Job personnel knowledge extraction method, device and equipment based on large language model

Abstract

The application relates to a post personnel knowledge extraction method, device and equipment based on a large language model. The method comprises the following steps: defining and formalizing description of an entity set, a relation set and a rule constraint set in a knowledge extraction process, and constructing a prompt word template; on the basis of the prompt word template, inputting an obtained resume data set into a large model, outputting a post personnel domain triple set through a multi-round chain iteration mechanism, and then performing a knowledge alignment strategy; performing entity-relation filtering through a regular expression, establishing an entity linking channel of cross-knowledge bases through an API interface of a Wikidata database, performing knowledge disambiguation, and calculating an edit distance between entities; calculating a similarity score between entities according to the edit distance, determining a knowledge alignment result by using the similarity score, and completing knowledge extraction. The method can improve the intelligent level and accuracy of person-post matching.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus and equipment for extracting job personnel knowledge based on a large language model. Background Technology

[0002] Traditional methods for extracting knowledge about job roles, such as rule-based text matching, dictionary methods, and statistical feature extraction, have achieved some success in specific scenarios. However, they rely heavily on predefined rules, manually constructed vocabularies, and feature templates, resulting in poor generalizability and often proving inadequate when dealing with diverse and non-standard text expressions. Rule-based methods are susceptible to linguistic ambiguity and word order changes in extraction accuracy, making it difficult to cover complex and varied expressions. Dictionary and feature-based methods, on the other hand, suffer from high maintenance costs and weak scalability, and struggle to achieve deep semantic understanding and extraction when processing unstructured corpora such as resumes and job descriptions.

[0003] However, current research on knowledge extraction for job personnel mainly focuses on traditional methods such as rule-based text matching, dictionary methods, and statistical feature extraction. While these methods have achieved certain results in specific scenarios, they lack sufficient flexibility and universality when dealing with diverse and non-standardized texts. Furthermore, they often rely on manually constructed rules and vocabularies, making it difficult to effectively capture deep semantic information. Therefore, research on data-driven knowledge extraction for job personnel remains insufficient. The widespread application of large language models in job personnel knowledge extraction practice still faces several challenges. On the one hand, the model's generalization ability in professional domains needs improvement, and there may be biases in understanding industry-specific terminology and implicit semantics. On the other hand, there is no unified paradigm for the design of prompt words, and the prompting effect highly depends on the designer's experience and task characteristics. Therefore, it is urgent to explore an efficient and universal prompting strategy and domain adaptation mechanism to fully unleash the potential of large language models in job personnel knowledge extraction, improve the intelligence and accuracy of job matching, and provide data support and technical assurance for scientific decision-making in human resource management. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, and equipment for extracting job knowledge based on a large language model to address the above-mentioned technical problems. This method can improve the intelligence and accuracy of job matching and provide data support and technical assurance for scientific decision-making in human resource management.

[0005] A method for extracting job-related knowledge based on a large language model, the method comprising:

[0006] Define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process;

[0007] A prompt word template is constructed based on the entity set, relation set, and rule constraint set; based on the prompt word template, the obtained resume dataset is input into a large model and outputs a set of job domain triples through a multi-round chain iteration mechanism;

[0008] A knowledge alignment strategy is applied to the domain triple set of job personnel. Regular expressions are used to filter entity-relationships. After knowledge disambiguation, a cross-knowledge base entity link channel is established based on the filtered entities and the API interface of the Wikidata database. The standard name of the entity in the Wikidata database is obtained and the edit distance between entities is calculated.

[0009] The similarity score between entities is calculated based on the edit distance, and the knowledge alignment result is determined using the similarity score to complete knowledge extraction.

[0010] A knowledge extraction device for job personnel based on a large language model, the device comprising:

[0011] The multi-round chained iteration module is used to define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process; construct prompt word templates based on the entity set, relation set, and rule constraint set; and input the obtained resume dataset into the large model based on the prompt word templates to output the job domain triple set through the multi-round chained iteration mechanism.

[0012] The knowledge alignment module is used to implement knowledge alignment strategies for the domain triple set of job personnel. It uses regular expressions to filter entity-relationships, establishes cross-knowledge base entity link channels based on the filtered entities and the API interface of the Wikidata database, performs knowledge disambiguation, obtains the standard name of the entity in the Wikidata database, and calculates the edit distance between entities.

