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

Abstract

The invention 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 formally describing an entity set, a relationship set and a rule constraint set in a knowledge extraction process to construct a cue word template; on the basis of a cue word template, an obtained resume data set is input into a large model, a post personnel field triple set is output through a multi-round chain type iteration mechanism, and then a knowledge alignment strategy is carried out; performing entity-relationship filtering by using a regular expression, establishing a cross-knowledge-base entity link channel with an API (Application Program Interface) of the Wikdata database, performing knowledge disambiguation, and calculating an editing distance between entities; and calculating a similarity score between the entities according to the editing distance, and determining a knowledge alignment result by using the similarity score to complete knowledge extraction. By adopting the method, the intelligent level and accuracy of man-post matching can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a post personnel knowledge extraction method, device and equipment based on a large language model. BACKGROUND

[0002] Traditional post personnel knowledge extraction methods, such as rule-based text matching, dictionary methods and statistical feature extraction, have achieved certain results in specific scenarios, but overall rely on predefined rules, manually constructed word tables and feature templates, have poor universality, and often lack the ability to deal with diversified and non-standardized texts. Rule-based methods are easily affected by language ambiguity and word order changes, and are difficult to cover complex and variable expression methods; dictionary and feature methods face high maintenance costs and weak scalability, and are difficult to achieve deep understanding and extraction at the semantic level when dealing with unstructured data such as resumes and job descriptions.

[0003] However, current research on post personnel knowledge extraction mainly focuses on traditional methods such as rule-based text matching, dictionary methods and statistical feature extraction. These methods have achieved certain results in specific scenarios, but lack sufficient flexibility and universality when dealing with diversified and non-standardized texts, and often rely on manually constructed rules and word tables, making it difficult to effectively capture deep semantic information. Therefore, there is still a lack of research on data-driven post personnel information knowledge extraction. The current widespread application of large language models in post personnel knowledge extraction practice still faces several challenges. On the one hand, the model's generalization ability in specialized fields still needs to be strengthened, and there may be biases in understanding industry-specific terminology and implicit semantics; on the other hand, there is no unified paradigm for prompt word engineering, and the effectiveness of prompts is highly dependent on the designer's experience and task characteristics. Therefore, it is necessary to explore an efficient and universal prompt strategy and domain adaptation mechanism to fully realize the potential of large language models in post personnel knowledge extraction and improve the intelligent level and accuracy of human resource management, providing data support and technical support for scientific decision-making in human resources management. SUMMARY

[0004] Therefore, it is necessary to provide a post personnel knowledge extraction method, device and equipment based on a large language model to improve the intelligent level and accuracy of human resource management, and to provide data support and technical support for scientific decision-making in human resources management.

[0005] A post personnel knowledge extraction method based on a large language model, the method comprising: defining and formalizing the entity set, the relationship set and the rule constraint set in the knowledge extraction process; construct a prompt word template according to the entity set, the relation set and the rule constraint set; input the obtained resume data set into the large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; perform knowledge alignment on the post personnel domain triple set, filter the entity-relation through a regular expression, establish an entity linking channel across knowledge bases according to the filtered entity and an API interface of a Wikidata database to perform knowledge disambiguation, obtain a standard name of the entity in the Wikidata database and calculate an edit distance between the entities; calculate a similarity score between the entities according to the edit distance, and determine the knowledge alignment result to complete knowledge extraction by using the similarity score.

[0006] A post personnel knowledge extraction device based on a large language model, the device comprising: a multi-round chain iteration module configured to define and formally describe an entity set, a relation set and a rule constraint set in a knowledge extraction process, construct a prompt word template according to the entity set, the relation set and the rule constraint set, and input an obtained resume data set into a large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; a knowledge alignment module configured to perform knowledge alignment on the post personnel domain triple set, filter the entity-relation through a regular expression, establish an entity linking channel across knowledge bases according to the filtered entity and an API interface of a Wikidata database to perform knowledge disambiguation, obtain a standard name of the entity in the Wikidata database and calculate an edit distance between the entities; a knowledge correction module configured to calculate a similarity score between the entities according to the edit distance, and determine the knowledge alignment result to complete knowledge extraction by using the similarity score.

