Human resources matching system based on big data

Abstract

The embodiment of the present application provides a human resources matching system based on big data, including: a construction module for performing entity recognition on a target text block using a rule library to obtain an initial entity; and performing entity recognition on the target text block using a first machine learning model, and adjusting the recognition process of the first machine learning model using the initial entity to obtain a target entity; building a graph database based on the target entity; the target text block includes a text block after the job application text is segmented and a text block after the recruitment text is segmented. A search module is used to generate search graph data based on user input operations and the graph data corresponding to the user. A matching module is used to calculate the similarity between the search graph data and the graph data in the graph database, and select matching graph data from the graph database based on the similarity. The fusion of entity recognition based on rules and contextual information is achieved in this way, which can improve the accuracy of entity recognition by the machine learning model.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a human resources matching system based on big data. Background Art

[0002] In recent years, the application of advanced technologies such as big data analysis and artificial intelligence has revolutionized the field of human resources management. In particular, intelligent matching systems, leveraging algorithms to automatically analyze the compatibility between applicants' abilities and experience and the requirements of available positions, have enabled accurate and efficient talent recommendations.

[0003] However, in the existing technology, a large amount of training data is needed to train the algorithm model to improve the accuracy of the algorithm model's data processing. However, it is difficult to obtain a large amount of labeled training data, which leads to the low accuracy of the data processing of the existing algorithm model. Summary of the Invention

[0004] The embodiment of the present application provides a human resources matching system based on big data to achieve the effect of improving the accuracy of the algorithm model.

[0005] Some embodiments of the present application provide a human resources matching system based on big data, characterized in that the human resources matching system includes a backend server and a front-end application end, and the backend server includes:

[0006] A construction module is configured to perform entity recognition on a target text block using a rule library to obtain an initial entity; perform entity recognition on the target text block using a first machine learning model, and adjust an intermediate recognition result using the initial entity during the recognition process of the first machine learning model to obtain a target entity; and construct a graph database based on the target entity; wherein the target text block includes a text block obtained by segmenting the job application text and a text block obtained by segmenting the recruitment text;

[0007] The search module is used to generate search graph data based on the user's input focus words and the user's corresponding graph data;

[0008] The matching module is used to calculate the similarity between the search graph data and the graph data to be selected in the graph database, and select the matching graph data from the graph database based on the similarity.

[0009] In the above technical solution, the human resources matching system includes a construction module, a search module, and a matching module. The construction module obtains a graph database by processing the target text block, the search module is used to generate search graph data, and the matching module is used to select graph data that best matches the search graph data from the graph database. Since job application texts and recruitment texts usually use more specialized terms and have relatively simple sentence structures, the construction module can first perform initial recognition of the target text block using rules to obtain initial entities; then, the target text block is input into the first machine learning model for recognition, and the ability of the first machine learning model to recognize information in combination with contextual information is used to recognize the target text block. The initial entity is used to adjust the recognition process of the first machine learning model, realizing the fusion of entity recognition based on rules and in combination with contextual information. This can improve the accuracy of entity recognition by the first machine learning model, especially in the case of insufficient training data, and still ensure the accuracy of entity recognition by the first machine learning model.

[0010] In some possible implementations, the first machine learning model includes a model input layer, a model encoding layer, a first fusion layer, a revised input layer, a revised encoding layer, and a model decoding layer;

[0011] Building blocks are used to:

[0012] Use the model input layer to vectorize the target text block to obtain the main vector; use the multi-layer model encoding layer to encode the main vector multiple times and then output it;

[0013] Use the correction input layer to vectorize the initial entity to obtain a correction vector; use the correction encoding layer to encode the correction vector and then output it;

[0014] Use the first fusion layer to fuse the data output by the multi-layer model coding layer and the data output by the correction coding layer;

[0015] The multi-layer model decoding layer is used to perform multiple decoding processes on the data output by the first fusion layer to obtain the target entity.

[0016] In the above technical solution, the correction input layer is used to vectorize the initial entity, the correction encoding layer encodes the vectorized initial entity so as to merge it with the data of the target text block after vectorization and encoding, thereby correcting the encoded data, and the decoding layer decodes the corrected encoded data and outputs the target entity, thereby improving the accuracy of entity recognition by the first machine learning model.

[0017] In some possible implementations, the first machine learning model further includes at least one second fusion layer, and the building module is configured to:

[0018] Use the i-th second fusion layer to perform weighted averaging on the output data of the corrected coding layer of the i+1-th layer and the output data of the model coding layer of the i+1-th layer;

[0019] The model encoding layer of the i-th last layer is used to encode the data output by the i-th second fusion layer and then output; wherein the weighting coefficient of the second fusion layer is adjustable.

[0020] In the above technical solution, in order to further improve the initial entity's ability to correct the first machine learning model, the first machine learning model is also provided with multiple second fusion layers. The second fusion layer is used to fuse the data output by the model coding layer and the data output by the correction coding layer and output them to the next model coding layer. The weighting coefficient of the second fusion layer is adjustable. By adjusting the size of the weighting coefficient, the initial entity's ability to correct the first machine learning model can be adjusted, thereby further improving the accuracy of data processing.

[0021] In some possible implementations, the human resource matching system includes multiple graph databases, each graph database corresponding to an industry;

[0022] The search module is used to receive the user's input of the focus word, obtain the user's corresponding graph database, use the focus word to modify the weight of the user's corresponding graph data, and generate search graph data;

[0023] The matching module is used to determine the search industry corresponding to the search graph data, calculate the similarity of each node between the search graph data and the graph data to be selected in the graph database corresponding to the search industry, and determine the similarity of the graph data.

[0024] In the above technical solution, the focus words are determined based on the focus words input by the user, and the focus words are used to correct the weights in the graph data corresponding to the user. In this way, when calculating the similarity between the search graph data and the graph data in the graph database, the similarity of the graph data matching the focus words is increased, ensuring that the graph data selected from the graph database has the highest matching degree with the focus words, thereby improving the matching accuracy.

[0025] In some possible implementations, the backend server receives an initial screening instruction sent by the frontend application, the initial screening instruction including an initial screening keyword, generates a recommended keyword based on the initial screening keyword, and sends the recommended keyword to the frontend application;

[0026] The backend server receives a progressive screening instruction sent by the front-end application, the progressive screening instruction including a progressive keyword, and generates a focus word according to the initial screening keyword and the progressive keyword.

[0027] In the above technical solution, recommended keywords are generated based on the keywords entered by the user for the first time, and the user is guided to enter again to obtain more filtering information from the user. In this way, the graph data in the graph database can be matched based on more filtering information, so that the matched graph data can be more in line with the user's filtering needs, thereby improving the human resources matching system's ability to match needs.

[0028] In some possible implementations, the building blocks are specifically used to:

[0029] Use the second machine learning model to extract the current year's project plan from the enterprise development report, generate a talent recruitment plan based on the current year's project plan and the existing talent pool, and update the user's corresponding graph data based on the talent recruitment plan;

[0030] The search module generates search graph data based on the focus words input by the user and the updated graph data.

[0031] In the above technical solution, after obtaining the enterprise development report, the machine learning model is triggered to extract the current year's project plan from the enterprise development report, and then a talent recruitment plan is generated based on the current year's project plan and the existing talent pool. This can automatically update the user's corresponding graph data based on the talent recruitment plan, and then trigger the search module to update the search graph data, and obtain the graph data of job seekers that match the end-of-year project plan from the job search graph database, thereby improving data matching efficiency.

