A method for intelligently adapting university course content to post skills based on a heterogeneous knowledge graph

Abstract

This invention relates to the field of computer information processing technology, specifically to a method for intelligently matching university course content with job skills based on heterogeneous knowledge graphs. The method includes the following steps: S1, acquiring job recruitment data and course syllabus data, denoising the job data, extracting skills, and semantically standardizing it, and parsing the syllabus to form BIO sequence labeled data; S2, constructing an industry-side demand knowledge graph and an education-side supply knowledge graph respectively, and storing them in a graph database to form a dual-tower structure; S3, semantically representing the knowledge points on both sides and mapping them to a unified vector space to determine the cross-domain knowledge point correspondences; S4, calculating the job skill coverage rate based on the correspondences and generating course optimization or supplementation schemes. This invention achieves accurate, quantitative, and dynamic matching of university courses with job skills through automatic multi-source data collection, a dual-tower heterogeneous knowledge graph, and a semantic alignment optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, and in particular to a method for intelligently adapting university course content and job skills based on heterogeneous knowledge graphs. Background Technology

[0002] With the development of big data and artificial intelligence technologies, the education sector and the job market have accumulated a large amount of textual data related to curriculum design, teaching content, job requirements, and skill requirements. University curriculum systems typically include various structured and unstructured information such as syllabi, course descriptions, and knowledge point structures, while corporate recruitment processes generate a large amount of textual information on job descriptions, skill requirements, and employment conditions. Achieving an effective match between university course content and job skill requirements is of great significance for optimizing talent training programs, improving graduate employment suitability, and assisting employers in accurately selecting talent.

[0003] Existing course-job matching technologies largely rely on keyword matching or similarity calculation methods based on statistical features and traditional machine learning. They typically assess the relevance between texts through word frequency overlap, vector distance, or pre-defined rules. However, in practical applications, there are significant differences in expression and focus between university teaching contexts and corporate recruitment contexts. This makes it difficult to accurately identify many semantically related but differently expressed concepts, leading to frequent missed or incorrect matching results. Furthermore, while some technologies introduce knowledge graphs to enhance semantic understanding, their utilization of knowledge graphs often remains at the level of local ternary relationships, primarily serving as a supplement to the textual context. This makes it difficult to characterize multi-hop associations and overall structural relationships between various entities such as courses, knowledge points, skills, and job positions, and thus fails to meet the refined adaptation needs of cross-domain, long-text, and heterogeneous entity scenarios. Summary of the Invention

[0004] This invention provides an intelligent adaptation method for university course content and job skills based on heterogeneous knowledge graphs. Addressing the multi-source heterogeneity and semantic gap between university teaching data and enterprise job demand data at the source, structure, and semantic levels, this method constructs a heterogeneous knowledge graph that integrates multiple types of entities such as courses, chapters, knowledge points, skills, and job positions. By mapping graph nodes to a unified semantic representation space, it achieves automated, fine-grained semantic alignment between course content and job skills. This provides quantifiable, data-driven technical support for optimizing university curriculum systems and talent cultivation decisions, overcoming the shortcomings of existing technologies such as low matching accuracy, insufficient structural utilization, and poor adaptability.

[0005] A method for intelligently matching university course content with job skills based on heterogeneous knowledge graphs includes the following steps: S1. Obtain job recruitment data from the industry side and course syllabus data from the education side. Denoise and extract skill information from the job recruitment data, and perform semantic standardization on the job skill descriptions. Parse the course syllabus data, reorganize the unstructured text into text blocks with hierarchical relationships, and convert it into BIO sequence labeling format data. S2. Construct an industry-side demand knowledge graph and an education-side supply knowledge graph respectively. The industry-side demand knowledge graph is used to describe the relationship between job positions, skills, knowledge points and courses, while the education-side supply knowledge graph is used to describe the hierarchical relationship between courses, chapters and knowledge points. The industry-side demand knowledge graph and the education-side supply knowledge graph are stored in a graph database to form a dual-tower structure. S3 performs semantic representation of knowledge points in the industry-side demand knowledge graph and knowledge points in the education-side supply knowledge graph, and maps them to a unified vector space. It then determines the correspondence between cross-domain knowledge points by calculating semantic similarity and combining rule matching. S4, based on the correspondence between cross-domain knowledge points, calculates the coverage of the existing curriculum system with the knowledge points required for the target position, obtains the job skills coverage rate, and generates curriculum optimization or supplementary plans to improve the job skills coverage rate based on the uncovered knowledge points.

