Recruitment post course system recommendation method and system based on knowledge graph

Abstract

The invention provides a recruitment post course system recommendation method and system based on a knowledge graph, and the method comprises the following steps: collecting recruitment post information data issued by recruitment enterprises in different regions, extracting recruitment post demand entities in the recruitment post information data to construct a post knowledge graph, and storing the post knowledge graph in a Neo4j graph database; according to professional course system teaching resources and course knowledge structures of vocational colleges, constructing a professional course system knowledge base through an LDA model, extracting professional course entities to construct a course knowledge graph, and storing the course knowledge graph into a Neo4j graph database; performing entity vectorization representation on the post knowledge graph and the course knowledge graph, combining similar entities, and constructing a complete post and course knowledge graph; constructing an on-duty recommendation model based on the on-duty knowledge graph, and in a recommendation stage, inputting a post name needing to be queried in the on-duty recommendation model and recommending a course system for the post; according to the method, the intelligent matching model of the post and the course is constructed based on the knowledge graph, so that the accurate butt joint of the post demand and the course system is realized, the intelligence, the accuracy and the matching degree of course recommendation are improved, and the pertinence of vocational education and the market adaptability of talent training are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and particularly to a method and system for recommending a recruitment position curriculum system based on a knowledge graph. Background Art

[0002] In the current rapidly developing social and economic environment, enterprises have put forward higher requirements for the professional skills, knowledge structure and comprehensive qualities of talents. However, the curriculum settings of vocational colleges often fail to closely follow the changes in market demands, resulting in a certain disconnection between the skills of graduates and the needs of enterprises. At the same time, traditional curriculum recommendation methods mainly rely on manual experience or simple matching based on historical learning records, lacking in-depth understanding of job requirements and intelligent analysis, and it is difficult to accurately recommend a curriculum system that meets job requirements. Therefore, how to effectively connect the job requirements of enterprises with the curriculum system of vocational colleges and improve the pertinence of talent cultivation and employment competitiveness has become an urgent problem to be solved in the current fields of vocational education and human resource management.

[0003] After reviewing relevant publicly disclosed technical solutions, the technology with the publication number CN119359504A proposes a vocational teaching planning method and system based on individual advantage degree analysis. This solution is based on a recruitment database and a student learning database, and statistics are made on the supply and demand of positions and personnel, the deviation between social recruitment positions and campus recruitment positions, the matching degree between positions and majors, and the supply and demand prediction value of positions; the individual advantage degree of students on each position is calculated; based on the supply and demand of positions and personnel, the deviation between social recruitment positions and campus recruitment positions, the matching degree between positions and majors, the supply and demand prediction value of positions, and the individual advantage degree of students, vocational planning of students based on individual advantage degree is carried out; the matching evaluation result of positions and professional courses is obtained, and curriculum planning and teacher allocation at the department level are carried out. This solution realizes the effect of updating the curriculum arrangement in a timely manner by associating learning content with job requirements. However, this solution mainly conducts vocational planning based on statistical analysis and calculation of individual advantage degree, lacking in-depth semantic understanding and intelligent matching between job requirements and the curriculum system, and failing to utilize knowledge graphs and deep learning models to accurately mine and recommend the curriculum system that best meets job requirements, thus having certain limitations in the accuracy and flexibility of recommendation. Summary of the Invention

[0004] The purpose of the present invention is to propose a method and system for recommending a recruitment position curriculum system based on a knowledge graph in view of the current deficiencies.

[0005] The present invention adopts the following technical solutions:

[0006] A method for recommending a recruitment position curriculum system based on a knowledge graph, the method comprising the following steps:

[0007] S1: Collect the recruitment position information data released by recruitment enterprises in different regions, clean and format the preprocessing of the data, extract the recruitment position demand entities in the recruitment position information data through the bert_bilstm_crf model to construct a position knowledge graph, and store it in the Neo4j graph database;

[0008] S2: According to the teaching resources of the professional curriculum system of vocational colleges and the curriculum knowledge structure, construct a professional curriculum system knowledge base through the LDA model, extract professional curriculum entities through the bert_bilstm_crf model to construct a curriculum knowledge graph, and store it in the Neo4j graph database;

[0009] S3: Use the TransE model to vectorize the entities of the position knowledge graph and the curriculum knowledge graph, calculate the semantic similarity of each entity through cosine similarity, merge similar entities, and construct a complete position-course knowledge graph;

[0010] S4: Build a position-course recommendation model based on the position-course knowledge graph, and in the recommendation stage, input the position name to be queried in the position-course recommendation model and recommend a curriculum system for this position.