[0032] In some possible implementations, the existing talent pool includes at least one piece of existing talent information, and each piece of existing talent information includes mastered skills and project experience;

[0033] The search module is specifically used to:

[0034] Use the second machine learning model to extract talent demand information from this year's project plan and generate a talent demand database. The talent demand information includes skill requirements and project requirements.

[0035] For each required talent information, match the required talent information with the existing talent information in the existing talent pool, determine the talent gap information based on the matching results, and use the talent gap information as a recruitment plan.

[0036] In the above technical solution, a second machine learning model is used to extract required talent information from this year's project plan, and the required talent information is matched with the existing talent information. A recruitment plan is generated based on the matching results, thereby ensuring that the recruited talents can meet this year's project plan and improving the human resources matching system's ability to match demand.

[0037] This embodiment provides a human resource matching method based on big data, including:

[0038] Performing entity recognition on a target text block using a rule library to obtain an initial entity; performing entity recognition on the target text block using a first machine learning model, and adjusting an intermediate recognition result using the initial entity during the recognition process of the first machine learning model to obtain a target entity; constructing a graph database based on the target entity; wherein the target text block includes a text block obtained by segmenting a job application text and a text block obtained by segmenting a recruitment text;

[0039] Generate search graph data based on the user's input focus words and the user's corresponding graph data;

[0040] The similarity between the search graph data and the graph data to be selected in the graph database is calculated, and matching graph data is selected from the graph database according to the similarity.

[0041] The embodiment of the present application provides a backend server, comprising: a memory, a processor;

[0042] Memory stores computer-executable instructions;

[0043] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the implementation methods involved in the above embodiments.

[0044] An embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the implementation methods involved in the above embodiments.

[0045] An embodiment of the present application provides a computer program product, including a computer program, which implements the implementation methods involved in the above embodiments when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0047] Figure 1 A structural block diagram of a big data-based human resources matching system provided in some embodiments of the present application;

[0048] Figure 2 A schematic diagram of job search graph data provided in some embodiments of the present application;

[0049] Figure 3 A schematic diagram of recruitment graph data provided in some embodiments of the present application;

[0050] Figure 4 A schematic diagram of the architecture of a first machine learning model provided in some embodiments of the present application;

[0051] Figure 5 A schematic diagram of the architecture of a first machine learning model provided in some embodiments of the present application;

[0052] Figure 6 A schematic diagram of the architecture of a backend server provided for some embodiments of the present application.

[0053] Reference numerals:

[0054] 100, backend server; 110, construction module; 120, search module; 130, matching module; 200, front-end application.

[0055] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0056] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0057] In recent years, the application of advanced technologies such as big data analysis and artificial intelligence has revolutionized the field of human resources management. In particular, intelligent matching systems, leveraging algorithms to automatically analyze the compatibility between applicants' abilities and experience and the requirements of available positions, have enabled accurate and efficient talent recommendations.

[0058] In the existing technology, a large amount of training data is needed to train the algorithm model to improve the accuracy of the algorithm model's data processing. However, it is difficult to obtain a large amount of labeled training data, which leads to low accuracy of data processing by the existing algorithm model.

[0059] This application provides a human resources matching system based on big data. Due to the large number of specialized terms and simple sentence structures in job application and recruitment documents, some corpora are suitable for entity recognition using rule bases. In view of this, by integrating the entity recognition of machine learning models with the entity recognition of rule bases, entity recognition based on rules and contextual information can be achieved. In the case of insufficient training data, the accuracy of entity recognition performed by machine learning models can be improved.

[0060] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0061] Figure 1 This is a structural diagram of a human resources matching system based on big data provided by some embodiments of this application. Figure 1 As shown, the human resources matching system includes a backend server 100 and a frontend application 200 . The backend server 100 includes a construction module 110 , a search module 120 and a matching module 130 .

[0062] Construction module 110 is configured to perform entity recognition on the target text block using a rule library to obtain initial entities. Entity recognition is performed on the target text block using a first machine learning model. During the recognition process of the first machine learning model, the initial entities are used to adjust the intermediate recognition results to obtain target entities. A graph database is constructed based on the target entities. The target text blocks include text blocks obtained by segmenting the job application text and the recruitment text.

[0063] The search module 120 is used to generate search graph data based on the focus word input by the user and the graph data corresponding to the user.

[0064] The matching module 130 is used to calculate the similarity between the search graph data and each candidate graph data in the graph database, and select matching graph data from the graph database based on the similarity.

[0065] More specifically, construction module 110 is configured to collect resume information and pre-process it to obtain a job application text. In some examples, construction module 110 obtains user-submitted resume information, removes irrelevant information from the resume, such as headers and footers, and converts the resume information into a format to obtain a job application text.

[0066] Subsequently, the construction module 110 divides the job application text into text blocks corresponding to personal basic information, text blocks corresponding to educational background, text blocks corresponding to work experience, text blocks corresponding to skill certificates, text blocks corresponding to job application information, and text blocks corresponding to project experience.

[0067] For each of the above text blocks, the rule library is used to identify the text block and obtain initial entities. For example, the text block is segmented and the segmented text is matched with words in the dictionary to obtain initial entities. The first machine learning model is used to perform entity recognition on the text block. During the recognition process of the first machine learning model, the initial entities are used to adjust the intermediate recognition results to obtain the target entity.

[0068] The job search entity includes basic personal information, job search information, education background, work experience, project experience, and skill certificates. Basic personal information includes name, gender, age, and contact information. Job search information includes job title, work location, and salary range. Education background includes school name, major, degree, and graduation date. Work experience includes company name, position, work hours, and work content. Project experience includes project name and content. Skill certificates include skills mastered and certificates obtained.

[0069] The construction module 110 generates job-seeking graph data based on the extracted job-seeking entities. Figure 2 A schematic diagram of job search graph data provided in some embodiments of the present application, such as Figure 2 As shown, the job search node includes personal information, job search information, education background, work experience, project experience, and skill certificates. Each job search node is connected to the job search graph data by an edge. Each edge has a weight value, which can be set to the default value or modified as required.

[0070] Name, gender, age, and contact information are attributes of the basic personal information node. Job title, work location, and salary range are attributes of the job application information node. School name, major, education level, and graduation date are attributes of the education background node. Company name, position, work hours, and work content are attributes of the work experience node. Project name and project content are attributes of the project experience node. Skills mastered and certificates obtained are attributes of the skill certificate node.

[0071] Construction module 110 is specifically configured to collect recruitment information and pre-process it to obtain recruitment text. More specifically, construction module 110 collects recruitment information from recruitment websites and company websites. It removes redundant and irrelevant information, such as advertisements, from the recruitment information. It then converts the format of the recruitment information to obtain recruitment text.

[0072] Construction module 110 uses a rule library to identify the job text and obtain initial entities. For example, the job text is segmented and matched with words in a dictionary to obtain initial entities. Entity recognition is performed on the job text using a first machine learning model. During the recognition process, the initial entities are used to adjust the intermediate recognition results of the first machine learning model to obtain the target entity.

[0073] Recruitment entities include position information, skill requirements, work experience requirements, project experience requirements, and educational requirements. Position information includes job title, job category, location, salary range, job description, job responsibilities, team description, company name, industry, size, and location. Skill requirements include required skills and certifications. Work experience requirements include the company and required years of experience. Project experience requirements include required project experience. Education requirements include bachelor's and master's degrees.