[0006] Optionally, S1 includes: S11, by calling the DrissionPage automation framework, simulates browser behavior to access the recruitment platform and retrieves the corresponding original job posting HTML text data for the target position. S12, perform regular expression cleaning on the obtained raw job recruitment HTML text data to remove HTML tags and redundant text information unrelated to skills, and obtain plain text job description content. S13. For non-standard or mixed expression skill texts in plain text job descriptions, input the skill texts into the Large Language Model (LLM), perform semantic normalization processing, output unified standard skill entities and label their types; S14: Read the PDF byte stream of the college course syllabus, construct a parsing state machine based on regular expression rules (including chapter title matching) and keywords, identify different semantic regions in the syllabus, and reorganize the unstructured text into hierarchical text blocks that include course information and knowledge points. S15 converts the reorganized hierarchical text blocks into BIO (Begin-Inside-Outside) sequence label format data.

[0007] Optionally, S12 includes: S121, Perform tag matching and stripping operations on the original job posting HTML text data. By deleting character substrings that match the HTML tag pattern, obtain the text content excluding HTML tags. ; S122, For the text content excluding HTML tags, based on a preset set of non-skill-related stop words, perform redundant text matching and deletion operations to obtain the job description text after filtering out non-skill-related redundant text. ; S123 will filter out redundant non-skill text from the job description text. As a plain text job description Output.

[0008] Optionally, S13 includes: S131, from plain text job description content Extract the set of candidate fragments belonging to the skill description ,in, For the first A number of candidate skill fragments, The number of candidate segments; S132, each candidate skill fragment Input a large language model and obtain its corresponding standard skill entity name. This enables semantic normalization of non-standard or mixed expressions; S133, for each standard skill entity name Assign entity type labels and output standardized results for As a result of job-side skills standardization.

[0009] Optionally, S14 includes: S141, Read the PDF byte stream of university course syllabi. And parse it to obtain the text line sequence arranged in the original order. ,in, For the first Line text content, Total number of lines in the text; S142, construct a parsing state machine based on chapter title regular expression matching and keyword triggering rules, and process each line of text content. Generate semantic state This allows for the identification of different semantic regions, such as chapters and reference books. S143, based on semantic state sequence Adjacent lines of text with the same semantic state or that satisfy a hierarchical nesting relationship are aggregated into a hierarchical text block set. Each level of text block includes course information and content related to knowledge points.

[0010] Optionally, S2 includes: S21 takes job skill entities as input, performs semantic reasoning and decomposition on job skills based on a large language model, maps skills to corresponding course names and refines them into multiple knowledge points, and builds the relationship between job positions, skills, knowledge points and courses in a graph database to form an industry demand knowledge graph. S22 performs semantic encoding and entity recognition on BIO sequence labeling format data, extracts course, chapter and knowledge point entities, identifies the inclusion relationship between chapter and knowledge point, constructs a hierarchical structure of courses, chapters and knowledge points in graph database, and forms an educational supply knowledge graph.

[0011] Optionally, S21 includes: S211: Extract the skill entities corresponding to the job positions from the job recruitment data, and use them as input objects for industry-side knowledge reasoning and graph construction. S212, For job skill entities, construct prompt words for knowledge reasoning, perform semantic reasoning on the skill entities through a large language model, map the skills to the corresponding academic course names, and decompose them into multiple atomic knowledge points; S213, based on the reasoning results between job positions, skills, courses and knowledge points, constructs a multi-level association path of job position - skill - knowledge point - course in the graph database to form an industry-side demand knowledge graph.

[0012] Optionally, S22 includes: S221. Input the BIO sequence labeling format data into the BERT-Base-Chinese model to perform contextual semantic modeling of the text and obtain character-level dynamic semantic vectors. ; S222: Input the character-level dynamic semantic vector into the Conditional Random Field (CRF) layer to calculate the label transition probability matrix. Global decoding constraints are applied to the BIO sequence label format data to ensure that the output label sequence conforms to the BIO labeling rules; S223 uses the Entity-Masked BERT model to insert entity tags before and after the identified chapter entities and knowledge point entities, concatenates the vector representations of the corresponding entities, and inputs them into a classifier to determine whether there is an inclusion relationship between chapters and knowledge points; S224. Based on the entity recognition results and relationship determination results, construct the hierarchical inclusion relationship between courses, chapters and knowledge points in the graph database to form an educational supply knowledge graph.