[0013] The knowledge correction module is used to calculate the similarity score between entities based on the edit distance, and to use the similarity score to determine the knowledge alignment result to complete knowledge extraction.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0015] Define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process;

[0016] A prompt word template is constructed based on the entity set, relation set, and rule constraint set; based on the prompt word template, the obtained resume dataset is input into a large model and outputs a set of job domain triples through a multi-round chain iteration mechanism;

[0017] A knowledge alignment strategy is applied to the domain triple set of job personnel. Regular expressions are used to filter entity-relationships. After knowledge disambiguation, a cross-knowledge base entity link channel is established based on the filtered entities and the API interface of the Wikidata database. The standard name of the entity in the Wikidata database is obtained and the edit distance between entities is calculated.

[0018] The similarity score between entities is calculated based on the edit distance, and the knowledge alignment result is determined using the similarity score to complete knowledge extraction.

[0019] The aforementioned method, apparatus, and equipment for extracting job personnel knowledge based on a large language model, in this application, establishes a unified domain knowledge framework by clearly defining entities, relations, and rule constraint sets. Through ontology design, the entities, relations, and rules of the job personnel domain are clearly defined, providing a clear target anchor for large model extraction and avoiding bias caused by semantic ambiguity. Based on the entity set, relation set, and rule constraint set, a prompt word template is constructed. The standardized template binds domain knowledge with the extraction task, clarifying the extraction direction. Based on the prompt word template, the acquired resume dataset is input into the large model, and a multi-round chained iterative mechanism is used to output a set of job personnel domain triples. Multiple rounds of iteration accurately capture professional terms and implicit semantics, reduce extraction errors, and construct a more comprehensive candidate knowledge profile. A knowledge alignment strategy is applied to the triple set of job personnel domains. Redundant tags are filtered out using regular expressions, and Wikidata links are used to unify the standard names of entities. Edit distance between entities is calculated, and similarity scores between entities are calculated based on the edit distance. Spelling variants are corrected using similarity scores, which solves the problems of ambiguous entity references and inconsistent names in traditional methods. This forms a standardized knowledge graph, which solves the problem of job personnel knowledge extraction from a data-driven perspective and provides more scientific and reliable data support for decision support in related fields such as job personnel recommendation. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for extracting job-related knowledge based on a large language model in one embodiment.

[0021] Figure 2 This is a schematic diagram illustrating the construction of a prompt word template in one embodiment;

[0022] Figure 3 This is a structural block diagram of a job personnel knowledge extraction device based on a large language model in one embodiment;

[0023] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] In one embodiment, such as Figure 1 As shown, a method for extracting job-related knowledge based on a large language model is provided, including the following steps:

[0026] Step 102: Define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process; construct a prompt word template based on the entity set, relation set, and rule constraint set; based on the prompt word template, input the obtained resume dataset into the large model and output the job personnel domain triple set through a multi-round chain iteration mechanism.

[0027] The main concepts and knowledge scope involved in the extracted knowledge are defined and formally described, forming an entity set, a relation set, and a rule constraint set. The entity set contains subsets composed of various data types, including a job set. personnel gathering Skills Collection and other entity sets; relation sets Includes relationship types such as role, skill, knowledge, education, work experience, awards, and skill certification; rule constraint set. This approach encompasses core constraints for entity relationship verification, including rules for job uniqueness, skill consistency, and work experience duration. The explicit definitions and formal descriptions of these entity sets, relationship sets, and rule constraint sets provide a unified domain language framework for knowledge extraction, completely overcoming the limitations of poor generalization in traditional rule-based methods and high maintenance costs in dictionary-based methods. In traditional methods, manually constructed rules or vocabularies often only cover a single scenario. However, by designing a clearly defined entity set (including job titles, personnel, skills, work experience, etc.), relationship sets (such as personnel-possess-skills, personnel-hold-job titles, etc.), and rule constraint sets, the knowledge system of the job and personnel domain can be systematically organized. This allows the large language model to have clear target anchors during extraction, avoiding extraction bias caused by semantic ambiguity. For example, explicitly requiring skill entities to include skill name + proficiency level accurately distinguishes between "knowing Python" and "being proficient in Python," laying the foundation for precise matching of skill requirements and candidate abilities in subsequent job-person matching.