[0074] The construction module 110 generates recruitment graph data based on the extracted recruitment entities. Figure 3 A schematic diagram of recruitment graph data provided in some embodiments of the present application, such as Figure 3 As shown, the recruitment node includes position information, skill requirements, work experience requirements, project experience requirements, and educational requirements. Each recruitment node is connected to the recruitment graph data by an edge. Each edge has a weight value, which can be set to the default value or modified as required.

[0075] Position title, position category, location, salary range, job description, job responsibilities, team profile, company name, industry, size, and location are attributes of the position information node. Required skills and certifications are attributes of the skill requirements node. Company and required years of work experience are attributes of the work experience requirements node. Project name and content are attributes of the project experience requirements node. Bachelor's and postgraduate degrees are attributes of the education requirements node.

[0076] The search module 120 is used to receive the focus word input by the user, obtain the graph data corresponding to the user, use the focus word to modify the nodes in the graph data corresponding to the user, and obtain search graph data.

[0077] Matching module 130 is configured to traverse the candidate graph data in the graph database and calculate the similarity between the search graph data and the candidate graph data in the graph database. In some examples, candidate graph data with a similarity greater than a preset similarity threshold is selected as the matched graph data output. In other examples, candidate graph data with a higher similarity ranking is selected as the matched graph data output.

[0078] In the above technical solution, the human resources matching system includes a construction module 110, a search module 120, and a matching module 130. The construction module 110 obtains a graph database by processing the target text block, the search module 120 is used to generate search graph data, and the matching module 130 is used to select graph data that best matches the search graph data from the graph database. Since job application texts and recruitment texts usually use more specialized terms and have relatively simple sentence structures, the construction module 110 can first perform initial recognition of the target text block using rules to obtain initial entities; then, the target text block is input into the first machine learning model for recognition, and the ability of the first machine learning model to recognize information in combination with contextual information is used to recognize the target text block. The initial entity is used for adjustment during the recognition process of the first machine learning model to achieve a fusion of rule-based and contextual information recognition technologies. In the case of insufficient training data, the accuracy of entity recognition by the first machine learning model can be improved.

[0079] Specifically, the system uses a rule library to perform initial entity recognition on target text blocks and uses this initial entity recognition to adjust the machine learning model's recognition process, achieving a fusion of rule-based and contextual information. This approach is particularly well-suited to the terminology-rich and simple sentence structures found in job search and recruitment texts, effectively improving entity recognition accuracy and maintaining high recognition precision even when training data is insufficient.

[0080] The system constructs a structured graph database based on the identified target entities, with each entity as a node in the graph and the relationships between entities as edges. This structured data representation facilitates efficient subsequent retrieval and matching operations.

[0081] The search module generates personalized search graph data based on the user's input keywords and the corresponding graph data. By adjusting the weights of relevant nodes in the user graph data, the system can more accurately reflect the user's search intent, thereby improving the personalization and relevance of the matching.

[0082] The matching module calculates the similarity between the search graph data and the candidate graph data in the graph database, quickly screening the options that best match the user's needs. This approach not only improves matching efficiency but also ensures the accuracy and relevance of matching results, providing users with more precise job or talent recommendations.

[0083] The system supports dynamic updates of the graph database based on corporate development reports, ensuring data timeliness and accuracy. Furthermore, the system can continuously expand the graph database content based on new recruitment needs and job search information, improving the system's scalability and practicality.

[0084] Through precise matching and recommendation functions, the system can help companies quickly find suitable talent, reducing recruitment costs and time. It also provides features such as talent gap analysis, providing strong support for companies' human resource planning and optimizing recruitment and decision-making processes.

[0085] Furthermore, by providing personalized search and matching services, the system significantly improves the user experience for both job seekers and recruiters. Job seekers can more quickly find positions that meet their expectations, while companies can more easily find talent with the skills and experience they need, thereby increasing satisfaction and collaboration efficiency for both parties.

[0086] In some possible implementations, the human resources matching system further includes a rule library comprising multiple dictionaries, one for each industry. Construction module 110 is further configured to obtain a dictionary for the industry to which the target text block belongs. The target text block is segmented, and the segmented corpus is matched with words in the dictionary for the industry to which the target text block belongs to obtain initial entities.

[0087] Among them, when obtaining job application information, the company where the applicant once worked is extracted from the job application information, and the industry to which the job application information belongs is determined based on the industry to which the company where the applicant once worked belongs, and then the industry corresponding to the job application text is obtained. The construction module 110 divides the job application text into blocks to obtain text blocks corresponding to personal basic information, text blocks corresponding to educational background, text blocks corresponding to work experience, text blocks corresponding to skill certificates, text blocks corresponding to job application information, and text blocks corresponding to project experience. For each of the above text blocks, the corpus in each text block is segmented, and the segmentation results are matched with the words in the dictionary library of the industry to which the text block belongs. If they match, the matched segmentation results are used as the initial entity. For example: the text block corresponding to the skill certificate is segmented to obtain one of the segmentation results as Java. Java is the same as the word in the dictionary library of the computer industry, so Java is used as an initial entity.

[0088] In the above technical solution, by classifying the dictionaries in the rule base according to industry, the construction module 110 obtains the industry corresponding to the target text block when using the rule base to identify the target text block, and then focuses on the dictionary of the corresponding industry. Using the dictionary of the corresponding industry for identification can reduce the amount of data processing and improve data processing efficiency.

[0089] In some possible implementations, the human resources matching system includes multiple graph databases, each corresponding to an industry. The search module 120 is configured to receive a focus word input by a user, obtain the graph database corresponding to the user, modify the weight of the graph database corresponding to the user using the focus word, and generate search graph data.

[0090] The matching module 130 is used to determine the industry corresponding to the search graph data, use the industry corresponding to the search graph data as the search industry, traverse the graph data to be selected in the graph database corresponding to the search industry, calculate the similarity of corresponding nodes between the search graph data and the graph data to be selected in the graph database of the search industry, and calculate the graph data similarity according to the first formula;

[0091] The first formula specifically includes:

[0092]

[0093] Among them, x1 represents the similarity of the first node, w 11 Represents the weight of the first node in the search graph data, w 21 represents the weight of the first node of the selected graph data; x2 represents the similarity of the first node, w 12 Represents the weight of the second node in the search graph data, w 22 Indicates the weight of the second node in the graph data to be selected.

[0094] It's worth noting that the first formula above calculates the similarity of each node between the search graph data and the candidate graph data in the graph database and takes a weighted average of these similarities to arrive at the similarity of the entire graph data. This formula comprehensively considers the matching of multiple nodes and, compared to the matching of a single node, better reflects the overall matching level of the graph data, thereby improving matching accuracy.

[0095] The weight terms in the first formula (e.g., the weight of the first node in the search graph, the weight of the first node in the candidate graph, etc.) allow the system to flexibly adjust the importance of each node based on actual needs. For example, if a user is particularly interested in a node (such as work location or salary range), the weight of that node can be increased to make the matching results more in line with the user's expectations.

[0096] Furthermore, human resource matching often involves multiple dimensions of information, such as personal information, educational background, work experience, and skill requirements. The first formula calculates the weighted average of the similarities across each node, achieving a comprehensive assessment of this multi-dimensional information. This allows for a more comprehensive reflection of the match between job seekers and recruiters, improving the comprehensiveness and accuracy of matching.