[0011] Furthermore, before extracting the recruitment position demand entities, first annotate the recruitment position demand entities in the recruitment position information data to obtain the training data required by the model. The recruitment position demand entity is the specific demand entity information of the enterprise for this position. The specific process of annotating the recruitment position demand entity is as follows:

[0012] S111: Manually collect the demand types of knowledge, skills, qualities and majors in the recruitment position requirements, and label each demand type with corresponding tags to construct a knowledge base;

[0013] S112: Load the constructed knowledge base into the word segmentation dictionary as a custom word dictionary, and use the tag as the custom word part of speech;

[0014] S113: Segment the recruitment position information data text by sentence;

[0015] S114: Annotate the word segmentation obtained in step S13 according to the custom word part of speech: If the part of speech in the word segmentation contains the custom word part of speech, then further split the word into several single characters, label the first character with B-custom word part of speech, and label the other single characters with I-custom word part of speech; If the part of speech in the word segmentation does not contain the custom word part of speech, then label each character in the word with O, and each line has one character, and the annotation label after the character is separated by a space; The annotations between sentences are separated by blank lines.

[0016] S115: Organize the annotated word segmentation as training data and save it in a format that can be used for deep learning model training.

[0017] Further, the specific types of the recruitment position requirement entities include position knowledge requirement entities, position skill requirement entities, position quality requirement entities, and position specialty entities.

[0018] Further, in the step S1, the bert_bilstm_crf model includes a bert layer, a bilstm layer, and a crf layer; the process of extracting recruitment position requirement entities by the bert_bilstm_crf model is as follows:

[0019] S121: Obtain the labeled training data, and perform word segmentation on each piece of training data to obtain a word sequence;

[0020] S122: Input the word sequence into the bert layer of the model, and the bert layer maps each word into a corresponding word embedding vector according to the pre-trained model;

[0021] S123: Send the word embedding vector into the bilstm layer, and the bilstm layer processes the word embedding vector through forward and backward lstm respectively, and generates forward and backward directional vector representations respectively; splice the forward and backward directional vectors, and send the spliced vector into the softmax function to calculate the probability of each word corresponding to the entity label;

[0022] S124: Send the probability of each word corresponding to the entity label into the crf layer, so as to obtain the optimal label sequence and extract the position requirement entity.

[0023] Further, in the step S1, a position knowledge graph is established in the following specific way:

[0024] S131: After the position requirement entities are extracted, store each position requirement entity in the form of a triple; the triple specifically includes a head entity, a relationship name, and a tail entity; the head entity is the position name, the relationship name is the entity type, and the tail entity is the position requirement entity; each triple represents the relationship between a position name and the corresponding position requirement entity of the position;

[0025] S132: Store all the constructed triples in csv format, with each triple occupying one line;

[0026] S133: Connect to the Neo4j graph database using the py2neo library, and use the pandas library in Python to parse the CSV file of triples. Store the job names and job requirement entities as nodes in the Neo4j graph database, use the entity type as the associated edge between the job name and job requirement entities to connect the corresponding nodes, and merge duplicate nodes, thereby constructing a job knowledge graph and storing it in the Neo4j graph database.

[0027] Further, in step S2, the professional curriculum system knowledge base is constructed in the following specific way:

[0028] S211: According to the professional settings of vocational colleges, collect the teaching resources corresponding to the basic courses, professional courses, and training courses of different majors.

[0029] S212: Use jieba to perform word segmentation preprocessing on the text of the teaching resources, and use the LDA model to extract the keywords in each chapter of each course in the text.

[0030] S213: Classify the keywords according to their positions in the teaching resource text. The classification categories include knowledge point core words, ability keywords, and literacy keywords; and label the keywords of each category with corresponding tags and store them in a txt file to form a professional curriculum system knowledge base for vocational colleges.

[0031] Further, the specific types in the professional curriculum entity include curriculum knowledge entity, curriculum skill entity, curriculum literacy cultivation entity, and curriculum major entity.

[0032] 8. According to the method for recommending a recruitment position curriculum system based on a knowledge graph described in claim 1, it is characterized in that in step S3, the job-course knowledge graph is constructed in the following specific way:

[0033] S31: Represent the job knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors.

[0034] S32: Represent the curriculum knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors.

[0035] S33: Classify the job requirement entities and professional curriculum entities according to the consistency of entity types in the job knowledge graph and the curriculum knowledge graph, and establish corresponding relationships; the specific content of the classification and corresponding relationship establishment is as follows:

[0036] Knowledge category: The job knowledge requirement entities correspond to the curriculum knowledge entities:

[0037] Skill category: The job skill requirement entities correspond to the curriculum skill entities;

[0038] For the quality category: the job quality requirement entity corresponds to the course quality cultivation entity;

[0039] For the professional category: the job professional entity corresponds to the course professional entity;

[0040] S34: Calculate the cosine similarity of the corresponding entity word embedding vectors within each category in step S33 as the semantic similarity between entities;

[0041] S35: Extract entity pairs with semantic similarity greater than the specified threshold in each category; each entity pair is specifically two corresponding entities in a certain category;

[0042] S36: Merge the entity pairs extracted in step S35, eliminate redundancy, form job-course triples, and use the job-course triples as the job-course knowledge graph.