[0028] The collected publicly available resume data is represented as a dataset. Specifically, the resume data contains the information defined above. Based on the information defined above, a design is created as follows: Figure 2The prompt word template shown is used as a basis for constructing the prompt word template, and then, based on the constructed prompt word template, the dataset is... The knowledge graph of job personnel is obtained by importing it into a large language model and using a multi-round chained iterative mechanism. Specifically, the multi-round chained iterative process of the job personnel knowledge graph is a process of progressively correcting, supplementing, and validating the initial extraction results through the multi-round interaction mechanism of the large language model. We design multi-round prompt word feedback to guide the model to gradually deepen semantic understanding, and optimize subsequent input instructions based on the feedback of previous outputs, ultimately achieving accurate extraction of high-quality triples. The specific implementation steps are as follows:

[0029] (1) First round of selection: based on Figure 2 The designed prompts are used to perform the first round of triple extraction from the resume text. This stage focuses on entity boundary recognition and basic relation classification, performing basic semantic parsing to generate initial structured data. Due to the diversity of expressions in resume texts, such as abbreviations of skill names and non-standardized descriptions of work experience, the first round of results may contain entity type errors or missed relationship judgments.

[0030] (2) Second-round verification: The results of the first round are matched and verified against the seven relation types in the ontology, including: role, skill, knowledge, education, work experience, award, certificate, and entity type. Ontology constraints are injected through prompt words to guide the model to identify errors that do not conform to the ontology rules, such as the award entity lacking a specific award, and a corrected dataset is generated. .

[0031] (3) Three rounds of fine-tuning: Based on the implicit logical relationships in the resume, multiple rounds of follow-up questions are asked (e.g., if someone has skills such as "C language", "embedded development" and "microcontroller debugging", it can be gradually inferred that they have worked as an "embedded engineer", which is consistent with their work experience). The driving model uses contextual semantics to infer the correct information and generate a dataset. .

[0032] (4) Termination Decision: Set an iteration termination threshold to avoid efficiency degradation caused by excessive iteration. The final output is a high-quality set of triples that has undergone multiple rounds of validation. This completes the knowledge filling process from multi-source heterogeneous resumes to knowledge graph triples.

[0033] The prompt word template, built upon entities, relationships, and rules, along with a multi-round chained iteration mechanism, addresses the challenges of large language models lacking a unified paradigm for prompt words and exhibiting weak domain generalization ability. Traditional methods rely on human experience to design rules, while standardized prompt word templates deeply bind domain knowledge to the extraction task, allowing the model to clearly define what to extract and how to extract it without relying on the designer's subjective experience. The multi-round chained iteration mechanism enables progressive processing from shallow extraction to deep completion. For example, after the first round extracts basic skills, the second round can verify the consistency of skill descriptions based on rule constraints, preventing "Python programming" and "Python development" from appearing simultaneously in the same person's resume. Further iterations can then supplement any missing implicit relationships. This mechanism allows large language models to more accurately capture the professional terminology and implicit semantics of job positions, reducing extraction errors caused by domain misunderstanding biases and providing a more comprehensive and accurate candidate knowledge profile for subsequent person-job matching.

[0034] Step 104: Apply a knowledge alignment strategy to the domain triple set of job personnel, use regular expressions to filter entity-relationship pairs, establish cross-knowledge base entity link channels based on the filtered entities and the API interface of the Wikidata database, perform knowledge disambiguation, obtain the standard names of entities in the Wikidata database, and calculate the edit distance between entities.