[0097] Furthermore, by calculating the similarity between graph data, the system can quickly select the candidate graph data that best matches the search graph data. This approach avoids traditional traversal searches, reduces computational effort, and improves search efficiency. Furthermore, due to the high matching accuracy, the system can more quickly find matching results that meet user needs.

[0098] In the following example, a user who is a job seeker is retrieved from the job seeker graph database. This includes personal information, job search information, educational background, work experience, and skill certifications. If the user focuses on the job location in the job search information, the weight of this job search information in the user graph data is increased, thus generating search graph data. The candidate graph data is the recruitment graph data, which includes position information, skill requirements, work experience requirements, and educational requirements.

[0099] The job search information nodes in the search graph data correspond to the job position information nodes in the candidate graph data. The job position similarity between the job search information and the job position information in the candidate graph data is calculated. The skill certificate nodes in the search graph data correspond to the skill requirement nodes in the candidate graph data. The skill similarity between the skill certificate in the search graph data and the skill requirement in the candidate graph data is calculated. The work experience nodes in the search graph data correspond to the work experience requirement nodes in the candidate graph data. The experience similarity between the work experience in the search graph data and the work experience requirement in the candidate graph data is calculated. The education background nodes in the search graph data correspond to the educational requirement nodes in the candidate graph data. The educational similarity between the education background in the search graph data and the educational requirement in the candidate graph data is calculated.

[0100] Calculate the average weight of the job search information node in the user graph data and the weight of the position information node in the recruitment graph data As the weight of the job similarity. Calculate the average weight of the skill certificate node in the user graph data and the weight of the skill requirement node in the recruitment graph data As the weight of skill similarity. Calculate the average weight of the work experience node in the user graph data and the weight of the work experience requirement node in the recruitment graph data As the weight of experience similarity. Calculate the average weight of the education requirement node in the recruitment graph data and the weight of the education background node in the search graph data As the weight of the educational similarity, calculate the weighted average of the similarities of the above four nodes to obtain the similarity between the search graph data and the candidate graph data.

[0101] In the above technical solution, the focus word is determined based on the focus word input by the user, and the focus word is used to correct the weight in the graph data corresponding to the user. In this way, when calculating the similarity between the search graph data and the graph data to be selected in the graph database, the similarity of the graph data matching the focus word is increased, ensuring that the graph data selected from the graph database has the highest matching degree with the focus word, thereby improving the matching accuracy.

[0102] Specifically, the system can establish multiple graph databases based on industry segmentation. This segmentation strategy allows each graph database to focus on human resources data for a specific industry, thereby improving the targetedness and accuracy of matching. When a user searches, the system can quickly locate the graph database related to the user's industry, laying a solid foundation for the subsequent matching process.

[0103] The search module receives user-entered keywords and modifies the weights of the corresponding nodes in the user's graph data based on these keywords. This mechanism assigns higher weights to nodes highly correlated with the user's keywords, further emphasizing their contribution when calculating similarity. This weight adjustment strategy significantly enhances the relevance of matching results to user needs.

[0104] The matching module calculates the similarity between the search graph data and the candidate graph data using the first formula. This formula considers the similarity of each node and its corresponding weight, comprehensively evaluating the similarity between the graph data through a weighted average. This refined calculation method more accurately reflects the actual differences and similarities between the graph data, thereby improving matching accuracy.

[0105] Furthermore, because focus terms and weights are modified based on real-time user input, the system has strong dynamic adjustment capabilities. Regardless of how user needs change, the system can quickly respond and adjust matching strategies to ensure that users are provided with the human resource matching results that best meet their needs.

[0106] By implementing the above technical solutions, the system can provide users with more personalized and accurate human resource matching services. This not only improves users' search efficiency, but also greatly enhances their satisfaction and trust in the system's matching results.

[0107] Furthermore, the introduction of multiple graph databases and their segmentation strategy enable the system to more efficiently manage and allocate resources. During the search and matching process, the system can quickly locate relevant graph databases, avoiding unnecessary resource waste and computational overhead. This optimization strategy improves the overall efficiency and responsiveness of the system.

[0108] In some possible implementations, the matching module is used to:

[0109] Generate an academic degree conversion database based on the academic degree information of existing talents in the existing talent database, and use the academic degree conversion database to train a vector conversion model so that each academic degree information in the academic degree conversion database is converted into the same vector;

[0110] Generate a work experience conversion database based on the work experience information of existing talents in the existing talent pool, and use the work experience conversion database to train a vector conversion model so that each work experience information in the work experience conversion database is converted into the same vector;

[0111] The search graph data will be converted into a search vector using the vector conversion model, and the candidate graph data will be converted into a candidate vector using the vector conversion model;

[0112] Calculate the Euclidean distance between the search vector and the candidate vector to obtain the similarity between the search graph data and the candidate graph data.

[0113] For each enterprise user, we obtain information about existing talent, such as the talent's graduate school and previous work experience, and add the talent's graduate school to the enterprise user's education conversion database. We use this database to train a vector conversion model, mapping each graduate school in the database to the same vector. This allows us to calculate the similarity between any two graduate schools in the database when calculating the similarity between educational background nodes in two graphs.

[0114] Add the companies where the existing talent has worked to the company user's work experience conversion database. Use this work experience conversion database to train a vector conversion model so that each company in the work experience conversion database is mapped to the same vector. This way, when calculating the similarity between work experience nodes in two graphs, the similarity between any two companies in the work experience conversion database is high.

[0115] The job search nodes in the search graph correspond to the job position nodes in the candidate graph. A vector conversion model is used to convert the job search nodes in the search graph into job search vectors. Similarly, a vector conversion model is used to convert the job position nodes in the candidate graph into job candidate vectors. The Euclidean distance between the job search vectors and the job candidate vectors is calculated as the job similarity.

[0116] The skill certificate nodes in the search graph data correspond to the skill requirement nodes in the candidate graph data. A vector conversion model is used to convert the skill certificate nodes in the search graph data into skill search vectors. A vector conversion model is used to convert the skill requirement nodes in the candidate graph data into skill candidate vectors. The Euclidean distance between the skill search vector and the skill candidate vector is calculated as the skill similarity.

[0117] The work experience nodes in the search graph data correspond to the work experience requirement nodes in the candidate graph data. A vector conversion model is used to convert the work experience nodes in the search graph data into experience search vectors. A vector conversion model is used to convert the work experience requirement nodes in the candidate graph data into experience candidate vectors. The Euclidean distance between the experience search vector and the experience candidate vector is calculated as the experience similarity.

[0118] The educational requirement nodes in the candidate graph data correspond to the educational background nodes in the search graph data. A vector conversion model is used to convert the educational background nodes in the search graph data into educational search vectors. A vector conversion model is used to convert the educational requirement nodes in the candidate graph data into educational candidate vectors. The Euclidean distance between the educational search vector and the educational candidate vector is calculated as the educational similarity.

[0119] Calculate the weighted average of the similarities of the above four nodes to obtain the similarity between the search graph data and the candidate graph data.