[0043] Furthermore, in the said step S4, the job-course recommendation model is established in the following manner:

[0044] S411: According to the connection of the head entity, relationship name, and tail entity in the Neo4j graph database, use Cypher query statements to query the relationship name and tail entity corresponding to each head entity, and represent them as triples (h, l, t); where h represents the head entity, l represents the relationship name, and t represents the tail entity;

[0045] S412: Construct all triples into a training set S, merge all the head entities and tail entities in the training set and remove duplicate entity words to form an entity set E, and the entity set contains course names, job names, professional course entities, and job requirement entities; represent all the relationship names in the training set as a relationship set L, then at this time h ∈ E, t ∈ E, l ∈ L;

[0046] S413: Use the TransE model as the job-course recommendation model to learn the relationship mapping between job requirement entities and professional course entities, and this model satisfies the following conditions:

[0047] S4131: Assign vector representations to all entities and relationship types and construct the target relationship:

[0048] h + l ≈ t;

[0049] Define the distance from h + l to t as d(h + l, t), and calculate this distance using the L1 norm;

[0050] S4132: Define the loss function of the TransE model as:

[0051] Loss = ∑ (h,l,t)∈s ∑ (h′,l,t′)∈s′[γ + d(h + l, t) - d(h′, l, t′)] + ;

[0052] S′ = {(h′, l, t)|h′ ∈ E} ∪ {(h′, l, t′)|t′ ∈ E};

[0053] wherein, + when the content in [] is positive, take the value in []; when [] is negative or 0, take 0; γ is a hyperparameter with a trend greater than 0, representing the interval correction distance between a correct triple and an incorrect triple. The correct triple is the known true relationship in the post-course knowledge graph, and the vector distance calculated by it is expressed as d(h + l, t); the incorrect triple is an unreasonable relationship generated by replacing the head entity or the tail entity in the correct triple, and the vector distance calculated by it is expressed as d(h′, l, t′); S′ is a negative sample set; (h′, l, t) and (h, l, t′) represent triples generated by random recombination.

[0054] (h′, l, t) means that when the relationship name and the tail entity remain unchanged, the original head entity is replaced by another head entity h′; (h, l, t′) means that when the relationship name and the head entity remain unchanged, the original tail entity is replaced by another tail entity t ′ ; The purpose of model training is to maximize the distance between positive samples and negative samples, making the relationship mapping of the knowledge graph more accurate; the positive samples are the set of correct triples, and the negative samples are the set of incorrect triples.

[0055] A recruitment post course system recommendation system based on a knowledge graph is applicable to a recruitment post course system recommendation method based on a knowledge graph. The system includes a data collection module, a knowledge extraction and knowledge graph construction module, a post-course knowledge graph fusion module, a post-course recommendation model construction module, and a post-course recommendation query module; the data collection module is used to obtain recruitment post information of different positions and teaching resources of different professional course systems through a third-party data source; the knowledge extraction and knowledge graph construction module is used to construct a post knowledge graph and a course knowledge graph in combination with the data collected by the data collection module; the post-course knowledge graph fusion module is used to vectorize and fuse the post knowledge graph and the course knowledge graph to form a complete post-course knowledge graph; the post-course recommendation model construction module is used to construct a post-course recommendation model in combination with the post-course knowledge graph; the post-course recommendation query module is used to provide a recommended course system corresponding to the input position for the user.

[0056] The beneficial effects achieved by the present invention:

[0057] The present invention achieves the precise matching of job requirements and course resources by constructing a recruitment position and curriculum system matching model based on knowledge graph; compared with traditional course recommendation methods, the present invention can automatically extract job requirement information, and combine semantic similarity calculation and knowledge graph fusion to provide personalized and precise course recommendations, effectively improving the matching degree between vocational education and market demand, and enhancing the pertinence of talent training and employment competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The present invention can be further understood from the following description in conjunction with the accompanying drawings. The components in the figures are not necessarily drawn to scale, but rather the emphasis is placed on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0059] Figure 1 This is a flow chart of the method for recommending a recruitment position course system based on knowledge graphs of the present invention.

[0060] Figure 2 This is a flow chart of the method for establishing a job knowledge graph according to the present invention.

[0061] Figure 3 A flowchart of the method for establishing a course knowledge graph of the present invention.

[0062] Figure 4 This is a flow chart of the method for establishing a job-course knowledge graph according to the present invention.

[0063] Figure 5 This is a flow chart of a specific method for the job recommendation course system based on the present invention.

[0064] Figure 6 It is a schematic diagram of the overall module of the system of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with its embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention; for those skilled in the art, after reviewing the following detailed description, other systems, methods and / or features of the present embodiment will become apparent; it is intended that all such additional systems, methods, features and advantages are included in this specification; included within the scope of the present invention and protected by the appended claims; additional features of the disclosed embodiments are described in the following detailed description, and these features will be apparent from the following detailed description.

