simultaneously support semantic retrieval and keyword retrieval, and can automatically fall back to the keyword path when vector anomalies occur, thereby improving the robustness, robustness, and recall coverage of job matching; the hybrid query index enables the retrieval results to have higher semantic accuracy and structural consistency, improving the overall matching quality.

[0065] Optionally, the steps of validating the job query vector and generating the corresponding vector index based on the validation result may also include determining the validation result of the job query vector by performing dimensional consistency and non-zero checks on the job query vector; when the validation result is valid, the job query vector is written into a preset vector index structure to generate a vector index that can be used for subsequent retrieval processing; when the validation result is invalid, the generation of the vector index is prevented, causing the hybrid query index to degenerate into query and matching processing based solely on keywords.

[0066] In this embodiment of the invention, the above-mentioned personnel job matching platform can perform basic quality checks on the generated job query vector to determine whether the vector can be used as a valid semantic retrieval input. Specifically, it can be done through the following two steps, wherein, dimension consistency check: used to detect whether the dimension of the job query vector is consistent with the system's preset vector dimension requirements, such as whether it is 512 dimensions or 768 dimensions. If the dimension is abnormal, it means that the model output is incomplete or the encoding fails. Non-zero check: Used to detect whether all values ​​in a vector are zero or nearly zero, such as a vector of [0,0,0,...,0] or extremely low value distribution. This usually indicates that the input content is missing, the model has not effectively parsed the text, or semantic extraction has failed.

[0067] When the above verification result is deemed valid, the personnel job matching platform performs the following operations: Write the job query vector as an available search term into the preset vector index structure; Generate corresponding vector index entries so that semantic similarity retrieval can be performed subsequently with the vector library of the candidate object set.

[0068] Specifically, job query vectors can be registered according to the format of a vector index structure, including assigning vector numbers, recording vector dimensions, storing vector values, and associating them with identifiers corresponding to job parsing data. This process writes the job query vectors into a preset vector index structure, generating a corresponding vector index, which enables the vectors to be quickly accessed and used during the retrieval phase.

[0069] When the verification result is invalid, the above personnel job matching platform performs the following operations: Prevent vector index generation: This means not registering any vector entries related to the position in the vector index structure to avoid using incorrect vectors in the retrieval process; Degradation: This means that the mixed query index automatically switches to using only the keyword index for retrieval and matching, and no longer calls the vector path.

[0070] The degradation mechanism allows the aforementioned job matching platform to continue effective matching through keyword paths even when job descriptions are incomplete, text noise is too high, or the model is not effectively parsed, thus avoiding retrieval failures or abnormal recommendation results.

[0071] By following the above steps, invalid or abnormal job query vectors can be automatically identified, enabling dynamic switching of semantic retrieval paths; ensuring the reliability and robustness of matching results, avoiding erroneous recommendations caused by abnormal model encoding, and ensuring that basic matching capabilities can still be maintained by relying on keyword retrieval in extreme cases, thereby improving the stability of matching retrieval.

[0072] The aforementioned preset vector index structure can be a fixed retrieval index structure set up to store job query vectors, and can adopt vector databases, ANN structures (such as HNSW, FAISS), inverted vector directories or other high-dimensional vector retrieval structures.

[0073] Optionally, the step of matching the job content corresponding to the job parsing data in the target job set based on keyword index and vector index to obtain matching data further includes: performing semantic similarity retrieval based on vector index to obtain a first candidate set corresponding to the job query vector; performing keyword matching retrieval based on keyword index to obtain a second candidate set corresponding to the keyword information; performing uniqueness processing on the first and second candidate sets to obtain a unique candidate set; and performing fusion score calculation based on the vector similarity score, keyword matching score, and skill coverage score corresponding to each candidate object in the unique candidate set to generate matching data.

[0074] In this embodiment of the invention, the semantic similarity retrieval can be performed by the personnel and job matching platform based on the vector index structure, which calculates the similarity between the job query vector and the pre-built vectors corresponding to each candidate job in the target job set. For example, the semantic distance between the query vector and the candidate vector can be calculated based on cosine similarity or vector dot product to identify the candidate job that is closest to the job requirements at the semantic level.