[0120] In the above technical solution, by collecting educational information from the existing talent pool, determining the schools that the company tends to recruit, collecting work experience from the existing talent pool, determining the companies that the company tends to recruit, and using the above information to train the vector conversion model, when calculating the similarity between the search graph data and the selected graph data, that is, when calculating the job seeker's corresponding graph data and the recruiter's corresponding graph data, if the job seeker's educational experience is the same as the recruiter's preferred recruiting school, the job seeker's corresponding graph data and the recruiter's corresponding graph data are the same after conversion, the Euclidean distance between the two is also the smallest, and the similarity is the highest. In this way, the best job seeker can be matched for the company, and the matching accuracy of the human resources matching system is improved.

[0121] Specifically, the system collects educational background information and work experience information from the existing talent pool to generate a separate educational background conversion database and a work experience conversion database. These two conversion databases enable the previously complex and diverse educational background and work experience information to be uniformly converted into vector form. This unified data representation not only simplifies the data processing process but also significantly improves matching efficiency, as comparisons and calculations between vectors are more efficient than with raw text or structured data.

[0122] The academic qualification and work experience conversion databases are constructed based on actual data from existing talent pools. This means they accurately reflect companies' hiring preferences, such as the schools and companies they favor. By training the vector conversion model, the system learns these preferences and utilizes them in subsequent matching processes, enabling more accurate recommendations for candidates that meet companies' requirements.

[0123] When calculating the similarity between the search graph data and the candidate graph data, the system uses Euclidean distance as a metric. Because educational background and work experience information have been converted into vectors in the same vector space through a vector conversion model, Euclidean distance accurately reflects the degree of similarity between these vectors. When a job seeker's educational background matches the institution preferred by the recruiter, or their work experience matches the company preferred by the recruiter, their corresponding vectors will be very close, with the minimum Euclidean distance and the highest similarity. This matching method significantly improves the accuracy of the human resources matching system.

[0124] Furthermore, the introduction of the vector transformation model makes the system more adaptable and scalable. As companies' recruitment needs change and their talent pools are constantly updated, the system can adapt to new data environments by retraining or adjusting the vector transformation model. This flexibility enables the system to maintain efficient and accurate matching capabilities.

[0125] By implementing the above technical solution, the system can provide companies with more accurate and targeted job seeker recommendations that better meet their needs. This not only improves recruitment efficiency but also reduces recruitment costs. Furthermore, job seekers can find suitable job opportunities more quickly.

[0126] In some possible implementations, the first machine learning model includes a model input layer, a model encoding layer, a first fusion layer, a modified input layer, a modified encoding layer, and a model decoding layer. Construction module 110 is used to:

[0127] Use the model input layer to vectorize the target text block to obtain the main vector; use the multi-layer model encoding layer to encode the main vector multiple times and then output it;

[0128] Using a correction input layer to vectorize the initial entity to obtain a correction vector; using at least one correction encoding layer to encode the correction vector and then output it;

[0129] Using a first fusion layer to fuse data output by a multi-layer model coding layer and data output by at least one layer of correction coding layer;

[0130] The multi-layer model decoding layer is used to perform multiple decoding processes on the data output by the first fusion layer to obtain the target entity.

[0131] More specifically, Figure 4 A schematic diagram of the architecture of a first machine learning model provided in some embodiments of the present application, such as Figure 4 As shown, the model input layer receives the target text block and vectorizes it into a first vector. The first model encoding layer receives the model input layer output vector, encodes it, and outputs it to the second model encoding layer. The second model encoding layer encodes the data received at its input and outputs it. This continues in this order until the last model encoding layer encodes the data received at its input and outputs it.

[0132] The rule base is used to perform entity recognition on the target text block to obtain the initial entity. The correction input layer receives the initial entity, vectorizes the initial entity and outputs it to the first correction coding layer. The first correction coding layer encodes the data received at its input end and outputs it to the second correction coding layer. The second correction coding layer encodes the data received at its input end and outputs it to the third correction coding layer, until the last correction coding layer encodes the data received at its input end and outputs it.

[0133] The first fusion layer receives the data output by the last model encoding layer and the last correction encoding layer. It fuses the output data of the last model encoding layer and the last correction encoding layer, and then outputs it to the first model decoding layer. The second model decoding layer receives the data output by the first model decoding layer, decodes the output data of the second model decoding layer, and then outputs it. This continues until the last model decoding layer decodes and outputs the data output by the second-to-last model decoding layer.

[0134] Each model encoding layer has the same structure, as does each revised encoding layer. Both the model encoding layer and the revised encoding layer have the same structure, including a multi-head self-attention mechanism and a feedforward neural network. The model decoding layer includes an encoder-decoder attention mechanism and a feedforward neural network. The first fusion layer takes a weighted average of the output data of the last model encoding layer and the output of the revised encoding layer.

[0135] In the above technical solution, the correction input layer is used to vectorize the initial entity, the correction encoding layer encodes the vectorized initial entity so that it can be fused with the data of the target text block after vectorization and encoding, thereby correcting the encoded data, and the decoding layer decodes the corrected encoded data and outputs the target entity, thereby realizing the fusion of rule-based and contextual information recognition technology, thereby improving the accuracy of entity recognition by the machine learning model.

[0136] Specifically, the model input layer first converts the target text block into a vectorized representation (primary vector), which is then deeply encoded through a multi-layer model encoding layer. This process captures the deep semantic information in the text and provides rich feature representations for subsequent entity recognition.

[0137] The revised input layer vectorizes the initial entity to generate a revised vector, which is then encoded by the revised encoding layer. The introduction of the revised vector provides the model with additional context or domain knowledge, helping the model to strengthen this key information during the encoding process, thereby improving the accuracy of entity recognition.

[0138] The first fusion layer fuses the data output from the multi-layer model's encoding layer with the data output from the correction encoding layer. This fusion strategy combines the global features automatically learned by the model with the local features acquired through the rule base, leveraging the complementary strengths of the two different recognition paths. Adjusting the fusion weights can further optimize the model's recognition performance, particularly when processing domain text with specific patterns or terminology.

[0139] The fused data is decoded through a multi-layer model decoding layer, gradually recovering the specific form of the target entity. This multi-layer decoding structure not only improves decoding accuracy but also enhances the model's robustness and generalization capabilities. Even when faced with unseen entities or new text patterns, the model's powerful encoding and decoding capabilities enable it to effectively handle them.

[0140] Furthermore, the first machine learning model utilizes a modular design, allowing each layer to be trained and optimized independently. This design reduces model complexity and improves training efficiency. Furthermore, by adjusting the parameters and structure of each layer, the model can be flexibly tuned to suit different application scenarios and requirements.

[0141] Furthermore, by combining a rule base with a machine learning model, the system can achieve better recognition results even with limited training data. This hybrid approach leverages the accuracy and stability of the rule base with the adaptability and generalization capabilities of the machine learning model.

[0142] Furthermore, in different fields or application scenarios, trained model parameters can be transferred to new tasks through methods such as transfer learning, thereby achieving rapid transfer and sharing of knowledge. This significantly reduces the development cost and time of new tasks.

[0143] In some possible implementations, the construction module 110 is configured to:

[0144] Use the i-th second fusion layer to perform weighted averaging on the output data of the corrected coding layer of the i+1-th layer and the output data of the model coding layer of the i+1-th layer;

[0145] The model encoding layer of the i-th layer from the end is used to encode the data output by the i-th second fusion layer and then output it; wherein the weighting coefficient of the second fusion layer is adjustable.