[0035] After obtaining the knowledge graph triples, issues such as data representation diversity and spelling variations can affect the accuracy and consistency of the knowledge graph. Therefore, a knowledge alignment strategy needs to be implemented on the extracted results. Specifically, the input is the original set of triples. Each triple It usually begins with a serial number. (e.g., Arabic numerals, special symbols, etc.), followed by an "entity-relationship-entity" structure, with the goal of constructing a mapping function. To remove the serial number marker. The core characteristic of the serial number marker is its prefix property, and its regular expression... It needs to cover various formats, such as matching Arabic numerals with separators, Chinese numerals with separators, and special characters, and handle optional spaces by ensuring that only the prefix at the beginning of the string is matched. The filtering process is performed using a matching function. Identify the starting sequence number marker substring If it exists, then execute. The process involves deleting the original triplet if necessary, thus achieving precise filtering of redundant sequence numbers. The extracted entities are then mapped to the Wikidata database, and their standard names in Wikidata are obtained via an API interface. This effectively solves the problems of inconsistent entity names and ambiguous referencing. For example, "Apple Inc.", "Apple Inc.", and "Apple (China) Co., Ltd." are uniformly mapped to standard entity IDs in Wikidata, avoiding matching omissions caused by variations in company names.

[0036] Step 106: Calculate the similarity score between entities based on the edit distance, and use the similarity score to determine the knowledge alignment result to complete knowledge extraction.

[0037] The system calculates the edit distance between entities and then uses this distance to calculate a similarity score. When the similarity score of two entity strings reaches or exceeds a set threshold, they are identified as the same entity. This effectively solves the problem of spelling variations in resume data, ensuring the effectiveness of knowledge extraction and improving the quality of the knowledge graph. By calculating edit distance and similarity scores, spelling variations such as "Data Analyst" and "Data Analyst" are corrected to ensure consistency in skill names. This multi-dimensional knowledge alignment allows resume data from different sources and in different formats to form a standardized knowledge graph. When matching people to jobs, it can directly compare the entities required for the job, their relationships, and the entities that the candidate possesses. For example, if the job requirement is "proficient in Java," it can quickly match candidates in the knowledge graph who "possess - Java - proficiency: proficient." This avoids the "missed qualified candidates" or "misjudged unqualified candidates" caused by "inconsistent names" in traditional matching.

[0038] The above-mentioned multi-dimensional knowledge alignment strategy can effectively integrate and standardize the extracted triplet data, providing a foundation for the construction and application of knowledge graphs for job personnel recommendations.

[0039] In the aforementioned method for extracting job personnel knowledge based on a large language model, this application establishes a unified domain knowledge framework by clearly defining entities, relations, and rule constraint sets. Through ontology design, the entities, relations, and rules in the job personnel domain are clearly defined, providing a clear target anchor for large-scale model extraction and avoiding bias caused by semantic ambiguity. Based on the entity set, relation set, and rule constraint set, a prompt word template is constructed. The standardized template binds domain knowledge to the extraction task, clarifying the extraction direction. Based on the prompt word template, the obtained resume dataset is input into the large model, and a multi-round chained iterative mechanism is used to output a set of job personnel domain triples. Multiple rounds of iteration accurately capture professional terms and implicit semantics, reduce extraction errors, and construct a more comprehensive candidate knowledge profile. A knowledge alignment strategy is applied to the triple set of job personnel domains. Redundant tags are filtered out using regular expressions, and Wikidata links are used to unify the standard names of entities. Edit distance between entities is calculated, and similarity scores between entities are calculated based on the edit distance. Spelling variants are corrected using similarity scores, which solves the problems of ambiguous entity references and inconsistent names in traditional methods. This forms a standardized knowledge graph, which solves the problem of job personnel knowledge extraction from a data-driven perspective and provides more scientific and reliable data support for decision support in related fields such as job personnel recommendation.

[0040] In one embodiment, the entity set comprises a subset consisting of multiple data types, including a job set. personnel gathering Skills Collection and other entity sets; relation sets Includes relationship types such as role, skill, knowledge, education, work experience, awards, and skill certification; rule constraint set. It covers the core constraints for entity relationship verification, including job uniqueness rules, skill consistency rules, and work experience time rules.

[0041] In one embodiment, the prompt word template is a pre-designed standardized text framework that includes a task description, relation categories and definitions, entity categories and definitions, rule constraints, and an output format example. The task description clarifies the task that the large language model needs to complete. The relation categories and definitions define the various types of relations to be extracted, explaining the connotation of each relation. The entity categories and definitions clarify the definitions of various entities to clarify the types of entities to be extracted. The output format example specifies the format of the model output.