[0075] The aforementioned first candidate set can be a set of candidate positions obtained through semantic similarity retrieval, containing several candidate objects that rank highly similar to the job query vector. For example, the aforementioned personnel-job matching platform can select the top N candidate positions with the highest similarity as the first candidate set.

[0076] The aforementioned keyword matching retrieval can be performed by the job matching platform based on a keyword index, matching keyword information in the job parsing data with the keyword fields of the target job set. For example, the job matching platform can perform filtering based on skill terms, industry terms, or region terms to find job content containing at least some key skills or tags.

[0077] The aforementioned second candidate set can be a set of candidate positions obtained through keyword matching retrieval, containing candidates that meet the matching conditions with the keyword information, and can be used to supplement relevant job content that may be missed in semantic retrieval.

[0078] In one possible embodiment, the aforementioned personnel job matching platform can perform unified identification and normalization processing on the first candidate set and the second candidate set to remove duplicate candidates, including but not limited to the following steps: performing case normalization on the candidate job identifiers; deleting duplicates caused by format differences; and performing unique mapping on the candidate data based on unified encoding rules, thereby obtaining a unique candidate set.

[0079] The aforementioned unique candidate set can contain all candidates from semantic retrieval and keyword retrieval without duplicate entries. It will serve as the unified input set for the fusion score calculation and will be used for subsequent sorting and truncation.

[0080] In another possible embodiment, the aforementioned personnel-job matching platform can also perform a fusion score calculation based on the vector similarity score, keyword matching score, and skill coverage score corresponding to the candidate. Specifically, according to preset fusion weights, the vector similarity score, keyword matching score, and skill coverage score are weighted and calculated to obtain the fusion score of the candidate job. For example, the aforementioned personnel-job matching platform can set weight parameters for different dimensions, and dynamically adjust the semantic weight, keyword weight, and skill weight according to the job characteristics to generate a comprehensive score that better meets business needs.

[0081] The above vector similarity score can be used to measure the semantic similarity between job query vectors and candidate job vectors, reflecting the degree of closeness between the two in the semantic space; The keyword matching score mentioned above can be used to measure the degree of matching between candidate positions and keyword information, such as the overall performance of the number of matching keywords, matching weight, and term weight; The skill coverage score mentioned above can be used to measure the coverage between the skills included in the candidate position and the skills required for the position, such as skill hit rate and skill importance coverage ratio.

[0082] More specifically, the specific fusion score can be explained using the following formula: score = a1 * vector score + a2 * keyword score + a3 * skill score; Where a1∈[0.2,0.8], a2∈[0.1,0.6], a3∈[0.1,0.6], it is recommended to satisfy a1+a2+a3=1; Scenario-based weighting: Increase A1 when the job description is clear; increase A3 when the job description emphasizes skill coverage and relevant industry experience; moderately increase A2 when there is noisy text or many synonyms. Threshold explanation: When the score is 60 (basic match): ≥40% of the hard requirements are met, key skills coverage is ≥50%, and industry / regional restrictions are allowed; When the score is 80 (high match): ≥60% of the hard requirements are met, key skills coverage is ≥70%, and relevant experience is available in the last 3–12 months; When the score is 90 (preferred): ≥70% of the hard requirements are met, core skills cover ≥80%, strong relevant industry experience, and the region / salary is highly compatible.

[0083] It should be noted that during the above matching process, online fine-tuning of (a1, a2, a3) can be performed by combining query readability, terminology density, industry priors, and historical backtesting effects. This constrains the results within a given range and maintains a sum of 1 to avoid overfitting. Specifically, the input length threshold is configurable, such as a character limit of 100k–300k. When the recall set exceeds the limit, truncation is performed based on the comprehensive score and the confidence score after deduplication. The output table contains a fixed set of columns (such as ranking / name / comprehensive matching degree / skill matching / experience matching / advantages / items to be verified), and the list uses a fixed numbering and square bracket format for easy front-end parsing and robot push.

[0084] Optionally, the steps of performing semantic analysis on job requirements and candidate objects using a preset language model based on the matching result set, and re-ranking the matching result set according to the analysis results to obtain matching data, further include: calculating a weighted score based on preset fusion weights for vector similarity scores, keyword matching scores, and skill coverage scores; ranking the candidate objects in the unique candidate set based on the fusion score to obtain ranked data; truncating the ranked data based on an output control threshold to obtain a target candidate set; and generating matching data based on the fusion score corresponding to the target candidate set and the corresponding candidate object data.