[0013] Optionally, S3 includes: S31, Knowledge points addressing industry-side needs Knowledge points supplied by the education side To address the semantic ambiguity problem, during the semantic encoding stage, information about the parent nodes of knowledge points is concatenated to construct context-enhanced input text, including the input text itself. With input text ; S32, using the Sentence-BERT model to respectively... and Mapped to 768-dimensional semantic vectors and and calculate and Cosine similarity between them as a semantic similarity feature Execute substring inclusion detection rules based on strings and calculate rule similarity features. The semantic similarity features and rule similarity features are weighted and fused to obtain a comprehensive matching score for cross-domain knowledge points. ; S33 will combine the matching score With preset threshold ( ) for comparison, when At this point, the large language model auditing stage is entered, where cases with high similarity but semantic inconsistency are verified and eliminated. After the audit is passed, the correspondence between demand-side knowledge points and supply-side knowledge points is established in the graph database, and the matching confidence level is recorded.

[0014] Optionally, S4 includes: S41, targeting specific positions Obtain the set of knowledge points required by them. Based on the cross-domain knowledge point correspondence, the system queries the graph database for the set of knowledge points already covered by the existing curriculum. Calculate the job skills coverage rate ; S42, Based on the difference between the set of job requirement knowledge points and the set of covered knowledge points, calculate the set of job requirement knowledge points not yet covered by the current curriculum. ; S43, for each unselected course in the candidate course pool Calculate its pair The coverage quantity is used as the marginal gain, and a greedy strategy is employed for iterative selection, specifically including: S431, Initialize the newly added course collection ; S432, in each iteration, select the course with the highest marginal return. join in ; S433, from Remove from Covered knowledge points and updated status; S434, repeat S431-S433 until... Empty or has reached the maximum number of recommended courses (Top-N); S44, based on the newly added course set Generate a report with suggestions for course additions and optimizations.

[0015] The beneficial effects of this invention are: This invention constructs an automatic data collection and processing flow for multi-source heterogeneous data, encompassing job recruitment data and university teaching syllabi. It also performs noise reduction, extraction, and semantic standardization on job skill information, enabling continuous acquisition and automatic updating of massive recruitment data. This overcomes the problems of small sample size, long cycle, and insufficient timeliness of traditional manual surveys, and elevates job demand analysis from annual static analysis to weekly dynamic analysis, thereby improving the comprehensiveness and timeliness of talent demand identification.

[0016] This invention employs a BERT-CRF-based syllabus entity recognition method, combined with an explicit entity tagging relation extraction mechanism and a parsing state machine based on regular expressions and keywords, to perform high-precision extraction and hierarchical modeling of courses, chapters, and knowledge points in unstructured Chinese syllabi. This effectively solves problems such as ambiguous boundaries of long and complex terms, logical conflicts in labels, and interference from contextual noise. While reducing computational complexity, it significantly improves the accuracy and stability of entity and relation extraction.