[0066] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the accompanying drawings. This is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only for illustrative purposes and cannot be construed as a limitation of this patent. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0067] Embodiment 1:

[0068] As Figure 1 shown, this embodiment provides a method for recommending a recruitment position curriculum system based on a knowledge graph. The method includes the following steps:

[0069] S1: Collect recruitment position information data released by recruitment enterprises in different regions, perform cleaning and formatting preprocessing on the data, extract recruitment position requirement entities in the recruitment position information data through the bert_bilstm_crf model to construct a position knowledge graph, and store it in the Neo4j graph database;

[0070] S2: According to the teaching resources and curriculum knowledge structure of the professional curriculum system of vocational colleges, construct a professional curriculum system knowledge base through the LDA model, extract professional curriculum entities through the bert_bilstm_crf model to construct a curriculum knowledge graph, and store it in the Neo4j graph database;

[0071] S3: Use the TransE model to perform entity vectorization representation on the position knowledge graph and the curriculum knowledge graph, calculate the semantic similarity of each entity through cosine similarity, merge similar entities, and construct a complete position-course knowledge graph;

[0072] S4: Construct a position-course recommendation model based on the position-course knowledge graph, and in the recommendation stage, input the position name to be queried in the position-course recommendation model and recommend a curriculum system for this position;

[0073] Further, in the step S1, the main process of collecting recruitment position information data includes using the requests library to obtain the website addresses and web page data of recruitment websites in different cities and regions, and using the bs4 library and the json library to parse and extract the web page data, and extracting data such as position numbers, position names, recruitment companies, work locations, educational requirements, position requirements, and release dates in the web page as recruitment position information data;

[0074] Furthermore, in step S1, the main process of cleaning and preprocessing the recruitment position information data includes removing duplicate data, special punctuation data and additional symbol data in the recruitment position information data, and using periods or semicolons to separate the recruitment position information data into sentences;

[0075] Furthermore, in step S1, before extracting the recruitment position requirement entities, the recruitment position requirement entities in the recruitment position information data are first labeled to obtain the training data required by the model. The recruitment position requirement entities are the specific requirement entity information of the enterprise for the position. The specific process of labeling the recruitment position requirement entities is as follows:

[0076] S111: Manually collect the required types of knowledge, skills, qualities, and expertise in recruitment positions, and label each requirement type with a corresponding label to build a knowledge base;

[0077] S112: Load the constructed knowledge base into the word segmentation dictionary as a custom word dictionary, and use the label as the custom word part of speech;

[0078] S113: Segmenting the recruitment position information data text into sentences;

[0079] S114: Mark the segmented words obtained in step S13 according to the custom word part of speech: if the part of speech in the segmented word includes the custom word part of speech, the word is further split into several individual characters, the first character is marked with B-custom word part of speech, and the other individual characters are marked with I-custom word part of speech; if the part of speech in the segmented word does not include the custom word part of speech, each character in the word is marked with O, and one character is placed per line, and the marking labels after the characters are separated by spaces; the markings between sentences are separated by blank lines;

[0080] S115: Organize the annotated word segments as training data and save them in a format that can be used for deep learning model training;

[0081] Furthermore, the specific types of the recruitment position requirement entities include position knowledge requirement entities, position skill requirement entities, position quality requirement entities and position professional entities;

[0082] Specifically, the recruitment position requirement entities include but are not limited to entity information such as the company's job requirements, educational requirements, professional requirements, work experience requirements, skill requirements, knowledge requirements, and professional quality requirements for the position. For example, in the recruitment position information data, the job responsibilities and job requirements for an "industrial robot debugging engineer" posted by a company are as follows:

[0083] Job Responsibilities: 1. Research on robot-related technologies and solve technical problems; 2. Teach and debug processes such as robot welding, gluing, hemming, and vision; 3. Conduct joint debugging, program programming, system configuration, and fault troubleshooting of industrial robots; 4. Handle pre- and post-project work related to robot applications, including customer communication, equipment selection, after-sales service, technical training support, etc.

[0084] Job Requirements: 1. Have more than 3 years of experience in white body debugging; 2. Have good English oral and written expression skills; 3. Have good communication skills and an optimistic and positive personality; 4. Those who are familiar with KUKA, ABB, and FANUC robots are preferred; 5. Have a college degree or above, majoring in electrical, automation-related fields; 6. Be able to independently be responsible for the robot application debugging work of relevant projects; 7. Have a strong sense of responsibility, work carefully and seriously, have teamwork spirit, and have organizational and coordination abilities; be able to accept frequent business trips.

[0085] Then the recruitment position requirement entity types and entities included in this position are as follows:

[0086]

[0087] Furthermore, in the step S1, the bert_bilstm_crf model includes a bert layer, a bilstm layer, and a crf layer; the process of extracting recruitment position requirement entities by the bert_bilstm_crf model is specifically as follows:

[0088] S121: Obtain the labeled training data, and perform word segmentation on each piece of training data in units of words to obtain a word sequence;

[0089] S122: Input the word sequence into the bert layer of the model, and the bert layer maps each word into a corresponding word embedding vector according to the pre-trained model;

[0090] S123: Send the word embedding vector into the bilstm layer, and the bilstm layer processes the word embedding vector through forward and backward lstm respectively, and generates forward and backward direction vector representations respectively; Concatenate the forward and backward direction vectors, and send the concatenated vector into the softmax function to calculate the probability of each word corresponding to the entity label;

[0091] S124: Send the probability of each word corresponding to the entity label into the crf layer, so as to obtain the optimal label sequence and extract the position requirement entity;

[0092] Furthermore, as Figure 2 shown, in the step S1, the position knowledge graph is specifically established in the following way:

[0093] S131: After the job requirement entities are extracted, each job requirement entity is stored in the form of a triple; the triple specifically includes a head entity, a relationship name, and a tail entity; the head entity is the job name, the relationship name is the entity type, and the tail entity is the job requirement entity; each triple represents the relationship between a job name and the corresponding job requirement entity of that job.