[0007] A computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program: define and formally describe an entity set, a relation set and a rule constraint set in a knowledge extraction process; construct a prompt word template according to the entity set, the relation set and the rule constraint set; input the obtained resume data set into the large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; The knowledge alignment strategy is performed on the post personnel field triple set, the regular expression is used for entity-relation filtering, the entity link channel is established between the knowledge bases according to the API interface of the filtered entity and the Wikidata database to perform knowledge disambiguation, the standard name of the entity in the Wikidata database is obtained, and the edit distance between entities is calculated; The similarity score between entities is calculated according to the edit distance, and the knowledge alignment result is determined by using the similarity score to complete knowledge extraction.

[0008] The above-mentioned post personnel knowledge extraction method, device and equipment based on a large language model, the present application defines the entity, relation and rule constraint set, builds a unified domain knowledge framework, clearly defines the entity, relation and rule of the post personnel field for the large model extraction, avoids the deviation caused by semantic ambiguity, constructs the prompt word template according to the entity set, the relation set and the rule constraint set, the standardized template binds the domain knowledge and the extraction task, and the extraction direction is clear; on the basis of the prompt word template, the obtained resume data set is input into the large model to output the post personnel field triple set through a multi-round chain iteration mechanism, the professional terms and implicit semantics are accurately captured through multi-round iteration, the extraction error is reduced, and a more comprehensive candidate knowledge portrait is constructed. The knowledge alignment strategy is performed on the post personnel field triple set, the regular expression is used for filtering and stripping redundant marks, the Wikidata link is used to unify the entity standard name, and the edit distance between entities is calculated; the similarity score between entities is calculated according to the edit distance, the spelling variants are corrected by using the similarity score, the problems of entity reference ambiguity and non-uniform name in the traditional method are solved, a standardized knowledge graph is formed, and the problem of post personnel knowledge extraction in the data-driven perspective is solved, which provides more scientific and reliable data support for decision support in post personnel recommendation and other related fields. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 A flowchart of a post personnel knowledge extraction method based on a large language model in one embodiment is shown in the figure; Figure 2 A schematic diagram of constructing a prompt word template in one embodiment is shown in the figure; Figure 3 A structural block diagram of a post personnel knowledge extraction device based on a large language model in one embodiment is shown in the figure; Figure 4 An internal structure diagram of a computer device in one embodiment is shown in the figure. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0011] In one embodiment, as shown in Figure 1 , a large language model-based post personnel knowledge extraction method is provided, comprising the following steps: Step 102, defining and formally describing the entity set, relationship set and rule constraint set in the knowledge extraction process; constructing a prompt word template according to the entity set, relationship set and rule constraint set; inputting the obtained resume data set into the large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism.

[0012] The main concepts and knowledge range involved in the extracted knowledge are defined and formally described to form an entity set, a relationship set and a rule constraint set. The entity set contains subsets of various data types, including a post set , a personnel set , a skill set and other entity sets; the relationship set contains role relationship, skill relationship, knowledge relationship, education relationship, work experience relationship, award relationship and skill certificate relationship; the rule constraint set covers the core constraint conditions of entity relationship verification, including post uniqueness rule, skill consistency rule and work experience time rule. The above-mentioned definite definition and formal description of the entity set, relationship set and rule constraint set provide a unified domain language framework for knowledge extraction, and completely break through the limitations of poor generalization of traditional rule method and high maintenance cost of dictionary method. In the traditional method, the rules or word tables constructed manually can only cover a single scene, while through the entity set, including post, personnel, skill, work experience, etc., the relationship set, such as personnel-having-skill and personnel-holding-post, and the rule constraint set, the knowledge system of post personnel domain can be systematically sorted out, so that the large language model has a clear target anchor point when extracting, avoiding the extraction deviation caused by semantic ambiguity, for example, the skill entity needs to contain skill name + proficiency level, which can accurately distinguish the difference between mastering Python and being proficient in Python, which lays a foundation for accurately matching skill demand and candidate ability in subsequent person-post matching.

