Heterogeneous graph-based man-post matching method and device, equipment and medium

Abstract

The invention discloses a person and post matching method and device based on a heterogeneous graph, equipment and a medium. Comprising the steps of obtaining position information, resume information and historical matching behavior data between the position information and the resume information; taking the position information as a position node, taking the resume information as a resume node, and establishing a first edge relationship between the position node and the resume node based on historical matching behavior data; based on the similarity of each piece of position information, the similarity of each piece of resume information and historical matching behavior data, establishing a second edge relationship between position nodes, a third edge relationship between resume nodes and a fourth edge relationship between the position nodes and the resume nodes; constructing a global heterogeneous graph based on each position node, each resume node and the edge relationship thereof; training a GLAN model based on the global heterogeneous graph; and converting the position information to be matched into heterogeneous graph features based on the GLAN model, and screening matched target resume information for the position information to be matched based on the heterogeneous graph features.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, in particular to the field of information recommendation technology, and in particular to a method, device, equipment and medium for matching people and jobs based on heterogeneous graphs. Background Art

[0002] In order to provide users with better services and a better service experience, existing recruitment platforms usually recommend positions to job seekers or recommend resumes to recruiters with authorization. In this recommendation process, accurate person-job matching is the key to improving recruitment efficiency and job search success rate. Current matching methods are usually based on keyword matching between resumes and job descriptions (JDs), which makes it difficult to fully capture the complex relationship between resumes and positions.

[0003] To this end, a deep model-based person-job matching algorithm has been designed in the relevant technology, which can capture the complex relationship between resumes and positions in combination with contextual information and perform person-job matching. However, when the samples of interaction behaviors between users and positions are sparse, the existing person-job matching process is difficult to achieve good results, thus affecting the accuracy of person-job matching recommendations. Summary of the invention

[0004] In view of this, the embodiments of the present application provide a person-job matching method, device, equipment, medium and product based on a heterogeneous graph, which can improve the accuracy of person-job matching and achieve accurate matching.

[0005] In a first aspect, an embodiment of the present application provides a method for matching people and jobs based on a heterogeneous graph, the method comprising: obtaining each position information, each resume information, and historical matching behavior data between the position information and the resume information; taking the position information as the position node and the resume information as the resume node, and establishing a first edge relationship between the position node and the resume node based on the historical matching behavior data, wherein the first edge relationship is used to characterize the matching relationship; establishing a second edge relationship between the position node and the position node, a third edge relationship between the resume node and the resume node, and a fourth edge relationship between the position node and the resume node based on the similarity between each position information, the similarity between each resume information, and the historical matching behavior data, wherein the second edge relationship is used to characterize the position similarity relationship, the third edge relationship is used to characterize the resume similarity relationship, and the fourth edge relationship is used to characterize the potential matching relationship; constructing a global heterogeneous graph based on each position node, each resume node and their edge relationships; training a global-local attention network GLAN model based on the global heterogeneous graph; converting the position information to be matched into heterogeneous graph features based on the GLAN model, and screening matching target resume information for the position information to be matched based on the heterogeneous graph features corresponding to the position information to be matched.

[0006] In some implementations of the first aspect, a second side relationship between position nodes and position nodes, a third side relationship between resume nodes and resume nodes, and a fourth side relationship between position nodes and resume nodes are established, including: establishing a second side relationship between position nodes and position nodes based on the similarity between each position information; establishing a third side relationship between resume nodes and resume nodes based on the similarity between each resume information; for the first position node of each position node, obtaining a set of position nodes that have a second side relationship with the first position node; obtaining a first set of resume nodes that have a first side relationship with the position nodes in the position node set, and a second set of resume nodes that have a third side relationship with the resume nodes in the first resume node set; establishing a fourth side relationship between the first position node and all resume nodes in the first resume node set and the second resume node set.

[0007] In some implementable methods of the first aspect, based on the heterogeneous graph features corresponding to the job information to be matched, matching target resume information is screened for the job information to be matched, including: determining the text features corresponding to the job information to be matched based on the text feature learning model LSTM; fusing the heterogeneous graph features with the text features to obtain the fused features corresponding to the information to be matched; and determining the target resume information that matches the job information to be matched based on the fused features corresponding to the information to be matched.

[0008] In some implementation methods of the first aspect, before inputting the to-be-matched position information into the text feature learning model, it also includes: screening, from the global heterogeneous graph, a first position resume node combination with a first edge relationship, a second position resume node combination with a fourth edge relationship, and a third position resume node combination without an edge relationship, wherein the edge weight of the first edge relationship is greater than the edge weight of the fourth edge relationship; using the word vectors corresponding to the first position resume node combination and the second position resume node combination as positive samples, and using the word vectors corresponding to the third position resume node combination and position information and resume information in different technical fields as negative samples, and inputting them into the text feature learning model for model training.