[0094] S132: Store all the constructed triples in csv format, with each triple occupying one line.

[0095] S133: Use the py2neo library to connect to the Neo4j graph database, and apply the pandas library in python to parse the csv file of triples. Store the job name and job requirement entities as nodes in the Neo4j graph database, use the entity type as the association edge between the job name and the job requirement entity to connect the corresponding nodes, and merge duplicate nodes, thereby constructing a job knowledge graph and storing it in the Neo4j graph database.

[0096] Specifically, the py2neo library is a third-party library used to interact between the python language and the Neo4j graph database. By providing interfaces for graph database connection, query, storage, and management, the py2neo library enables python code to directly control the content in the Neo4j graph database for applications such as constructing, querying, and maintaining knowledge graphs; the Neo4j graph database is a relational database based on a graph structure, specifically used for storing and querying graph data; the pandas library is an open-source library in python for data processing and analysis.

[0097] Further, in step S2, the professional curriculum system knowledge base is constructed specifically in the following way:

[0098] S211: According to the professional settings of vocational colleges, collect the teaching resources corresponding to the basic courses, professional courses, and training courses of different majors.

[0099] S212: Use jieba to segment the text of the teaching resources for preprocessing, and use the LDA model to extract the keywords in each chapter of each course in the text.

[0100] S213: Classify the keywords according to their positions in the teaching resource text where they are located. The classification categories include knowledge point core words, ability keywords, and literacy keywords; and label the keywords of each category with corresponding tags and store them in a txt file to form a professional curriculum system knowledge base for vocational colleges.

[0101] Further, the specific types in the professional curriculum entities include curriculum knowledge entities, curriculum skill entities, curriculum quality cultivation entities, and curriculum major entities;

[0102] Further, in step S2, the professional curriculum entities are extracted in the following specific way:

[0103] S221: Segment the text in the knowledge base of the professional curriculum system of vocational colleges into words to obtain a word sequence;

[0104] S222: Input the word sequence into the bert layer of the model, and the bert layer maps each word into a corresponding word embedding vector according to the pre-trained model;

[0105] S223: Send the word embedding vectors into the bilstm layer. The bilstm layer processes the word embedding vectors through the forward and backward lstm respectively, and generates the forward and backward directional vector representations respectively; Concatenate the forward and backward directional vectors, and send the concatenated vector into the softmax function to calculate the probability of each word corresponding to the entity label;

[0106] S224: Send the probability of each word corresponding to the entity label into the crf layer, so as to obtain the optimal label sequence and extract the professional curriculum entities;

[0107] Further, as Figure 3 shown, in step S2, the curriculum knowledge graph is established in the following specific way:

[0108] S231: After the professional curriculum entities are extracted, store each professional curriculum entity in the form of a triple; The triple specifically includes a head entity, a relationship name, and a tail entity; The head entity is the curriculum name, the relationship name is the entity type, and the tail entity is the professional curriculum entity; Each triple represents the relationship between a curriculum name and the corresponding professional curriculum entity;

[0109] S232: Store all the constructed triples in csv format, with each triple occupying one line;

[0110] S233: Use the py2neo library to connect to the Neo4j graph database, and use the pandas library in python to parse the csv file of the triples. Store the curriculum name and the professional curriculum entity as nodes in the Neo4j graph database, use the entity type as the association edge between the curriculum name and the professional curriculum entity to connect the corresponding nodes, and merge the duplicate nodes, and then construct the curriculum knowledge graph and store it in the Neo4j graph database;

[0111] Further, as Figure 4As shown in the figure, in step S3, the post-course knowledge graph is constructed in the following specific way:

[0112] S31: Represent the post knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors;

[0113] S32: Represent the course knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors;

[0114] S33: Classify the post demand entities and professional course entities according to the consistency of entity types in the post knowledge graph and the course knowledge graph, and establish corresponding relationships; The specific content of classification and corresponding relationship establishment is as follows:

[0115] Knowledge category: The post knowledge demand entities correspond to the course knowledge entities:

[0116] Skill category: The post skill demand entities correspond to the course skill entities;

[0117] Quality category: The post quality demand entities correspond to the course quality cultivation entities;

[0118] Major category: The post major entities correspond to the course major entities;

[0119] S34: Calculate the cosine similarity of the word embedding vectors of the corresponding entities in each category in step S33 as the semantic similarity between entities;

[0120] S35: Extract the entity pairs with semantic similarity greater than the specified threshold in each category; Each entity pair is specifically two corresponding entities in a certain category;