[0013] The collected public resume data is represented as a data set , specifically, the resume data contains the information defined above. According to the information defined above, a prompt word template is designed as Figure 2The prompt word template is shown, and then, on the basis of the constructed prompt word template, the data set The large language model is imported and the post personnel knowledge graph is obtained through a multi-round chain iteration mechanism. Specifically, the multi-round chain iteration of the post personnel knowledge graph is a process of gradually correcting, supplementing and verifying the initial extraction results through the multi-round interaction mechanism of the large language model. We gradually deepen the semantic understanding by designing a multi-round prompt word feedback guide model, and optimize the subsequent input instructions based on the feedback of the previous output, and finally realize the accurate extraction of high-quality triples. The specific implementation steps are as follows: (1) First extraction: based on the prompt word designed by the present application, the first triple extraction is performed on the resume text. This stage focuses on entity boundary identification and basic relationship classification, and performs basic semantic analysis to generate initial structured data Figure 2 . Due to the diversity of expression in the resume text, such as the abbreviation of skill name and the non-standardized description of work experience, the first round result may have entity type errors or relationship omissions.

[0014] (2) Second verification: the first round result is matched and verified with the 7 relationship types in the ontology library, including: role, skill, knowledge, education, work experience, award, certificate and entity type. By injecting the ontology constraint through the prompt word, the model is guided to identify errors that do not conform to the ontology rules, such as the lack of specific awards in the award entity, to generate the corrected data set .

[0015] (3) Third fine-tuning: for the logical relationships implied in the resume, through multiple rounds of questioning (such as someone mastering "C language", "embedded development" and "single-chip debugging" skills, which can gradually infer that he has worked as an "embedded engineer", which is consistent with his work experience), the model uses the context semantic inference to drive the model to infer the correct information, and generates the data set .

[0016] (4) Termination determination: set the iteration termination threshold to avoid the efficiency decline caused by excessive iteration. Finally, output the high-quality triple set after multi-round verification , complete the knowledge filling from multi-source heterogeneous resumes to knowledge graph triples.

[0017] ​The prompt word template based on entities, relations and rules, and the multi-round chain iteration mechanism solve the challenges of no unified paradigm and weak domain generalization ability of large language model prompt words. Traditional methods rely on artificial experience to design rules, while standardized prompt word templates deeply bind domain knowledge and extraction tasks, allowing the model to clearly extract and extract without relying on the subjective experience of the designer; The multi-round chain iteration mechanism can realize progressive processing from shallow extraction to deep completion, such as extracting basic skills of personnel in the first round, and then checking whether the skill description is consistent based on rule constraints in the next round to avoid “Python programming” and “Python development” appearing in the same person's resume at the same time, and then supplementing the missing implicit relationships through iteration. This mechanism allows large language models to more accurately capture professional terminology and implicit semantics in the domain of the post personnel, reducing extraction errors caused by domain understanding bias, and providing more comprehensive and accurate candidate knowledge portraits for subsequent person-job matching.

[0018] In step 104, the knowledge alignment strategy is performed on the set of post personnel domain triples. Regular expressions are used to filter entities and relationships, and after knowledge disambiguation, the standard name of the entity in the Wikidata database is obtained and the edit distance between entities is calculated.