[0009] In some implementation methods of the first aspect, based on the text feature learning model LSTM, the text features corresponding to the to-be-matched position information are determined, including: by inputting the to-be-matched position information into the word vector model BERT, the position information to be matched is converted into a word vector; by inputting the word vector into the LSTM, the word vector is converted into the text features corresponding to the to-be-matched position information.

[0010] In some possible implementations of the first aspect, target resume information that matches the position information to be matched is determined based on the fusion features corresponding to the information to be matched, including: inputting the fusion features into the detector and the discriminator respectively, evaluating the matching degree between the position information to be matched and the resume information in the resume library in combination with the detector and the discriminator, and outputting multiple target resume information in order of the matching scores from high to low.

[0011] In some implementable embodiments of the first aspect, the target person-job matching model includes BERT, LSTM, GLAN model, detector and discriminator, converts the to-be-matched job information into heterogeneous graph features based on the GLAN model, and screens matching target resume information for the to-be-matched job information based on the heterogeneous graph features corresponding to the to-be-matched job information, including: inputting the to-be-matched job information into the target person-job matching model, so that the target person-job matching model screens matching target resume information for the to-be-matched job information based on the heterogeneous graph features output by the GLAN model and the text features output by the LSTM; wherein the target person-job matching model is obtained by introducing an adversarial training mechanism.

[0012] In a second aspect, an embodiment of the present application provides a person-job matching device based on a heterogeneous graph, the device comprising: an acquisition module for acquiring each position information, each resume information, and historical matching behavior data between the position information and the resume information; a node relationship establishment module for taking the position information as a position node and the resume information as a resume node, and establishing a first edge relationship between the position node and the resume node based on the historical matching behavior data, wherein the first edge relationship is used to characterize the matching relationship; the node relationship establishment module is also used to establish a second edge relationship between the position node and the position node based on the similarity between the each position information, the similarity between the each resume information, and the historical matching behavior data. The third edge relationship between resume nodes and resume nodes, and the fourth edge relationship between position nodes and resume nodes, wherein the second edge relationship is used to characterize position similarity relationships, the third edge relationship is used to characterize resume similarity relationships, and the fourth edge relationship is used to characterize potential matching relationships; a graph construction module, used to construct a global heterogeneous graph based on each position node, each resume node and their edge relationships; a model training module, used to train a global-local attention network GLAN model based on the global heterogeneous graph; a person-job matching module, used to convert the position information to be matched into heterogeneous graph features based on the GLAN model, and based on the heterogeneous graph features corresponding to the position information to be matched, screen matching target resume information for the position information to be matched.

[0013] In some implementations of the second aspect, the node relationship establishment module is specifically used to: establish a second side relationship between position nodes and position nodes based on the similarity between each position information; establish a third side relationship between resume nodes and resume nodes based on the similarity between each resume information; for the first position node of each position node, obtain a position node set that has a second side relationship with the first position node; obtain a first resume node set that has a first side relationship with the position nodes in the position node set, and a second resume node set that has a third side relationship with the resume nodes in the first resume node set; establish a fourth side relationship between the first position node and all resume nodes in the first resume node set and the second resume node set.

[0014] In some implementable methods of the second aspect, the person-job matching module includes: a determination unit, used to determine the text features corresponding to the job information to be matched based on the text feature learning model LSTM; a feature fusion unit, used to fuse the heterogeneous graph features with the text features to obtain the fused features corresponding to the information to be matched; the determination unit is also used to determine the target resume information that matches the job information to be matched based on the fused features corresponding to the information to be matched.

[0015] In some implementations of the second aspect, it also includes: a screening module, which is used to screen, from the global heterogeneous graph, a first position resume node combination with a first edge relationship, a second position resume node combination with a fourth edge relationship, and a third position resume node combination without an edge relationship, before inputting the to-be-matched position information into the text feature learning model, wherein the edge weight of the first edge relationship is greater than the edge weight of the fourth edge relationship; a model training module, which is used to use the word vectors corresponding to the first position resume node combination and the second position resume node combination as positive samples, and use the word vectors corresponding to the third position resume node combination and position information and resume information in different technical fields as negative samples, and input them into the text feature learning model for model training.

[0016] In some implementation methods of the second aspect, the determination unit is specifically used to: convert the to-be-matched position information into a word vector by inputting the to-be-matched position information into a word vector model BERT; and convert the word vector into a text feature corresponding to the to-be-matched position information by inputting the word vector into an LSTM.

[0017] In some implementations of the second aspect, the determination unit is also used to: input fusion features to the detector and the discriminator respectively, evaluate the matching degree between the job information to be matched and the resume information in the resume library in combination with the detector and the discriminator, and output multiple target resume information in order of the matching scores from high to low.

[0018] In some implementable embodiments of the second aspect, the target person-job matching model includes BERT, LSTM, GLAN model, detector and discriminator, and the person-job matching module is specifically used to: input the job information to be matched into the target person-job matching model, so that the target person-job matching model screens the target resume information that matches the job information to be matched based on the heterogeneous graph features output by the GLAN model and the text features output by the LSTM; wherein the target person-job matching model is obtained by introducing an adversarial training mechanism.