[0121] S36: Merge the entity pairs extracted in step S35, eliminate redundancy, form post-course triples, and use the post-course triples as the post-course knowledge graph;

[0122] Furthermore, in step S4, the post-course recommendation model is established in the following way:

[0123] S411: According to the connection of the head entity, relationship name, and tail entity in the Neo4j graph database, use the Cypher query statement to query the relationship name and tail entity corresponding to each head entity, and represent them as triples (h, l, t); where h represents the head entity, l represents the relationship name, and t represents the tail entity;

[0124] S412: Construct all triples into a training set S. After merging all the head entities and tail entities in the training set and removing duplicate entity words, an entity set E is formed. The entity set contains course names, job names, professional course entities, and job requirement entities. Represent all the relation names in the training set as a relation set L. At this time, h ∈ E, t ∈ E, and l ∈ L.

[0125] S413: Use the TransE model as the job-course recommendation model to learn the relationship mapping between job requirement entities and professional course entities. This model satisfies the following conditions:

[0126] S4131: Assign vector representations to all entities and relation types and construct the target relation:

[0127] h + l ≈ t;

[0128] Define the distance from h + l to t as d(h + l, t), and calculate this distance using the L1 norm.

[0129] S4132: Define the loss function of the TransE model as:

[0130] Loss = ∑ (h,l,t)∈S ∑ (h′,l,t′)∈S′ [γ + d(h + l, t) - d(h′, l, t′)] + ;

[0131] S′ = {(h′, l, t)|h′ ∈ E} ∪ {(h′, l, t′)|t′ ∈ E};

[0132] where, + it means taking the value inside [] when the value inside [] is positive, and taking 0 when [] is negative or 0; γ is a hyperparameter with a trend greater than 0, representing the interval correction distance between a correct triple and an incorrect triple. The correct triple is the known true relationship in the job-course knowledge graph, and the vector distance calculated for it is expressed as d(h + l, t); the incorrect triple is an unreasonable relationship generated by replacing the head entity or tail entity in the correct triple, and the vector distance calculated for it is expressed as d(h′, l, t′); S′ is the negative sample set; (h′, l, t) and (j, l, t′) represent randomly reorganized triples.

[0133] (h′, l, t) means replacing the original head entity with another head entity h′ while keeping the relation name and tail entity unchanged; (h, l, t′) means replacing the original tail entity with another tail entity t while keeping the relation name and head entity unchanged. ′ ; The purpose of model training is to maximize the distance between positive samples and negative samples, making the relationship mapping of the knowledge graph more accurate. The positive samples are the set of correct triples, and the negative samples are the set of incorrect triples.

[0134] Further, as Figure 5 shown, in the step S4, the specific process of recommending a course system for the input position includes:

[0135] S421: Input the name of the position to be recommended into the Neo4j graph database and query the position requirement entities related to this position;

[0136] S422: Based on the trained position-course recommendation model, extract the word embedding vectors corresponding to the position requirement entities obtained in step S421;

[0137] S423: Extract the word embedding vectors corresponding to all professional course entities in the position-course recommendation model, and calculate the cosine similarity between each position requirement entity obtained in step S421 and all professional course entities in combination with the mapping relationship obtained by the position-course recommendation model;

[0138] S424: Sort the calculated cosine similarities corresponding to each position requirement entity in descending order, and select the top three professional course entities corresponding to the cosine similarities as the course system recommended for this position.

[0139] Embodiment 2:

[0140] This embodiment should be understood as including at least all the features of any one of the foregoing embodiments, and further improved on this basis;

[0141] As Figure 6 shown, this embodiment provides a recruitment position course system recommendation system based on a knowledge graph. The system includes a data collection module, a knowledge extraction and knowledge graph construction module, a position-course knowledge graph fusion module, a position-course recommendation model construction module, and a position-course recommendation query module; the data collection module is used to obtain recruitment position information for different positions and teaching resources for different professional course systems through third-party data sources; the knowledge extraction and knowledge graph construction module is used to construct a position knowledge graph and a course knowledge graph in combination with the data collected by the data collection module; the position-course knowledge graph fusion module is used to vectorize and represent the position knowledge graph and the course knowledge graph and fuse them into a complete position-course knowledge graph; the position-course recommendation model construction module is used to construct a position-course recommendation model in combination with the position-course knowledge graph; the position-course recommendation query module is used to provide the recommended course system corresponding to the input position for the user.

[0142] Further, the system further includes a server side, a data storage side, and a user terminal, which are used to support the computing, storage, and query functions of each module; the server side includes a high-performance computing server equipped with a multi-core CPU, GPU, large-capacity RAM, and SSD storage, and is used to execute data processing, knowledge graph construction, and recommendation calculation; the data storage side includes a Neo4j graph database and a MySQL / PostgreSQL database; the user terminal supports the PC side, the Web side, and the mobile side, and the user can input the job name through the front-end interface to query and obtain the recommended curriculum system.