[0085] In this embodiment of the invention, the aforementioned preset fusion weight can refer to a weight parameter used to control the degree of influence of different scoring dimensions. Generally, corresponding weights are configured for vector similarity score, keyword matching score, and skill coverage score.

[0086] Specifically, based on job type or business needs, the semantic search weight can be set to 0.5, the keyword matching weight to 0.3, and the skill coverage weight to 0.2. It should be noted that these weights can be automatically adjusted after analyzing user needs, and the fusion weights that meet the corresponding needs can be dynamically set. For example, when the job matching platform parses the user's job requirement text and determines that the user values ​​a certain skill, the weight of the skill coverage score can be set higher, so that the sum of the other weights equals the weight of the skill coverage score.

[0087] In this embodiment, multiple score values ​​corresponding to each candidate object in the unique candidate set can be weighted and summed according to the preset fusion weight. The personnel job matching platform will multiply each score value by its corresponding weight factor and accumulate the results to obtain the fusion score of each candidate object.

[0088] The sorted data mentioned above can be an ordered list of candidates obtained by sorting the candidates in the unique candidate set according to the fusion score. Generally, it can be arranged from high to low according to the fusion score, so that the candidate that best meets the job requirements is at the front of the list.

[0089] The aforementioned output control threshold can be a threshold rule used to control the scale of sorted data output. Generally, it can be a fixed number of entries (e.g., returning the first 20 results), a lower limit of the score (e.g., only retaining candidates with a fusion score higher than 0.6), or a dynamic threshold combined with the business scenario.

[0090] For example, when the job parsing data contains a large number of skills, strong industry requirements, or rich semantic elements, the aforementioned job matching platform automatically raises the threshold strategy to retain more high-quality candidates. Specifically, when the score difference between the top few candidates in the fusion score is small (e.g., the difference in the fusion score of the top 10 candidates is less than the preset fluctuation threshold of 0.03), the aforementioned job matching platform will dynamically expand the output range, taking the top 15 or top 20 candidates in the sorted data as the target candidate set, in order to avoid accidentally deleting potentially high-matching items due to scores being too close, thereby achieving adaptive control of the output scale.

[0091] In one possible embodiment, the aforementioned personnel job matching platform can extract a portion of candidate objects that meet the threshold rules from the sorted data based on the output control threshold, and use this to generate the final target candidate set.

[0092] The aforementioned target candidate set can be a truncated set of candidate objects containing the candidate positions or candidates that best meet the job requirements. This target candidate set is characterized by clear ordering, controllable quantity, and consistent structure, and can serve as input for further generating matching data.

[0093] In another possible embodiment, the aforementioned personnel job matching platform forms the final matching data structure for output based on the fusion score of the candidate objects in the target candidate set and the corresponding job content, including but not limited to candidate object identifiers, fusion scores, job feature fields, and other relevant display information.

[0094] By using the above methods and steps, more accurate and stable candidate ranking results can be generated based on multi-dimensional scoring. By controlling the output threshold, the scale and quality of the results can be kept under control, thereby improving the reliability, interpretability and consistency of the structured output of job matching.

[0095] Optionally, in the steps of generating structured matching results based on matching data and performing push and record management on the structured matching results, the method further includes: performing structured processing on the matching data according to a preset format to obtain the target structured matching result; encapsulating the target structured matching result in a preset format and sending it to the target terminal for display or invocation through a message push channel; and updating the corresponding record information so that the target terminal can track and manage the target structured matching result.

[0096] In this embodiment of the invention, the above-mentioned personnel job matching platform can perform structured output and subsequent management of the matching data. Specifically, based on preset fields and fixed headers, the matching data is processed in a structured manner to form a target structured matching result including candidate object identifiers, fusion scores, skill hit status and job characteristics. The generated structured matching results are formatted and encapsulated in the form of JSON, structured table objects, or pre-built visualization components. After encapsulation, the results are sent to the target terminal via a message push channel (enterprise robot interface, system notification interface, or API callback) so that the target terminal can directly display or call them. Simultaneously, the push status, return confirmation, error status, and retry records are recorded, generating corresponding log information to enable the target terminal to track and manage the structured matching results.