[0017] This invention constructs a dual-tower heterogeneous structure of an industry-side demand knowledge graph and an education-side supply knowledge graph. Based on a cross-domain alignment strategy combining semantic similarity calculation and rule matching, it bridges the semantic gap between job skill descriptions and curriculum content. Furthermore, it introduces a quantitative calculation model for job skill coverage and a greedy optimization algorithm based on maximizing marginal returns. This enables quantifiable assessment and optimal supplementation decisions of the curriculum system regarding job requirements, thereby maximizing the benefits of job skill coverage under conditions of limited teaching resources. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this 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 for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the adaptation method according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] like Figure 1 As shown, a method for intelligently matching university course content with job skills based on heterogeneous knowledge graphs includes the following steps: S1. Obtain job recruitment data from the industry side and course syllabus data from the education side. Denoise and extract skill information from the job recruitment data, and perform semantic standardization on the job skill descriptions. Parse the course syllabus data, reorganize the unstructured text into text blocks with hierarchical relationships, and convert it into BIO sequence labeling format data. S2. Construct an industry-side demand knowledge graph and an education-side supply knowledge graph respectively. The industry-side demand knowledge graph is used to describe the relationship between job positions, skills, knowledge points and courses, while the education-side supply knowledge graph is used to describe the hierarchical relationship between courses, chapters and knowledge points. The industry-side demand knowledge graph and the education-side supply knowledge graph are stored in a graph database to form a dual-tower structure. S3 performs semantic representation of knowledge points in the industry-side demand knowledge graph and knowledge points in the education-side supply knowledge graph, and maps them to a unified vector space. It then determines the correspondence between cross-domain knowledge points by calculating semantic similarity and combining rule matching. S4, based on the correspondence between cross-domain knowledge points, calculates the coverage of the existing curriculum system with the knowledge points required for the target position, obtains the job skills coverage rate, and generates curriculum optimization or supplementary plans to improve the job skills coverage rate based on the uncovered knowledge points.

[0022] S1 includes: S11, by calling the DrissionPage automation framework, simulates browser behavior to access the recruitment platform and retrieves the corresponding original job posting HTML text data for the target position. S12, perform regular expression cleaning on the obtained raw job recruitment HTML text data to remove HTML tags and redundant text information unrelated to skills, and obtain plain text job description content. S13. For non-standard or mixed expression skill texts in plain text job descriptions, input the skill texts into the Large Language Model (LLM), perform semantic normalization processing, output unified standard skill entities and label their types; S14: Read the PDF byte stream of the college course syllabus, construct a parsing state machine based on regular expression rules (including chapter title matching) and keywords, identify different semantic regions in the syllabus, and reorganize the unstructured text into hierarchical text blocks that include course information and knowledge points. S15 converts the reorganized hierarchical text blocks into BIO (Begin-Inside-Outside) sequence label format data.

[0023] S12 includes: S121, Perform tag matching and stripping operations on the original job posting HTML text data. By deleting character substrings that match the HTML tag pattern, obtain the text content excluding HTML tags. , represented as: ; in, The original job posting HTML text data, for Any substring of characters in, This refers to the regular expression matching pattern for HTML tags; S122, For the text content excluding HTML tags, based on a preset set of non-skill-related stop words, perform redundant text matching and deletion operations to obtain the job description text after filtering out non-skill-related redundant text. , represented as: ; in, For words or phrases in the text, This is a set of stop words that are not related to skills. S123 will filter out redundant non-skill text from the job description text. As a plain text job description Output.

[0024] S13 includes: S131, from plain text job description content Extract the set of candidate fragments belonging to the skill description ,in, For the first A number of candidate skill fragments, The number of candidate segments; S132, each candidate skill fragment Input a large language model and obtain its corresponding standard skill entity name. This achieves semantic normalization of non-standard or mixed expressions, represented as: ; in, For a predefined set of standard skill entities, For a candidate skill entity in the standard skill entity set, For semantic representation functions generated by large language models, This is a semantic similarity measurement function; S133, for each standard skill entity name Assign entity type labels and output standardized results for As a result of job-side skills standardization, it is represented as: ; ; in, For type annotation functions based on large language models, This is the set of standardized skill entities and type annotation results for output.

[0025] S14 includes: S141, Read the PDF byte stream of university course syllabi. And parse it to obtain the text line sequence arranged in the original order. ,in, For the first Line text content, Total number of lines in the text; S142, construct a parsing state machine based on chapter title regular expression matching and keyword triggering rules, and process each line of text content. Generate semantic state This allows for the identification of different semantic regions such as chapters and reference books, represented as: ; ; in, This is the state transition function. The set of regular expression rules for matching chapter titles. for Satisfies the chapter title regular expression rule. This is a set of keywords used for semantic region discrimination. Keywords Appeared middle, This is an indicator function; it returns 1 if the condition is true, and 0 otherwise. The feature vector is triggered by the regular expression / keyword of the text line; S143, based on semantic state sequence Adjacent lines of text with the same semantic state or that satisfy a hierarchical nesting relationship are aggregated into a hierarchical text block set. Each level of text block includes course information and knowledge point-related content, represented as: ; ; in, For the first Each level of text block For text concatenation / aggregation operators, For the first The line index range covered by each text block This represents the semantic state corresponding to the text block. This represents the number of text blocks.