[0019] After obtaining the knowledge graph triples, due to the existence of data diversity, spelling variants and other problems, the accuracy and consistency of the knowledge graph will be affected, so it is necessary to perform a knowledge alignment strategy on the extraction results. Specifically, the original triple set is input, where each triple is usually marked at the beginning with a serial number (such as Arabic numerals, special symbols, etc.), followed by an “entity-relation-entity” structure, and the goal is to construct a mapping function to strip the serial number mark. The core feature of the serial number mark is prefix, and its regular expression needs to cover multiple forms, such as matching Arabic numerals with separators, Chinese numerals with separators, and special symbols, and by ensuring that only the prefix at the beginning of the string is matched, optional spaces are handled. The filtering process identifies the starting serial number mark substring through the match function , and if it exists, it executes The original triple is retained, thereby achieving accurate filtering of the redundant sequence number label. After that, the extracted entity is mapped to the Wikidata database, and the standard name of the entity in Wikidata is obtained through the API interface, which can effectively solve the problem of non-uniform entity name and ambiguous reference. For example, “Apple Inc.”, “Apple (China) Co., Ltd.” are unified to the standard entity ID in Wikidata, avoiding the matching omission caused by the variation of enterprise name.

[0020] In step 106, the similarity score between entities is calculated according to the edit distance, and the knowledge alignment result is determined by using the similarity score to complete the knowledge extraction.

[0021] The edit distance between entities is calculated, and the similarity score between entities is calculated according to the edit distance. When the similarity score of two entity strings reaches or exceeds the set threshold, they are identified as the same entity, thereby effectively solving the spelling variation problem in the resume data, ensuring the effectiveness of knowledge extraction, and improving the quality of the knowledge graph. By calculating the edit distance and the similarity score, spelling variations such as “Data Analysist” and “Data Analyst” are corrected to ensure the consistency of skill names. This multi-dimensional knowledge alignment allows different sources and formats of resume data to form a standardized knowledge graph, and the precise comparison of entity - relationship - candidate entity based on job requirements can be directly used for person - job matching, for example, the requirement of “Java proficiency” can quickly match the candidate with “possess - Java - proficiency level: proficient” in the knowledge graph, avoiding the “qualified candidate screening” or “unqualified candidate misjudgment” caused by “non-uniform name” in traditional matching.

[0022] Through the above multi-dimensional knowledge alignment strategy, the extracted triple data can be effectively integrated and standardized, providing a basis for the construction and application of the knowledge graph for job personnel recommendation.

[0023] In the post personnel knowledge extraction method based on the large language model, the application builds a unified domain knowledge framework through the explicit definition of entity, relationship and rule constraint set. Through ontology design, the entity, relationship and rule of the post personnel domain are clearly defined to provide a clear target anchor for large model extraction, avoid bias caused by semantic ambiguity, and construct a prompt word template according to the entity set, relationship set and rule constraint set. The standardized template binds the domain knowledge and the extraction task, and clearly defines the extraction direction. On the basis of the prompt word template, the obtained resume data set is input into the large model to output the post personnel domain triple set through a multi-round chain iteration mechanism. The multi-round iteration accurately captures professional terms and implicit semantics, reduces extraction errors, and constructs a more comprehensive candidate knowledge portrait. The knowledge alignment strategy is performed on the post personnel domain triple set, the regular expression is used to filter and strip redundant marks, the Wikidata link is used to unify the entity standard name, and the edit distance between entities is calculated. According to the edit distance, the similarity score between entities is calculated, the spelling variants are corrected using the similarity score, the entity reference ambiguity and name inconsistency problems in the traditional method are solved, a standardized knowledge graph is formed, and the problem of post personnel knowledge extraction in the data-driven perspective is solved, providing more scientific and reliable data support for decision support in post personnel recommendation and other related fields.

[0024] In one embodiment, the entity set includes a subset of multiple data types, including a post set , a personnel set , a skill set , and other entity sets; the relationship set includes role relationship, skill relationship, knowledge relationship, education relationship, work experience relationship, award relationship, and skill certificate relationship; the rule constraint set covers the core constraint conditions of entity relationship verification, including post uniqueness rule, skill consistency rule, and work experience time rule.

[0025] In one embodiment, the prompt word template is a pre-designed standardized text framework, including task description, relationship category and definition, entity category and definition, rule constraint, and output format example; the task description indicates the task that the large language model needs to complete; the relationship category and definition define each type of relationship to be extracted, and explain the connotation of each type of relationship; the entity category and definition define each type of entity to clearly define the entity type to be extracted; and the output format example specifies the format of the model output.