[0097] Through the above methods and steps, the matching results are standardized in format, stable in structure, and machine-readable, improving the compatibility of the personnel and job matching platform for invocation and display. Push records and status management improve the reliability of the result output link, making the transmission process of matching results traceable and maintainable.

[0098] like Figure 2 As shown, this embodiment of the invention also provides a personnel job matching device 200, which includes: The first acquisition module 201 is used to acquire job parsing data; The first construction module 202 is used to construct a query index for the job parsing data through a multimodal query model, thereby obtaining a hybrid query index, which includes a keyword index and a vector index. The first matching module 203 is used to match the job content corresponding to the job parsing data in the target job set based on the keyword index and vector index, and obtain a matching result set. The second matching module 204 is used to perform semantic analysis on job requirements and candidate objects based on the matching result set using a preset language model, and to reorder the matching result set according to the analysis results to obtain matching data. The first execution module 205 is used to generate structured matching results based on the matching data, and to perform push and record management on the structured matching results.

[0099] Optionally, the first acquisition module 201 mentioned above includes: The first acquisition submodule is used to acquire at least one job description data from the target resume collection; The second acquisition submodule is used to preprocess at least one of the job description data to obtain clean job description data; The third acquisition submodule is used to extract semantic features from the clean job description data through a natural language model to determine the corresponding feature description elements. The feature description elements include extracted skill elements, qualification elements, regional elements and industry context elements. The fourth acquisition submodule is used to generate job parsing data for the corresponding resume positions based on the extracted skill elements, qualification elements, regional elements and industry context elements.

[0100] Optionally, the first building module 202 mentioned above includes: The first construction submodule is used to perform feature representation calculation on the job parsing data through the multimodal query model to generate the corresponding job query vector; The second construction submodule is used to extract keywords from the job parsing data, obtain keyword information, construct keyword query paths based on the keyword information, and generate corresponding keyword indexes. The third construction submodule is used to perform validity verification on the job query vector and generate the corresponding vector index based on the verification result; The fourth submodule is used to determine the hybrid query index based on the keyword index and the vector index.

[0101] Optionally, the third construction submodule mentioned above includes: The first construction unit is used to determine the verification result of the job query vector by performing dimension consistency and non-zero checks on the job query vector; The second construction unit is used to write the job query vector into a preset vector index structure when the verification result is valid, so as to generate a vector index that can be used for subsequent retrieval processing. The third building unit is used to prevent the generation of the vector index when the verification result is invalid, so that the hybrid query index degenerates into query and matching processing based only on keywords.

[0102] Optionally, the first matching module 203 mentioned above includes: The first matching submodule is used to perform semantic similarity retrieval based on the vector index to obtain a first candidate set corresponding to the job query vector; The second matching submodule is used to perform keyword matching retrieval based on the keyword index to obtain a second candidate set corresponding to the keyword information; The third matching submodule is used to perform uniqueness processing on the first candidate set and the second candidate set to obtain a unique candidate set; The fourth matching submodule is used to perform a fusion score calculation based on the vector similarity score, keyword matching score, and skill coverage score corresponding to each candidate object in the unique candidate set, and generate the matching result set.

[0103] Optionally, the second matching module 204 mentioned above includes: The first matching unit is used to perform a weighted calculation on the vector similarity score, keyword matching score and skill coverage score based on a preset fusion weight to obtain a fusion score. The second matching unit is used to sort the candidate objects in the unique candidate set based on the fusion score to obtain sorted data; The third matching unit is used to truncate the sorted data based on the output control threshold to obtain the target candidate set; The fourth matching unit is used to generate matching data based on the fusion score corresponding to the target candidate set and the corresponding candidate object data.

[0104] Optionally, the first execution module 205 mentioned above includes: The first execution submodule is used to perform structured processing on the matching data according to a preset format to obtain the target structured matching result; The second execution submodule is used to encapsulate the target structured matching result in a preset format and send it to the target terminal for display or invocation through a message push channel, and update the corresponding record information so that the target terminal can track and manage the target structured matching result.

[0105] like Figure 3 As shown, this embodiment of the invention also provides an electronic device 300, including a processor, which can execute any of the above-described personnel job matching methods.