[0146] More specifically, Figure 5 This is a schematic diagram of the architecture of the machine learning model provided in some embodiments of the present application, and Figure 4The difference is that the machine learning model also includes at least one second fusion layer, and each second fusion layer is used to fuse a correction coding layer output data and a model coding layer output data. The number of second fusion layers and the number of correction coding layers differ by one, and the number of second fusion layers is less than half the number of model coding layers. For example: there are two second fusion layers, and the second-to-last second fusion layer performs weighted averaging processing on the correction coding layer output data of the third-to-last layer and the model coding layer output data of the third-to-last layer, and outputs it to the model coding layer of the second-to-last layer. The first-to-last second fusion layer performs weighted averaging processing on the correction coding layer output data of the second-to-last layer and the model coding layer output data of the second-to-last layer, and outputs it to the model coding layer of the first-to-last layer.

[0147] The weighting coefficient of the second fusion layer ranges from [0 to 1]. The weighting coefficient of the second fusion layer includes the weighting coefficient of the model coding layer and the weighting coefficient of the correction coding layer. When the weighting coefficient of the correction coding layer is relatively small, the correction effect of the initial entity on the processing of the model coding layer is reduced. When the weighting coefficient of the correction coding layer is relatively large, the correction effect of the initial entity on the processing of the model coding layer is increased. The correction effect of the initial entity on the processing of the machine learning model can be adjusted by adjusting the weighting coefficient of the second fusion layer.

[0148] In the above technical solution, in order to further improve the initial entity's ability to correct the machine learning model, the machine learning model is also provided with multiple second fusion layers. The second fusion layer is used to fuse the data output by the model coding layer and the data output by the correction coding layer and output them to the next model coding layer. The weight coefficient of the second fusion layer has a value range of [0,1]. By adjusting the size of the weight coefficient, the initial entity's ability to correct the machine learning model can be adjusted, thereby further improving the accuracy of data processing. By making the number of second fusion layers and the number of correction coding layers differ by one, and the number of second fusion layers is less than half the number of model coding layers, the correction effect of the initial entity on the processing process of the machine learning model can be reduced, avoiding excessive correction effect that causes a decrease in data recognition accuracy.

[0149] Specifically, by adding a second fusion layer, the system can more meticulously and flexibly fuse the output data of the model encoding layer and the correction encoding layer. These fusion layers not only promote the deep integration of the features automatically learned by the model with the initial entity features provided by the rule base, but also further optimize the fusion process through weighted averaging, allowing the model to more accurately capture and utilize key information in the text.

[0150] Furthermore, the weighting coefficients of the second fusion layer are adjustable, meaning the system can dynamically adjust the strength of the initial entity's model correction based on actual needs and performance. If stronger correction capability is required, the weight of the corrected encoding layer output data can be increased; otherwise, it can be reduced. This flexibility allows the model to better adapt to different application scenarios and data characteristics.

[0151] The use of a multi-layer fusion strategy (including first and second fusion layers) enhances the model's robustness and generalization. By fusing data and extracting features at different levels, the model can more comprehensively understand text content and reduce recognition errors caused by data bias or noise. This design also improves the model's ability to handle new data or unseen entities.

[0152] The second fusion layer feeds the output data of the corrected encoding layer back to the model encoding layer, forming a closed loop of knowledge transfer. This design promotes knowledge sharing and deep learning across different layers of the model, enabling the model to more effectively leverage the contextual or domain knowledge provided by the initial entities to optimize recognition results. Adjustable weighting coefficients provide more freedom for model training. During training, these coefficients can be adjusted to optimize model performance, such as accuracy and recall. Furthermore, the multi-layer fusion structure enables the model to converge to the optimal solution more stably during training. Through multiple fusion and encoding processes, the model can more accurately identify entities in text. This multi-level, multi-stage processing approach helps eliminate ambiguity and misjudgments during the recognition process, thereby improving the accuracy and reliability of data processing.

[0153] In some possible implementations, the construction module 110 is specifically configured to:

[0154] Use the second machine learning model to extract this year's project plan from the corporate development report, generate a talent recruitment plan based on this year's project plan and the existing talent pool, and update the user's corresponding graph data based on the talent recruitment plan.

[0155] The architecture of the first machine learning model is identical to that of the second machine learning model. The backend server 100 trains the second machine learning model, enabling the geothermal machine learning model to extract the year-end project plan from the enterprise development report. The frontend application 200 removes sensitive information from the enterprise development report and uploads it to the backend server 100. The backend server 100 uses the trained second machine learning model to extract the year-end project plan from the enterprise development report.

[0156] The backend server 100 pre-processes the enterprise development report and removes irrelevant information. Subsequently, the pre-processed enterprise development report is segmented into text blocks to obtain multiple target text blocks. A second machine learning model is used to identify the target text blocks and obtain the year-end project plan.

[0157] Correspondingly, after determining to update the graph data corresponding to the user, the search module generates search graph data according to the focus word input by the user and the graph data corresponding to the user.

[0158] In this solution, the system automatically triggers a second machine learning model to conduct in-depth analysis of the company's development report and accurately extract the project plan for the current year. This automated process not only reduces the workload of the HR department but also avoids the errors and delays that can occur with manual analysis, ensuring the accuracy and timeliness of project plan information.

[0159] Based on the extracted information about the current year's project plan and the existing talent pool, the system can intelligently generate a talent recruitment plan. This plan fully considers the company's current and future project needs, as well as the skills and experience of existing talent, ensuring the recruitment plan is targeted and practical.

[0160] The generation of a talent recruitment plan triggers the system to automatically update the user's corresponding graph data. This dynamic update mechanism ensures that the user graph data is always consistent with the company's current project plans and recruitment needs, providing an accurate data foundation for subsequent search and matching processes.

[0161] The search module quickly generates search graph data based on user-entered keywords and updated graph data. Because the graph data is updated based on the latest recruitment plans, the search process can more accurately locate the graph data of job seekers that match the year-end project plan. This precise matching not only improves search efficiency but also reduces the likelihood of ineffective searches and mismatches.

[0162] Through automated extraction, intelligent generation, and dynamic updates, the system can rapidly respond to changes in corporate development reports and promptly adjust recruitment plans and search strategies. This rapid response and flexible adjustment capability enables the system to better adapt to the changing needs of enterprises, providing more efficient and accurate human resource matching services.

[0163] Subsequently, the backend server 100 matches the talent needed in the current year's project plan with the existing talent in the existing talent pool and generates a talent recruitment plan based on the matching results. The talent recruitment plan includes job postings for multiple positions. The first machine learning model is used to identify target entities from the job postings and the corresponding graph data for the user is updated based on the target entities.

[0164] In the above technical solution, after obtaining the enterprise development report, the second machine learning model is triggered to extract the current year's project plan from the enterprise development report, and then a talent recruitment plan is generated based on the current year's project plan and the existing talent pool. This can automatically update the user's corresponding graph data based on the talent recruitment plan, and then trigger the search module to update the search graph data, and obtain the graph data of job seekers that match the end-of-year project plan from the job search graph database, thereby improving data matching efficiency.

[0165] In some possible implementations, the existing talent pool includes at least one existing talent information, each of which includes skills and project experience. The search module is specifically used to:

[0166] Use the second machine learning model to extract talent demand information from this year's project plan and generate a talent demand database. The talent demand information includes skill requirements and project requirements.

[0167] For each required talent information, match the required talent information with the existing talent information in the existing talent pool, determine the talent gap information based on the matching results, and use the talent gap information as a recruitment plan.