[0143] The content disclosed above is only the preferred feasible embodiment of the present invention, and does not limit the protection scope of the present invention. Therefore, all equivalent technical changes made by using the content of the specification and drawings of the present invention are included in the protection scope of the present invention. In addition, with the development of technology, the elements therein can be updated.

Claims

1. A recruitment position curriculum system recommendation method based on a knowledge graph, characterized in that The method includes the following steps: S1: Collect the recruitment position information data released by recruitment enterprises in different regions, clean and format the data for preprocessing, extract the recruitment position demand entities in the recruitment position information data through the bert_bilstm_crf model to construct a position knowledge graph, and store it in the Neo4j graph database; S2: According to the teaching resources and curriculum knowledge structure of the professional curriculum system in vocational colleges, construct a professional curriculum system knowledge base through the LDA model, extract professional curriculum entities through the bert_bilstm_crf model to construct a curriculum knowledge graph, and store it in the Neo4j graph database; S3: Use the TransE model to perform entity vectorization representation on the position knowledge graph and the curriculum knowledge graph, calculate the semantic similarity of each entity through cosine similarity, merge similar entities, and construct a complete position-course knowledge graph; S4: Construct a position-course recommendation model based on the position-course knowledge graph, and in the recommendation stage, input the position name to be queried into the position-course recommendation model and recommend a curriculum system for this position.

2. The recruitment position curriculum system recommendation method based on a knowledge graph according to claim 1, wherein In the step S1, before extracting the recruitment position demand entities, first annotate the recruitment position demand entities in the recruitment position information data to obtain the training data required by the model. The recruitment position demand entity is the specific demand entity information of the enterprise for this position. The specific process of annotating the recruitment position demand entity is as follows: S111: Manually collect the demand types of knowledge, skills, qualities, and majors in the recruitment position requirements, and annotate corresponding labels for each demand type to construct a knowledge base; S112: Load the constructed knowledge base into the word segmentation dictionary as a custom word dictionary, and use the label as the custom word part of speech; S113: Segment the recruitment position information data text by sentence; S114: Annotate the word segmentation obtained in step S13 according to the custom word part of speech: If the part of speech in the word segmentation contains the custom word part of speech, then further split the word into several single characters, label the first character as B-custom word part of speech, and label other single characters as I-custom word part of speech; If the part of speech in the word segmentation does not contain the custom word part of speech, then label each character in the word as O, and each line has one character, and the annotation label after the character is separated by a space; The annotations between sentences are separated by blank lines; S115: Organize the annotated word segmentation as training data and save it in a format that can be used for deep learning model training.

3. The recruitment position curriculum system recommendation method based on a knowledge graph according to claim 2, wherein The specific types of the recruitment position demand entities include position knowledge demand entities, position skill demand entities, position quality demand entities, and position major entities.

4. The recruitment position curriculum system recommendation method based on a knowledge graph according to claim 3, wherein In the step S1, the bert_bilstm_crf model includes a bert layer, a bilstm layer, and a crf layer; the process of extracting the recruitment position demand entities through the bert_bilstm_crf model is as follows: S121: Obtain the annotated training data, and perform word segmentation processing on each piece of training data by character to obtain a character sequence; S122: Input the character sequence into the BERT layer of the model. The BERT layer maps each character into a corresponding character embedding vector according to the pre-trained model; S123: Send the character embedding vectors into the BiLSTM layer. The BiLSTM layer processes the character embedding vectors through the forward and backward LSTMs respectively, and generates the forward and backward directional vector representations respectively. Concatenate the forward and backward directional vectors, and send the concatenated vector into the softmax function to calculate the probability of each character corresponding to the entity label; S124: Send the probability of each character corresponding to the entity label into the CRF layer, so as to obtain the optimal label sequence and extract the job requirement entity.

5. A method for recommending a recruitment position curriculum system based on a knowledge graph according to claim 4, characterized in that, In step S1, the job knowledge graph is established in the following specific way: S131: After the job requirement entities are extracted, store each job requirement entity in the form of a triple. The triple specifically includes a head entity, a relationship name, and a tail entity. The head entity is the job name, the relationship name is the entity type, and the tail entity is the job requirement entity. Each triple represents the relationship between a job name and the corresponding job requirement entity; S132: Store all the constructed triples in the csv format, with each triple occupying one line; S133: Use the py2neo library to connect to the Neo4j graph database, and use the pandas library in Python to parse the csv file of triples. Store the job name and job requirement entity as nodes in the Neo4j graph database, use the entity type as the associated edge between the job name and the job requirement entity to connect the corresponding nodes, and merge duplicate nodes, thereby constructing a job knowledge graph and storing it in the Neo4j graph database.

6. The recruitment position curriculum system recommendation method based on a knowledge graph according to claim 5, characterized in that, In step S2, the professional curriculum system knowledge base is constructed in the following specific way: S211: According to the professional settings of vocational colleges, collect the teaching resources corresponding to the basic courses, professional courses, and training courses of different majors; S212: Use jieba to segment the text of the teaching resources for preprocessing, and use the LDA model to extract the keywords in each chapter of each course in the text; S213: Classify the keywords according to their positions in the teaching resource text. The classification categories include knowledge point core words, ability keywords, and literacy keywords. And label the corresponding tags for the keywords of each category and store them in a txt file to form a professional curriculum system knowledge base for vocational colleges.