[0026] In one embodiment, the obtained resume data set is input into the large model to output the post personnel domain triple set through a multi-round chain iteration mechanism, including: Based on the prompt word template, the obtained resume dataset is subjected to entity boundary recognition, basic relationship classification and basic semantic analysis, and initial structured data is generated; The initial structured data is matched and verified with the 7 relationship types and entity types of the ontology library, the prompt word is injected into the ontology constraint to guide the model to identify errors that do not conform to the ontology rules, and a corrected dataset is generated; The corrected dataset is subjected to multiple rounds of questioning and the model is driven to infer correct information using context semantics, and an inferred dataset is generated until a pre-set iteration termination threshold is reached, and a post-multi-round verification post personnel domain triple set is output.

[0027] In one of the embodiments, regular expressions are used for entity-relation filtering, including: Input post personnel domain triple set , wherein each triple is marked at the beginning with a serial number , followed by an entity-relation-entity structure; Construct a mapping function to match the serial number mark substring at the beginning of the triple by regular expression H , if matched , perform deletion operation; if not matched, keep the original triple to achieve accurate filtering of redundant serial number marks, denotes the triple set obtained after the operation.

[0028] In one of the embodiments, the edit distance between entities is calculated as

[0029] wherein denotes the first characters of entity , and denotes the first characters of entity .

[0030] In one of the embodiments, the similarity score between entities is calculated according to the edit distance, including: The similarity score between entities calculated according to the edit distance is

[0031] wherein , denotes the length of entity , and denotes the length of entity .

[0032] In one of the embodiments, the knowledge extraction is completed by determining the knowledge alignment result using the similarity score, including: When the similarity score of two entities reaches or exceeds the set threshold, the two entities are identified as the same entity, and the spelling variants are corrected to complete the knowledge extraction.

[0033] It should be understood that, although Figure 1 The steps in the flowchart of the method are displayed in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps in the method can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or sub-steps or stages of other steps.

[0034] In one embodiment, as shown in Figure 3 A large language model-based post personnel knowledge extraction device is provided, including: a multi-round chain iteration module 302, a knowledge alignment module 304, and a knowledge correction module 306, wherein: The multi-round chain iteration module 302 is configured to define and formally describe the entity set, the relationship set, and the rule constraint set in the knowledge extraction process; construct a prompt word template according to the entity set, the relationship set, and the rule constraint set; and input the obtained resume dataset into the large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; The knowledge alignment module 304 is configured to perform a knowledge alignment strategy on the post personnel domain triple set, filter entity-relationships using a regular expression, establish an entity linking channel across knowledge bases according to the filtered entities and an API interface of a Wikidata database to perform knowledge disambiguation, obtain the standard name of the entity in the Wikidata database, and calculate the edit distance between entities; The knowledge correction module 306 is configured to calculate the similarity score between entities according to the edit distance, and complete knowledge extraction by determining the knowledge alignment result using the similarity score.

[0035] The specific limitations of the post personnel knowledge extraction device based on a large language model can refer to the limitations of the post personnel knowledge extraction method based on a large language model in the foregoing, and will not be repeated here. Each module in the post personnel knowledge extraction device based on a large language model can be realized by software, hardware, and combinations thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0036] In one embodiment, a computer device, which can be a terminal, has an internal structure diagram as shown in Figure 4 The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a post personnel knowledge extraction method based on a large language model. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad provided on the computer device shell, or an external keyboard, touchpad, or mouse, etc.

[0037] Those skilled in the art can understand that Figure 4 The structure shown in the foregoing merely is a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the diagram, or combine certain components, or have a different component arrangement.

[0038] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0039] The technical features of the above embodiments can be combined in any way. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0040] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for extracting post personnel knowledge based on a large language model, characterized in that, The method comprises: defining and formally describing the entity set, the relation set and the rule constraint set in the knowledge extraction process; constructing a prompt word template according to the entity set, the relation set and the rule constraint set; inputting the obtained resume data set into a large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; performing a knowledge alignment strategy on the post personnel domain triple set, filtering entities and relations by using a regular expression, establishing an entity linking channel across knowledge bases through an API interface of a Wikidata database after knowledge disambiguation according to the filtered entities, obtaining standard names of the entities in the Wikidata database and calculating an edit distance between the entities; calculating a similarity score between the entities according to the edit distance, and determining a knowledge alignment result to complete knowledge extraction by using the similarity score.