[0106] Specifically, it includes a processor 301 and a memory 302, as well as a computer program for performing a personnel job matching method stored in the memory 302 and capable of running on the processor 301, wherein: The processor 301 executes the calculator program for the personnel job matching method stored in the memory 302, performing the following steps: Obtain job analysis data; By using a multimodal query model, a query index is constructed on the job parsing data to obtain a hybrid query index, which includes a keyword index and a vector index. Based on the keyword index and vector index, the job content corresponding to the job parsing data is matched in the target job set to obtain a set of matching results; Based on the matching result set, semantic analysis of job requirements and candidate objects is performed using a preset language model, and the matching result set is reordered according to the analysis results to obtain matching data; Based on the matching data, structured matching results are generated, and push and record management are performed on the structured matching results.

[0107] Optionally, processor 301 performs the acquisition of job parsing data, including: Extract at least one job description from the target resume set; Preprocess at least one of the job description data to obtain clean job description data; Semantic features are extracted from the clean job description data using a natural language model to determine the corresponding feature description elements, which include extracted skill elements, qualification elements, regional elements, and industry context elements. Based on the extracted skill elements, qualification elements, regional elements, and industry context elements, job analysis data for the corresponding resume positions is generated.

[0108] Optionally, processor 301 executes the multimodal query model to construct a query index for the job parsing data, resulting in a hybrid query index, including: The multimodal query model is used to perform feature representation calculations on the job parsing data to generate corresponding job query vectors. Keyword extraction is performed on the job parsing data to obtain keyword information, and a keyword query path is constructed based on the keyword information to generate a corresponding keyword index; The job query vector is validated, and a corresponding vector index is generated based on the validation result. Based on the keyword index and the vector index, a hybrid query index is determined.

[0109] Optionally, the processor 301 performs the validity check on the job query vector and generates a corresponding vector index based on the check result, including: The verification result of the job query vector is determined by performing dimension consistency and non-zero checks on the job query vector. When the verification result is valid, the job query vector is written into a preset vector index structure to generate a vector index that can be used for subsequent retrieval processing; When the verification result is invalid, the generation of the vector index is prevented, causing the hybrid query index to degenerate into query and matching processing based solely on keywords.

[0110] Optionally, processor 301 executes the matching of job content corresponding to the job parsing data in the target job set based on the keyword index and vector index, to obtain a matching result set, including: Based on the vector index, a semantic similarity retrieval is performed to obtain the first candidate set corresponding to the job query vector; Based on the keyword index, perform keyword matching retrieval to obtain a second candidate set corresponding to the keyword information; The first candidate set and the second candidate set are subjected to uniqueness processing to obtain a unique candidate set; Based on the vector similarity score, keyword matching score, and skill coverage score corresponding to each candidate object in the unique candidate set, a fusion score calculation is performed to generate the matching result set.

[0111] Optionally, the processor 301 executes the semantic analysis of job requirements and candidate objects based on the matching result set using a preset language model, and reorders the matching result set according to the analysis results to obtain matching data, including: Based on the preset fusion weights, the vector similarity score, keyword matching score, and skill coverage score are weighted and calculated to obtain the fusion score. Based on the fusion score, the candidate objects in the unique candidate set are sorted to obtain sorted data; Based on the output control threshold, the sorted data is truncated to obtain the target candidate set; Matching data is generated based on the fusion score and corresponding candidate object data of the target candidate set.

[0112] Optionally, the processor 301 executes the process of generating structured matching results based on the matching data, and performs push and record management on the structured matching results, including: According to a preset format, the matching data is processed in a structured manner to obtain the target structured matching result; The target structured matching results are pre-formatted and packaged, and sent to the target terminal for display or invocation through a message push channel, and the corresponding record information is updated so that the target terminal can track and manage the target structured matching results.