[0042] In one embodiment, the acquired resume dataset is input into a large model, and a multi-round chained iterative mechanism is used to output a set of job domain triples, including:

[0043] Based on the prompt word template, entity boundary recognition, basic relationship classification, and basic semantic parsing are performed on the acquired resume dataset to generate initial structured data;

[0044] The initial structured data is matched and validated against the seven relation types and entity types in the ontology library. Ontology constraints are injected through prompt words to guide the model to identify errors that do not conform to the ontology rules and generate a corrected dataset.

[0045] The corrected dataset is subjected to multiple rounds of follow-up questions, which drive the model to infer correct information using contextual semantics. The inferred dataset is generated until a pre-set iteration termination threshold is reached, and the set of domain triples for job personnel that has been verified through multiple rounds is output.

[0046] In one embodiment, entity-relationship filtering is performed using regular expressions, including:

[0047] Input the domain triple set of personnel positions In which each triple Begins with a serial number Followed by entity-relationship-entity structure;

[0048] Construct mapping function The regular expression H is used to match the ordinal marker substring that starts with a triplet. If a match is found Then execute Deletion operation; if no match is found, the original triplet is retained, achieving precise filtering of redundant sequence number markers. This represents the set of triplets obtained after performing the operation.

[0049] In one embodiment, the edit distance between entities is calculated as follows:

[0050]

[0051] in, Representing entities The former One character, Representing entities The former One character.

[0052] In one embodiment, a similarity score between entities is calculated based on the edit distance, including:

[0053] The similarity score between entities is calculated based on the edit distance.

[0054]

[0055] in, , Representing entities Length, Representing entities The length.

[0056] In one embodiment, knowledge extraction is completed by using similarity scores to determine knowledge alignment results, including:

[0057] When the similarity score of two entities reaches or exceeds a set threshold, the two entities are identified as the same entity, and the spelling variants are corrected to complete the knowledge extraction.

[0058] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0059] In one embodiment, such as Figure 3 As shown, a knowledge extraction device for job personnel based on a large language model is provided, including: a multi-round chain iteration module 302, a knowledge alignment module 304, and a knowledge correction module 306, wherein:

[0060] The multi-round chain iteration module 302 is used to define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process; construct prompt word templates based on the entity set, relation set, and rule constraint set; and input the obtained resume dataset into the large model based on the prompt word templates to output the job personnel domain triple set through the multi-round chain iteration mechanism.

[0061] The knowledge alignment module 304 is used to perform knowledge alignment strategies on the domain triple set of job personnel. It uses regular expressions to filter entity-relationships, establishes cross-knowledge base entity link channels based on the filtered entities and the API interface of the Wikidata database, performs knowledge disambiguation, obtains the standard name of the entity in the Wikidata database, and calculates the edit distance between entities.

[0062] The knowledge correction module 306 is used to calculate the similarity score between entities based on the edit distance, and to use the similarity score to determine the knowledge alignment result to complete knowledge extraction.

[0063] Specific limitations regarding the personnel knowledge extraction device based on large language models can be found in the limitations of the personnel knowledge extraction method based on large language models mentioned above, and will not be repeated here. Each module in the aforementioned personnel knowledge extraction device based on large language models can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0064] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for extracting job-specific knowledge based on a large language model. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0065] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0066] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for extracting job-related personnel knowledge based on a large language model, characterized in that, The method includes: Define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process; Based on the entity set, relation set, and rule constraint set, a prompt word template is constructed; based on the prompt word template, the obtained resume dataset is input into a large model and a multi-round chain iteration mechanism is used to output a set of job personnel domain triples; A knowledge alignment strategy is applied to the domain triple set of the personnel in the aforementioned positions. Entity-relationship filtering is performed using regular expressions. After knowledge disambiguation, a cross-knowledge base entity link channel is established based on the filtered entities and the API interface of the Wikidata database. The standard names of the entities in the Wikidata database are then obtained, and the edit distance between entities is calculated. The similarity score between entities is calculated based on the edit distance, and the knowledge alignment result is determined using the similarity score to complete knowledge extraction. The acquired resume dataset is input into a large model, and a multi-round chained iterative mechanism is used to output a set of domain triples for job applicants, including: Based on the prompt word template, entity boundary recognition, basic relationship classification, and basic semantic parsing are performed on the acquired resume dataset to generate initial structured data; The initial structured data is matched and validated against the seven relation types and entity types in the ontology library. Ontology constraints are injected through prompt words to guide the model to identify errors that do not conform to the ontology rules and generate a corrected dataset. The corrected dataset is subjected to multiple rounds of follow-up questions, which drive the model to infer correct information using contextual semantics. The inferred dataset is generated until a pre-set iteration termination threshold is reached, and the set of domain triples for job personnel that has been verified through multiple rounds is output.