[0026] S2 includes: S21 takes job skill entities as input, performs semantic reasoning and decomposition on job skills based on a large language model, maps skills to corresponding course names and refines them into multiple knowledge points, and builds the relationship between job positions, skills, knowledge points and courses in a graph database to form an industry demand knowledge graph. S22 performs semantic encoding and entity recognition on BIO sequence labeling format data, extracts course, chapter and knowledge point entities, identifies the inclusion relationship between chapter and knowledge point, constructs a hierarchical structure of courses, chapters and knowledge points in graph database, and forms an educational supply knowledge graph.

[0027] S21 includes: S211: Extract the skill entities corresponding to the job positions from the job recruitment data, and use them as input objects for industry-side knowledge reasoning and graph construction. Input example: ; S212: For job skill entities, construct prompt words for knowledge reasoning, perform semantic reasoning on skill entities through a large language model, map skills to corresponding academic course names, and break them down into multiple atomic knowledge points; Prompt inference example: The system build prompt reads: "Please map the skill 'Spark' to an academic course name and break it down into atomic knowledge points." Output data: Academic mapping: Spark → "Distributed Computing" (node ​​attribute type='course name'); Atomic decomposition: Spark → ["RDD", "DAG", "Transformation operator", "Action operator"] (node ​​attribute type='knowledge point'); S213, based on the reasoning results between job positions, skills, courses and knowledge points, constructs a multi-level association path of job position - skill - knowledge point - course in the graph database to form an industry-side demand knowledge graph.

[0028] Example of a graph relationship path: (Job: Big Data Engineer) → [REQUIRES] → (Skill:Spark) → [INCLUDES] → (KP:RDD) → [BELONGS_TO] → (Course: Distributed Computing).

[0029] S22 includes: S221. Input the BIO sequence labeling format data into the BERT-Base-Chinese model to perform contextual semantic modeling of the text and obtain character-level dynamic semantic vectors. ; S222: Input the character-level dynamic semantic vector into the Conditional Random Field (CRF) layer to calculate the label transition probability matrix. Global decoding constraints are applied to the BIO sequence label format data to ensure that the output label sequence conforms to the BIO labeling rules; S223, using the Entity-Masked BERT model, inserts entity tags before and after the identified chapter and knowledge point entities, concatenates the vector representations of the corresponding entities, and inputs them into a classifier to determine whether there is an inclusion relationship between chapters and knowledge points, represented as: ; S224. Based on the entity recognition results and relationship determination results, construct the hierarchical inclusion relationship between courses, chapters and knowledge points in the graph database to form an educational supply knowledge graph.

[0030] Example of a graph relationship path: (Course: Data Structures) → [CONTAINS] → (Chapter: Linear List) → [CONTAINS] → (KP: Reverse list).

[0031] S3 includes: S31, Knowledge points addressing industry-side needs Knowledge points supplied by the education side To address the semantic ambiguity problem, during the semantic encoding stage, information about the parent nodes of knowledge points is concatenated to construct context-enhanced input text, including the input text itself. With input text ; Input text example: Input text =“Distributed Computing [SEP] RDD”; Input text =“Traversal of a data structure [SEP] graph”; S32, using the Sentence-BERT model to respectively... and Mapped to 768-dimensional semantic vectors and and calculate and Cosine similarity between them as a semantic similarity feature Execute substring inclusion detection rules based on strings and calculate rule similarity features. The semantic similarity features and rule similarity features are weighted and fused to obtain a comprehensive matching score for cross-domain knowledge points. , represented as: ; ; ; in, These are weighting coefficients; S33 will combine the matching score With preset threshold ( ) for comparison, when At this point, the large language model auditing stage is entered, where cases with high similarity but semantic inconsistency are verified and eliminated. After the audit is passed, the correspondence between demand-side knowledge points and supply-side knowledge points is established in the graph database, and the matching confidence level is recorded.