7. A method for recommending a recruitment position curriculum system based on a knowledge graph according to claim 6, characterized in that, The specific types in the professional curriculum entity include course knowledge entity, course skill entity, course literacy cultivation entity, and course major entity.

8. A method for recommending a recruitment position curriculum system based on a knowledge graph according to claim 1, characterized in that In step S3, the job-course knowledge graph is constructed in the following specific way: S31: Represent the job knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors; S32: Represent the course knowledge graph as triples, and use the TransE model to convert the content in the triples into word embedding vectors; S33: Classify the job requirement entities and professional course entities according to the consistency of entity types in the job knowledge graph and the course knowledge graph, and establish the corresponding relationships. The specific content of classification and corresponding relationship establishment is as follows: Knowledge category: The job knowledge requirement entities correspond to the course knowledge entities: Skill category: The job skill requirement entities correspond to the course skill entities; Accomplishment category: The job accomplishment requirement entities correspond to the course accomplishment cultivation entities; Major category: The job major entities correspond to the course major entities; S34: Calculate the cosine similarity of the entity word embedding vectors within each category in step S33 as the semantic similarity between entities; S35: Extract the entity pairs with semantic similarity greater than the specified threshold in each category; each entity pair is specifically two corresponding entities in a certain category; S36: Merge the entity pairs extracted in step S35, eliminate redundancy, form job-course triples, and use the job-course triples as the job-course knowledge graph.

9. A method for recommending a recruitment position curriculum system based on a knowledge graph according to claim 8, characterized in that, In step S4, the job-course recommendation model is established in the following way: S411: According to the connection of the head entity, relationship name, and tail entity in the Neo4j graph database, use Cypher query statements to query the relationship name and tail entity corresponding to each head entity, and represent them as triples (h, l, t); where h represents the head entity, l represents the relationship name, and t represents the tail entity; S412: Construct all triples into a training set S. After merging all the head entities and tail entities in the training set and removing duplicate entity words, form an entity set E. The entity set contains course names, job names, professional course entities, and job requirement entities; Represent all the relationship names in the training set as a relationship set L. At this time, h ∈ E, t ∈ E, and l ∈ L; S413: Use the TransE model as the job-course recommendation model to learn the relationship mapping between job requirement entities and professional course entities. This model satisfies the following conditions: S4131: Assign vector representations to all entities and relationship types and construct the target relationship: h + l ≈ t; Define the distance from h + l to t as d(h + l, t), and calculate this distance using the L1 norm; S4132: Define the loss function of the TransE model as: Loss=∑ (h,l,t)∈s ∑ (h′,l,t′)∈s′ [γ + d(h + l, t) - d(h′, l, t′)] + ; S′ = {(h′, l, t)|h′ ∈ E} ∪ {(h′, l, t′)|t′ ∈ E}; Among them, + means that when the value inside [] is positive, the value inside [] is taken; when [] is negative or 0, 0 is taken; γ is a hyperparameter with a trend greater than 0, representing the interval correction distance between a correct triple and an incorrect triple. The correct triple is the known true relationship in the post-course knowledge graph, and the vector distance calculated by it is expressed as d9h+l,t); the incorrect triple is an unreasonable relationship generated by replacing the head entity or the tail entity in the correct triple, and the vector distance calculated by it is expressed as d(h′,l,t′); S′ is a negative sample set; (h′,l,t) and (h,l,t′) represent triples generated by random recombination. (h′, l, t) represents replacing the original head entity with another head entity h′ while keeping the relation name and the tail entity unchanged; (h, l, t′) represents replacing the original tail entity with another tail entity t′ while keeping the relation name and the head entity unchanged. ′ ; The purpose of model training is to maximize the distance between positive samples and negative samples, making the relation mapping of the knowledge graph more accurate; the positive samples are the set of correct triples, and the negative samples are the set of incorrect triples.

10. A knowledge graph-based recruitment position course system recommendation system, applicable to the knowledge graph-based recruitment position course system recommendation method according to claim 9, characterized in that: The system includes a data acquisition module, a knowledge extraction and knowledge graph construction module, a post-course knowledge graph fusion module, a post-course recommendation model construction module, and a post-course recommendation query module; the data acquisition module is used to obtain recruitment position information for different positions and teaching resources of different professional curriculum systems through third-party data sources; the knowledge extraction and knowledge graph construction module is used to construct a position knowledge graph and a course knowledge graph by combining the data collected by the data acquisition module; the post-course knowledge graph fusion module is used to vectorize and fuse the position knowledge graph and the course knowledge graph to form a complete post-course knowledge graph; the post-course recommendation model construction module is used to construct a post-course recommendation model by combining the post-course knowledge graph; the post-course recommendation query module is used to provide the user with a recommended curriculum system corresponding to the input position.

Images