[0113] This invention also provides a personnel job matching system. The personnel job matching system stores a computer program. When the computer program is executed by a processor, it implements the various processes of the personnel job matching method or the application-side personnel job matching method provided in this invention, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0114] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be done by a computer program instructing related hardware, and can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0115] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method of matching people to positions, characterized by, The method comprises the following steps: obtaining job analysis data; querying and indexing the job analysis data through a multi-modal query model to obtain a hybrid query index, wherein the hybrid query index comprises a keyword index and a vector index; based on the keyword index and the vector index, matching the job content corresponding to the job analysis data in a target job set to obtain a matching result set; based on the matching result set, performing semantic analysis on the job requirements and the candidate objects through a pre-set language model, and reordering the matching result set according to the analysis result to obtain matching data; based on the matching data, generating a structured matching result, and performing pushing and record management on the structured matching result.

2. The personnel position matching method of claim 1, wherein, The method comprises the following steps: obtaining at least one job description data from a target resume set; preprocessing at least one job description data to obtain clean job description data; performing semantic feature extraction on the clean job description data through a natural language model to determine corresponding feature description elements, wherein the feature description elements include extracted skill elements, experience elements, regional elements and industry context elements; based on the extracted skill elements, experience elements, regional elements and industry context elements, generating job analysis data corresponding to the resume job.

3. The personnel position matching method of claim 1, wherein, The method comprises the following steps: performing feature expression calculation on the job analysis data through the multi-modal query model to generate corresponding job query vectors; extracting keywords from the job analysis data to obtain keyword information, and constructing a keyword query path according to the keyword information to generate a corresponding keyword index; performing validity verification on the job query vectors, and generating a corresponding vector index according to the verification result; based on the keyword index and the vector index, determining a hybrid query index.

4. The personnel position matching method of claim 3, wherein, The method comprises the following steps: performing dimension consistency and non-zero verification on the job query vectors to determine the verification result of the job query vectors; when the verification result is valid, writing the job query vectors into a pre-set vector index structure to generate a vector index that can be used for subsequent retrieval processing; when the verification result is invalid, preventing the generation of the vector index, so that the hybrid query index is degraded to only perform query and matching processing based on keywords.

5. The personnel position matching method of claim 1, wherein, The method comprises the following steps: performing semantic similarity retrieval based on the vector index to obtain a first candidate set corresponding to the job query vectors; performing keyword matching retrieval based on the keyword index to obtain a second candidate set corresponding to the keyword information; performing unique processing on the first candidate set and the second candidate set to obtain a unique candidate set; Based on the vector similarity score, the keyword matching score and the skill coverage score corresponding to each candidate object in the unique candidate set, a fusion score calculation is performed to generate the matching result set.

6. The personnel job matching method of claim 5, wherein, Based on the matching result set, a semantic analysis of the job requirements and the candidate objects is performed through a preset language model, and the matching result set is reordered according to the analysis result to obtain matching data, including: Based on the preset fusion weight, the vector similarity score, the keyword matching score and the skill coverage score are weighted and calculated to obtain a fusion score; Based on the fusion score, the candidate objects in the unique candidate set are sorted to obtain sorting data; Based on the output control threshold, the sorting data is truncated to obtain a target candidate set; According to the fusion score corresponding to the target candidate set and the corresponding candidate object data, matching data is generated.

7. The personnel position matching method of claim 1, wherein, Based on the matching data, a structured matching result is generated, and the structured matching result is pushed and recorded, including: According to a preset format, the matching data is structured to obtain a target structured matching result; The target structured matching result is packaged in a preset format and sent to a target terminal through a message pushing channel for display or calling, and the corresponding record information is updated, so that the target terminal can track and manage the target structured matching result.

8. A personnel position matching apparatus characterized by comprising: Including: The first acquisition module is configured to acquire job analysis data. The first construction module is configured to construct a mixed query index by querying the job analysis data through a multi-modal query model, wherein the mixed query index includes a keyword index and a vector index. The first matching module is configured to match the job content corresponding to the job analysis data in a target job set based on the keyword index and the vector index to obtain a matching result set. The second matching module is configured to perform semantic analysis of the job requirements and the candidate objects based on the matching result set through a preset language model, and reorder the matching result set according to the analysis result to obtain matching data. The first execution module is configured to generate a structured matching result based on the matching data, and perform pushing and recording management on the structured matching result.

9. An electronic device, comprising: Including: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the personnel job matching method according to any one of claims 1 to 7.

10. A system for matching people to positions, the system comprising: The personnel job matching system stores a computer program, and the computer program is executed by the processor to implement the steps of the personnel job matching method according to any one of claims 1 to 7.

Images