[0032] S4 includes: S41, targeting specific positions Obtain the set of knowledge points required by them. Based on the cross-domain knowledge point correspondence, the system queries the graph database for the set of knowledge points already covered by the existing curriculum. Calculate the job skills coverage rate , represented as: ; S42, Based on the difference between the set of job requirement knowledge points and the set of covered knowledge points, calculate the set of job requirement knowledge points not yet covered by the current curriculum. , represented as: ; S43, for each unselected course in the candidate course pool Calculate its pair The coverage quantity is used as the marginal gain, and a greedy strategy is employed for iterative selection, specifically including: S431, Initialize the newly added course collection ; S432, in each iteration, select the course with the highest marginal return. join in ; S433, from Remove from Covered knowledge points and updated status; S434, repeat S431-S433 until... Empty or has reached the maximum number of recommended courses (Top-N); S45, based on the newly added course collection It generates a report with suggestions for course additions and optimizations, clearly indicating the recommended courses and their corresponding effects on improving job skill coverage.

[0033] Example explanation: "It is recommended to add a course on 'Big Data Processing Technology', which could increase the coverage of job skills by 25%."

[0034] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0035] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligently adapting university course content and job skills based on heterogeneous knowledge graphs, characterized in that, Includes the following steps: S1. Obtain job recruitment data from the industry side and course syllabus data from the education side. Denoise and extract skill information from the job recruitment data, and perform semantic standardization on the job skill descriptions. Parse the course syllabus data, reorganize the unstructured text into text blocks with hierarchical relationships, and convert it into BIO sequence labeling format data. S2. Construct an industry-side demand knowledge graph and an education-side supply knowledge graph respectively. The industry-side demand knowledge graph is used to describe the relationship between job positions, skills, knowledge points and courses, while the education-side supply knowledge graph is used to describe the hierarchical relationship between courses, chapters and knowledge points. The industry-side demand knowledge graph and the education-side supply knowledge graph are stored in a graph database to form a dual-tower structure. S3 performs semantic representation of knowledge points in the industry-side demand knowledge graph and knowledge points in the education-side supply knowledge graph, and maps them to a unified vector space. It then determines the correspondence between cross-domain knowledge points by calculating semantic similarity and combining rule matching. S4, based on the correspondence between cross-domain knowledge points, calculates the coverage of the existing curriculum system with the knowledge points required for the target position, obtains the job skills coverage rate, and generates curriculum optimization or supplementary plans to improve the job skills coverage rate based on the uncovered knowledge points.

2. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 1, characterized in that, S1 includes: S11, by calling the DrissionPage automation framework, simulates browser behavior to access the recruitment platform and retrieves the corresponding original job posting HTML text data for the target position. S12, perform regular expression cleaning on the obtained raw job recruitment HTML text data to remove HTML tags and redundant text information unrelated to skills, and obtain plain text job description content. S13: For non-standard or mixed expression skill texts in plain text job descriptions, input the skill texts into the large language model, perform semantic normalization processing, output unified standard skill entities and label their types; S14: Read the PDF byte stream of the college course syllabus, construct a parsing state machine based on regular expression rules and keywords, identify different semantic regions in the syllabus, and reorganize the unstructured text into hierarchical text blocks that include course information and knowledge points. S15 converts the reorganized hierarchical text blocks into BIO sequence label format data.

3. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 2, characterized in that, S12 includes: S121, Perform tag matching and stripping operations on the original job posting HTML text data. By deleting character substrings that match the HTML tag pattern, the text content excluding HTML tags is obtained. ; S122, For the text content excluding HTML tags, based on a preset set of non-skill-related stop words, perform redundant text matching and deletion operations to obtain the job description text after filtering out non-skill-related redundant text. ; S123 will filter out redundant non-skill text from the job description text. As a plain text job description Output.

4. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 3, characterized in that, S13 includes: S131, from plain text job description content Extract the set of candidate fragments belonging to the skill description ,in, For the first A number of candidate skill fragments, The number of candidate segments; S132, each candidate skill fragment Input a large language model and obtain its corresponding standard skill entity name. This enables semantic normalization of non-standard or mixed expressions; S133, for each standard skill entity name Assign entity type labels and output standardized results for As a result of job-side skills standardization.

5. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 4, characterized in that, S14 includes: S141, Read the PDF byte stream of university course syllabi. And parse it to obtain the text line sequence arranged in the original order. ,in, For the first Line text content, Total number of lines in the text; S142, construct a parsing state machine based on chapter title regular expression matching and keyword triggering rules, and process each line of text content. Generate semantic state This allows for the identification of different semantic regions, such as chapters and reference books. S143, based on semantic state sequence Adjacent lines of text with the same semantic state or that satisfy a hierarchical nesting relationship are aggregated into a hierarchical text block set. Each level of text block includes course information and content related to knowledge points.

6. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 5, characterized in that, S2 includes: S21 takes job skill entities as input, performs semantic reasoning and decomposition on job skills based on a large language model, maps skills to corresponding course names and refines them into multiple knowledge points, and builds the relationship between job positions, skills, knowledge points and courses in a graph database to form an industry demand knowledge graph. S22 performs semantic encoding and entity recognition on BIO sequence labeling format data, extracts course, chapter and knowledge point entities, identifies the inclusion relationship between chapter and knowledge point, constructs a hierarchical structure of courses, chapters and knowledge points in graph database, and forms an educational supply knowledge graph.

7. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 6, characterized in that, S21 includes: S211: Extract the skill entities corresponding to the job positions from the job recruitment data, and use them as input objects for industry-side knowledge reasoning and graph construction. S212, For job skill entities, construct prompt words for knowledge reasoning, perform semantic reasoning on the skill entities through a large language model, map the skills to the corresponding academic course names, and decompose them into multiple atomic knowledge points; S213, based on the reasoning results between job positions, skills, courses and knowledge points, constructs a multi-level association path of job position - skill - knowledge point - course in the graph database to form an industry-side demand knowledge graph.

8. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 7, characterized in that, S22 includes: S221. Input the BIO sequence labeling format data into the BERT-Base-Chinese model to perform contextual semantic modeling of the text and obtain character-level dynamic semantic vectors. ; S222, input the character-level dynamic semantic vector into the conditional random field layer, and calculate the label transition probability matrix. Global decoding constraints are applied to the BIO sequence label format data to ensure that the output label sequence conforms to the BIO labeling rules; S223 uses the Entity-Masked BERT model to insert entity tags before and after the identified chapter entities and knowledge point entities, concatenates the vector representations of the corresponding entities, and inputs them into a classifier to determine whether there is an inclusion relationship between chapters and knowledge points; S224. Based on the entity recognition results and relationship determination results, construct the hierarchical inclusion relationship between courses, chapters and knowledge points in the graph database to form an educational supply knowledge graph.

9. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 8, characterized in that, S3 includes: S31, Knowledge points addressing industry-side needs Knowledge points supplied by the education side To address the semantic ambiguity problem, during the semantic encoding stage, information about the parent nodes of knowledge points is concatenated to construct context-enhanced input text, including the input text itself. With input text ; S32, using the Sentence-BERT model to respectively... and Mapped to 768-dimensional semantic vectors and and calculate and Cosine similarity between them as a semantic similarity feature Execute substring inclusion detection rules based on strings and calculate rule similarity features. The semantic similarity features and rule similarity features are weighted and fused to obtain a comprehensive matching score for cross-domain knowledge points. ; S33 will combine the matching score With preset threshold When comparing, At this point, the large language model auditing stage is entered, where cases with high similarity but semantic inconsistency are verified and eliminated. After the audit is passed, the correspondence between demand-side knowledge points and supply-side knowledge points is established in the graph database, and the matching confidence is recorded.

10. The method for intelligent adaptation of university course content and job skills based on heterogeneous knowledge graphs according to claim 9, characterized in that, S4 includes: S41, targeting specific positions Obtain the set of knowledge points required by them. Based on the cross-domain knowledge point correspondence, the system queries the graph database for the set of knowledge points already covered by the existing curriculum. Calculate the job skills coverage rate ; S42, Based on the difference between the set of job requirement knowledge points and the set of covered knowledge points, calculate the set of job requirement knowledge points not yet covered by the current curriculum. ; S43, for each unselected course in the candidate course pool Calculate its pair The coverage quantity is used as the marginal benefit, and a greedy strategy is employed for iterative selection, specifically including: S431, Initialize the newly added course collection ; S432, in each iteration, select the course with the highest marginal return. join in ; S433, from Remove from Covered knowledge points and updated status; S434, repeat S431-S433 until... Empty or has reached the maximum number of recommended courses; S44, based on the newly added course set Generate a report with suggestions for course additions and optimizations.

Images