[0168] Among them, matching includes matching of skills and project experience, as well as matching of the number of required talents and the number of existing talents, and generating talent gap information based on the matching results.

[0169] For example, use the second machine learning model to extract a talent pool based on this year's project plan. This year's projects include Project A, which requires two types of talent. The first type requires talent proficiency in hardware development software and prior hardware development project experience. There are two people in this category. The second type requires talent proficiency in Java development and prior Java development project experience. There are three people in this category. The talent pool includes two people in this category and three people in this category.

[0170] By extracting the skills and project experience of existing employees, the company can create an existing talent pool. For example, the existing talent pool includes 20 categories of existing talent. Category A existing talent matches Category 1 talent demand, and there are three Category A existing talents. Category B existing talent matches Category 2 talent demand, and there are two Category B existing talents. The talent gap information indicates a lack of one Category 2 talent demand. Therefore, one Category 2 talent demand needs to be recruited, and the Category 2 talent demand information is used as recruitment information.

[0171] In the above technical solution, a second machine learning model is used to extract required talent information from this year's project plan, match the required talent information with the existing talent information, and generate a recruitment plan based on the matching results. This ensures that the recruited talents can meet this year's project plan and improves the human resources matching system's ability to match demand.

[0172] Specifically, the system leverages a second machine learning model to deeply analyze the annual project plan, accurately identifying and extracting the required skills and project requirements, and forming a talent pool. This step ensures that recruitment plans are based on actual project needs, avoiding blind hiring and wasted resources.

[0173] The search module conducts a comprehensive search and matching of existing talent pools for each required talent. By comparing the skills and project experience of existing talent with the skill and project requirements of the required talent, the system can quickly locate talent that meets or approaches the required requirements, providing strong support for subsequent recruitment plans.

[0174] Through the matching process, the system can clearly identify unmet needs within the existing talent pool, i.e., talent gap information. This information is crucial for developing targeted recruitment plans, as it directly guides the company in identifying the talent with the right skills and experience to recruit, as well as the number and priority of those recruited.

[0175] Based on accurate talent demand and talent gap information, the system can generate highly targeted recruitment plans. This plan not only clarifies recruitment goals and priorities, but also optimizes the allocation of recruitment resources, improving recruitment efficiency and success rates.

[0176] In some possible implementations, the search module 120 is specifically configured to:

[0177] Calculate the skill similarity between the skill requirements of the talent in demand and the skills mastered by existing talents; calculate the project similarity between the project requirements of the talent in demand and the project experience of existing talents;

[0178] Take a weighted average of skill similarity and project similarity to obtain the talent matching degree between the required talents and the existing talents;

[0179] If the talent matching degree is greater than the preset threshold, it is determined that the required talents and the existing talents are matched. If the number of required talents is less than or equal to the number of existing talents, it is determined that there is no gap in this type of required talents; if the number of required talents is greater than the number of existing talents, the gap information of this type of required talents is generated.

[0180] Among them, for each type of required talent, traverse each existing talent information in the existing talent database, convert the required talent information into a required vector, convert the existing talent information into an existing vector, calculate the Euclidean distance between the required vector and the existing vector, obtain the talent matching degree between the required talent and the existing talent, and achieve matching in the skill and project experience dimensions.

[0181] After the required talents are matched with the existing talents in terms of skills and projects, we will continue to determine whether the number of existing talents meets the required talents. If so, there is no gap in this type of talent. If not, the gap information of this type of required talents will be generated.

[0182] In the above technical solution, we first determine whether the existing talents and required talents match in terms of skills and project experience. If they match, we then confirm whether the number of existing talents can meet the number of required talents. Considering these two dimensions, we generate corresponding gap information to ensure that the recruited talents can meet this year's project plan and improve the human resources matching system's ability to match demand.

[0183] Specifically, for each type of talent in demand, the system accurately calculates the skill similarity between the required skills and the skills possessed by existing talent, and simultaneously calculates the project similarity between the project requirements and the project experience of existing talent. This dual matching mechanism ensures a comprehensive assessment of talent matching across two core dimensions: skills and project experience, improving the accuracy and reliability of matching.

[0184] By taking a weighted average of skill similarity and project similarity, the system generates a comprehensive talent matching index. This index comprehensively considers the similarity of skills and project experience, providing a scientific basis for determining whether existing talent meets the requirements of the required talent, avoiding the one-sidedness of single-dimensional evaluation.

[0185] The system intelligently identifies talent gaps based on the degree of talent match and the comparison of the number of required talent with the number of available talent. When the talent match exceeds a preset threshold, the system determines a match. Furthermore, if the number of required talent is less than or equal to the number of available talent, the system determines that there is no gap in that type of talent. Otherwise, gap information is generated. This process ensures the accuracy and timeliness of gap information, providing strong support for the formulation of recruitment plans.

[0186] Accurate talent gap information allows companies to clearly understand the types of talent they are short of, allowing them to adjust recruitment strategies and resource allocation accordingly. This helps avoid wasted resources and blind hiring, and improves recruitment efficiency and effectiveness.

[0187] Clearly, by deeply matching across the two dimensions of skills and project experience, and combining quantitative comparisons to generate gap information, the system has significantly improved its demand-matching capabilities in human resource matching. This enhanced capability enables the system to better meet the actual needs of enterprises and provide more accurate and effective support for their human resource management.

[0188] In some possible implementations, the backend server 100 is configured to receive an initial screening instruction sent by the frontend application 200, the initial screening instruction including an initial screening keyword, generate a recommended keyword based on the initial screening keyword, and send the recommended keyword to the frontend application 200;

[0189] The backend server 100 receives the progressive screening instruction sent by the front-end application 200, where the progressive screening instruction includes a progressive keyword, and generates a focus word according to the initial screening keyword and the progressive keyword.

[0190] The front-end application receives the keywords input by the user, generates an initial screening instruction based on the keywords input by the user, and sends the initial screening instruction to the back-end server 100. After extracting the initial screening keywords from the initial screening instruction, the back-end server 100 inputs the initial screening instruction into an existing recommendation model, causing the existing recommendation model to generate recommended keywords, and sends the recommended keywords to the front-end application 200.

[0191] The recommended keywords are displayed on the front-end application end 200, and the user input is received again. A progressive filtering instruction is generated according to the user input, and the progressive filtering instruction is sent to the back-end server. The focus words are generated according to the initial filtering keywords and the progressive keywords.

[0192] In the above technical solution, recommended keywords are generated based on the keywords entered by the user for the first time, and the user is guided to enter again to obtain more filtering information from the user. In this way, the graph data in the graph database can be matched based on more filtering information, so that the matched graph data can be more in line with the user's filtering needs, thereby improving the human resources matching system's ability to match needs.

[0193] Specifically, the system receives the initial filter keywords entered by the user and quickly generates related recommended keywords. This instant feedback mechanism not only simplifies the user input process but also guides users to more deeply express their needs by providing suggested keywords. This interactive approach enhances user engagement and makes the filter process more intuitive and efficient.

[0194] By introducing progressive filtering commands, the system encourages users to enter multiple entries, with each entry enabling more detailed filtering based on the previous one. This progressive filtering approach gradually accumulates user filtering information, enabling the system to more comprehensively understand user needs and providing a rich information foundation for subsequent graph data matching.