2. The method of claim 1, wherein, The entity set contains subsets of various data types, including a post set , a personnel set , a skill set , and other entity sets; the relationship set contains role relationships, skill relationships, knowledge relationships, education relationships, work experience relationships, award relationships, and skill certificate relationships; the rule constraint set covers the core constraint conditions for entity relationship verification, including post uniqueness rules, skill consistency rules, and work experience time rules.

3. The method of claim 1, wherein, The prompt word template is a pre-designed standardized text framework, which comprises a task description, relation categories and definitions, entity categories and definitions, rule constraints and an output format example; the task description indicates the task to be completed by the large language model; the relation categories and definitions define various types of relations to be extracted and explain the connotation of each type of relation; the entity categories and definitions define various types of entities to be extracted; and the output format example specifies the format of the model output.

4. The method according to any one of claims 1 to 3, characterized in that, The obtained resume data set is input into a large model to output a post personnel domain triple set through a multi-round chain iteration mechanism, which comprises: based on the prompt word template, performing entity boundary recognition, basic relation classification and basic semantic analysis on the obtained resume data set to generate initial structured data; matching and verifying the initial structured data with 7 relation types and entity types of an ontology library, injecting prompt words into ontology constraints to guide the model to identify and correct errors that do not conform to the ontology rules to generate a corrected data set; driving the model to infer correct information by using context semantics through multiple rounds of follow-up questions based on the corrected data set to generate an inferred data set until a pre-set iteration termination threshold is reached, and outputting a post personnel domain triple set that has passed multiple rounds of verification.

5. The method of claim 1, wherein, The filtering of entities and relations by using a regular expression comprises: input job personnel domain triple set wherein each triple is marked at the beginning with a sequence number and is followed by an entity-relation-entity structure; Constructing mapping function , match the number mark substring at the beginning of the triple by regular expression H , if match , execute deletion operation; if not match, keep the original triple to realize accurate filtering of redundant number marks, denotes the triple set obtained after executing the operation.

6. The method of claim 1, wherein, calculating the edit distance between the entities is wherein represents the first characters of the entity represents the first characters of the entity represents the first characters of the entity 7. The method of claim 1, wherein, calculating the similarity score between the entities according to the edit distance comprises: calculating the similarity score between the entities according to the edit distance is wherein, denotes the edit distance between entities A and B, denotes the length of entity , denotes the length of entity .

8. The method of claim 1, wherein, determining the knowledge alignment result to complete knowledge extraction by using the similarity score comprises: when the similarity score of two entities reaches or exceeds a set threshold, the two entities are identified as the same entity, and spelling variants are corrected to complete knowledge extraction. 9.A post personnel knowledge extraction device based on a large language model, characterized in that, The device comprises: a multi-round chain iteration module for defining and formally describing the entity set, the relation set and the rule constraint set in the knowledge extraction process; constructing a prompt word template according to the entity set, the relation set and the rule constraint set; inputting the obtained resume data set into a large model based on the prompt word template to output a post personnel domain triple set through a multi-round chain iteration mechanism; The knowledge alignment module is used for a knowledge alignment strategy on the post personnel domain triple set, filters entity-relation by using a regular expression, establishes an entity linking channel across knowledge bases according to an API interface of the filtered entity and a Wikidata database to perform knowledge disambiguation, obtains a standard name of the entity in the Wikidata database, and calculates an edit distance between entities. The knowledge correction module is used for calculating a similarity score between entities according to the edit distance, and determining a knowledge alignment result to complete knowledge extraction by using the similarity score. 10.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-9. The processor implements the steps of the method in any one of claims 1 to 8 when executing the computer program.

Images