2. The method according to claim 1, characterized in that, The entity set contains a subset composed of various data types, including job postings. personnel gathering Skills Collection and other entity sets; the set of relations This includes relationship relationships based on role, skill, knowledge, education, work experience, awards, and skill certificates; the set of rules constrains... It covers the core constraints for entity relationship verification, including job uniqueness rules, skill consistency rules, and work experience time rules.

3. The method according to claim 1, characterized in that, The prompt word template is a pre-designed standardized text framework, including task description, relation categories and definitions, entity categories and definitions, rule constraints, and output format examples. The task description clarifies the task that the large language model needs to complete. The relation categories and definitions define the various types of relations to be extracted, explaining the connotation of each relation. The entity categories and definitions clarify the definitions of various entities to determine the type of entity to be extracted. The output format examples specify the format of the model output.

4. The method according to claim 1, characterized in that, Entity-relationship filtering using regular expressions includes: Input the domain triple set of personnel positions In which each triple Begins with a serial number Followed by entity-relationship-entity structure; Construct mapping function The regular expression H is used to match the ordinal marker substring that starts with a triplet. If a match is found Then execute Deletion operation; if no match is found, the original triplet is retained, achieving precise filtering of redundant sequence number markers. This represents the set of triplets obtained after performing the operation.

5. The method according to claim 1, characterized in that, The edit distance between entities is calculated as follows: in, Representing entities The former One character, Representing entities The former One character.

6. The method according to claim 1, characterized in that, Calculate similarity scores between entities based on the edit distance, including: The similarity score between entities is calculated based on the edit distance. in, This indicates the calculation of the edit distance between entities A and B. Representing entities Length, Representing entities The length.

7. The method according to claim 1, characterized in that, Knowledge extraction is completed by using the similarity score to determine the knowledge alignment result, including: When the similarity score of two entities reaches or exceeds a set threshold, the two entities are identified as the same entity, and the spelling variants are corrected to complete the knowledge extraction.

8. A device for extracting job-related knowledge based on a large language model, characterized in that, The device includes: A multi-round chained iteration module is used to define and formally describe the entity set, relation set, and rule constraint set in the knowledge extraction process; it constructs a prompt word template based on the entity set, relation set, and rule constraint set; based on the prompt word template, the acquired resume dataset is input into the large model and outputs a set of job domain triples through a multi-round chained iteration mechanism, including: Based on the prompt word template, entity boundary recognition, basic relationship classification, and basic semantic parsing are performed on the acquired resume dataset to generate initial structured data; The initial structured data is matched and validated against the seven relation types and entity types in the ontology library. Ontology constraints are injected through prompt words to guide the model to identify errors that do not conform to the ontology rules and generate a corrected dataset. The corrected dataset is subjected to multiple rounds of follow-up questions and the model is driven to infer the correct information using contextual semantics. The inferred dataset is generated until a pre-set iteration termination threshold is reached, and the set of domain triples for job personnel that has been verified through multiple rounds is output. The knowledge alignment module is used to perform knowledge alignment strategies on the domain triple set of the personnel in the position. It uses regular expressions to filter entity-relationships, establishes cross-knowledge base entity link channels based on the filtered entities and the API interface of the Wikidata database, performs knowledge disambiguation, obtains the standard name of the entity in the Wikidata database, and calculates the edit distance between entities. The knowledge correction module is used to calculate the similarity score between entities based on the edit distance, and to use the similarity score to determine the knowledge alignment result to complete knowledge extraction.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images