[0195] Furthermore, the system generates focus terms based on the initial screening keywords and the progressive keywords. These focus terms directly reflect user needs. During the graph database matching process, the system prioritizes these focus terms, more accurately locating graph data that meets user needs. This precise matching mechanism improves the accuracy and efficiency of the human resources matching system.

[0196] By gradually accumulating user filtering information, the system can more accurately locate the graph data range that the user is interested in, avoiding unnecessary resource waste. This optimization strategy makes the system more efficient when processing large amounts of data and improves resource utilization.

[0197] Furthermore, because the system can more accurately match user needs, user satisfaction and trust in the system will also increase. Users can more quickly find the human resources information that meets their needs, thereby increasing the practicality and value of the entire system.

[0198] Furthermore, by collecting and analyzing users’ filter instructions and preferred keywords, the system can continuously learn and optimize its recommendation algorithms and matching strategies. This iterative optimization process enables the system to better adapt to users’ changing needs, maintaining its competitiveness and advancement.

[0199] Figure 6 This is a schematic diagram of the structure of the backend server 100 provided in this application. Figure 6 As shown, the electronic device 100 provided in this embodiment includes: at least one processor 101 and a memory 102. Optionally, the device 100 further includes a communication component 103. The processor 101, the memory 102 and the communication component 103 are connected via a bus 104.

[0200] During the specific implementation process, at least one processor 101 executes the computer-executable instructions stored in the memory 102, so that the at least one processor 101 performs the above method.

[0201] The specific implementation process of the processor 101 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0202] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0203] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory.

[0204] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified into address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0205] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0206] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0207] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0208] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0209] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, either through an interface, electrical, mechanical, or other means.

[0210] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0211] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0212] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0213] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented by hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0214] Finally, it should be noted that those skilled in the art will readily devise alternative embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. The present invention is not limited to the precise structure described above and illustrated in the accompanying drawings, and various modifications and variations may be made without departing from the scope thereof. The scope of the present invention is limited solely by the appended claims.

Claims

1. A human resources matching system based on big data, characterized in that: The human resources matching system includes a backend server and a front-end application end. The backend server includes: A construction module is configured to perform entity recognition on a target text block using a rule library to obtain an initial entity; perform entity recognition on the target text block using a first machine learning model, and adjust an intermediate recognition result using the initial entity during the recognition process of the first machine learning model to obtain a target entity; and construct a graph database based on the target entity; wherein the target text block includes a text block obtained by segmenting a job application text and a text block obtained by segmenting a recruitment text; The search module is used to generate search graph data based on the user's input focus words and the user's corresponding graph data; a matching module, configured to calculate the similarity between the search graph data and the graph data to be selected in the graph database, and select matching graph data from the graph database based on the similarity; The first machine learning model includes a model input layer, a model encoding layer, a first fusion layer, a correction input layer, a correction encoding layer, and a model decoding layer; The building blocks are used to: Performing vectorization processing on the target text block using the model input layer to obtain a main vector; and performing multiple encoding processing on the main vector using multiple layers of the model encoding layer and then outputting the result; Performing vectorization processing on the initial entity using the correction input layer to obtain a correction vector; and encoding the correction vector using the correction encoding layer and outputting the encoding vector; Using the first fusion layer to fuse the data output by the multiple layers of the model coding layer and the data output by the correction coding layer; Using multiple layers of the model decoding layer to perform multiple decoding processes on the data output by the first fusion layer to obtain the target entity; The first machine learning model further includes at least one second fusion layer, and the building module is configured to: Use the i-th second fusion layer to perform weighted averaging on the output data of the corrected coding layer of the i+1-th layer and the output data of the model coding layer of the i+1-th layer; The model coding layer of the penultimate i-th layer is used to encode the data output by the i-th second fusion layer and then output it; wherein the weighting coefficient of the second fusion layer is adjustable.

2. The human resources matching system based on big data according to claim 1 is characterized in that: The human resources matching system includes a plurality of graph databases, each of which corresponds to an industry; The search module is used to receive a focus word input by a user, obtain a graph database corresponding to the user, use the focus word to modify the weight of the graph data corresponding to the user, and generate the search graph data; The matching module is used to determine the search industry corresponding to the search graph data, and calculate the similarity of each node between the search graph data and the graph data to be selected in the graph database corresponding to the search industry to determine the similarity of the graph data.

3. The human resources matching system based on big data according to claim 2 is characterized in that: The backend server receives an initial screening instruction sent by the front-end application, the initial screening instruction including an initial screening keyword, generates a recommended keyword based on the initial screening keyword, and sends the recommended keyword to the front-end application; The backend server receives a progressive screening instruction sent by a front-end application, wherein the progressive screening instruction includes a progressive keyword, and generates the focus word according to the initial screening keyword and the progressive keyword.

4. The human resources matching system based on big data according to claim 1 is characterized in that: The building blocks are specifically used for: Using a second machine learning model to extract the current year's project plan from the enterprise development report, generating a talent recruitment plan based on the current year's project plan and an existing talent pool, and updating the graph data corresponding to the user based on the talent recruitment plan; The search module generates the search graph data according to the focus word input by the user and the updated graph data.

5. The human resources matching system based on big data according to claim 4 is characterized in that: The existing talent pool includes at least one existing talent information, and each existing talent information includes mastered skills and project experience; The search module is specifically used for: Extracting talent demand information from the current year's project plan using the second machine learning model to generate a talent demand database, wherein the talent demand information includes skill requirements and project requirements; For each required talent information, the required talent information is matched with each existing talent information in the existing talent database, and talent gap information is determined according to the matching result; and the talent gap information is used as a recruitment plan.

6. A human resource matching method based on big data, characterized in that: include: Use the rule base to perform entity recognition on the target text block to obtain the initial entity; Performing entity recognition on the target text block using a first machine learning model, adjusting an intermediate recognition result using the initial entity during the recognition process of the first machine learning model to obtain a target entity; constructing a graph database based on the target entity; wherein the target text block includes a text block obtained by segmenting the job application text and a text block obtained by segmenting the recruitment text; Generate search graph data based on the user's input focus words and the user's corresponding graph data; Calculating the similarity between the search graph data and the graph data to be selected in the graph database, and selecting matching graph data from the graph database based on the similarity; The first machine learning model includes a model input layer, a model encoding layer, a first fusion layer, a correction input layer, a correction encoding layer, and a model decoding layer; Performing vectorization processing on the target text block using the model input layer to obtain a main vector; and performing multiple encoding processing on the main vector using multiple layers of the model encoding layer and then outputting the result; Performing vectorization processing on the initial entity using the correction input layer to obtain a correction vector; and encoding the correction vector using the correction encoding layer and outputting the encoding vector; Using the first fusion layer to fuse the data output by the multiple layers of the model coding layer and the data output by the correction coding layer; Using multiple layers of the model decoding layer to perform multiple decoding processes on the data output by the first fusion layer to obtain the target entity; The first machine learning model further includes at least one second fusion layer; Use the i-th second fusion layer to perform weighted averaging on the output data of the corrected coding layer of the i+1-th layer and the output data of the model coding layer of the i+1-th layer; The model coding layer of the penultimate i-th layer is used to encode the data output by the i-th second fusion layer and then output it; wherein the weighting coefficient of the second fusion layer is adjustable.

7. A backend server, characterized in that: include: processor; as well as, a memory for storing executable instructions of the processor; The processor is configured to perform the method of claim 6 by executing the executable instructions.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to claim 6 when executed by a processor.

Images