[0019] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the heterogeneous graph-based person-job matching method of the first aspect are implemented.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the steps of the heterogeneous graph-based person-job matching method of the first aspect are implemented.

[0021] In a fifth aspect, an embodiment of the present application provides a computer program product, which is stored in a non-volatile storage medium. When the computer program product is executed by a processor, it implements the steps of the heterogeneous graph-based person-job matching method as in the first aspect.

[0022] In a sixth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the steps of the heterogeneous graph-based person-job matching method of the first aspect.

[0023] The present application provides a method, device, equipment, medium and product for matching people and jobs based on a heterogeneous graph, which obtains each job information, each resume information, and historical matching behavior data between the job information and the resume information; takes the job information as the job node and the resume information as the resume node, and establishes a matching relationship between the job node and the resume node, a similar job relationship between the job node and the job node, and a similar resume relationship between the resume node and the resume node based on the similarity between the job information, the similarity between the resume information, and the historical matching behavior data; further mining is performed based on the above three relationships to mine the potential matching relationship between the job node and the resume node, and a global heterogeneous graph is constructed based on each job node, each resume node and their edge relationships. In this way, when training the global-local attention network GLAN model based on the global heterogeneous graph, the features learned from the Resume and Job relationship graph constructed using the global heterogeneous graph are used to mine the potential Resume and Job relationship, and the position nodes and resume nodes with potential matching relationships are also used as training samples to expand the number of training samples, overcome the limitation that a large number of real samples must be required in the previous person-job matching, solve the problem that when there are few training samples for a certain position, it is difficult to train and the model effect is not ideal, and improve the training effect and person-job matching performance of the GLAN model. In addition, after converting the to-be-matched position information into heterogeneous graph features based on the GLAN model, the heterogeneous graph features can not only reflect the characteristics of the to-be-matched position information itself, but also retain the complex relationship between the to-be-matched position information and other position nodes and resume nodes. The feature content of the heterogeneous graph features is relatively rich and comprehensive, so the heterogeneous graph features can provide a reliable basis for subsequent person-job matching. Based on the heterogeneous graph features, the target resume information with a higher matching degree can be screened for the to-be-matched position information, and the matching degree between the target resume information and the to-be-matched position information, that is, the accuracy of person-job matching, can be improved to achieve accurate matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0025] Figure 1 It is a flowchart of a heterogeneous graph-based person-job matching method provided in one embodiment of the present application;

[0026] Figure 2 is an exemplary schematic diagram of a global heterogeneous graph provided by an embodiment of the present application;

[0027] Figure 3 is an exemplary schematic diagram of a global heterogeneous graph provided by another embodiment of the present application;

[0028] Figure 4 is an exemplary schematic diagram of a target person-job matching model provided in an embodiment of the present application;

[0029] Figure 5 is an exemplary schematic diagram of a person-job matching scenario provided in an embodiment of the present application;

[0030] Figure 6 It is a structural schematic diagram of a person-job matching device based on a heterogeneous graph provided in an embodiment of the present application;

[0031] Figure 7 It is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The principles and spirit of the present application will be described below with reference to several exemplary embodiments. It should be understood that the purpose of providing these embodiments is to make the principles and spirit of the present application clearer and more thorough, so that those skilled in the art can better understand and implement the principles and spirit of the present application. The exemplary embodiments provided herein are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments herein, all other embodiments obtained by ordinary technicians of the art without creative work are within the scope of protection of this application.

[0033] In this document, terms such as first, second, third, etc. are only used to distinguish one entity (or operation) from another entity (or operation), but not to require or imply any order or relationship between these entities (or operations).

[0034] The following is a brief description of the concepts and technical terms that may be involved in the embodiments of the present application.

[0035] Generally, the services provided by existing recruitment platforms to job seekers are called C-end services, which include but are not limited to: job seekers can create their own resumes on the platform, including their personal information, education background, skills and work experience; job seekers can also search and view positions according to various set conditions through the search engine provided by the platform; job seekers can submit their resumes on the job information page and communicate with the recruiter of the position online when the recruiter is online. The services provided by existing recruitment platforms to recruiters are called B-end services. B-end services include but are not limited to: recruiters can create job information posted by their own company on the platform, including company information, job responsibilities, various requirements or conditions for job seekers, salary and benefits, etc.; recruiters can also search, view and download resumes according to various set conditions through the search engine provided by the platform; recruiters can respond to job seekers' online communication requests and communicate with them online when job seekers view their posted job information pages, etc.

[0036] A global heterogeneous graph is a graph structure that contains different types of nodes and edges. Each node represents an entity, while the edge represents the relationship between entities. Global heterogeneous graph networks are designed to learn representations from these complex relationships and can be applied to multiple fields such as recommendation systems, social network analysis, bioinformatics, etc.

[0037] Global heterogeneous graph networks are heterogeneous and can handle different types of nodes and edges, each of which can have different attributes and characteristics.

[0038] Global heterogeneous graph networks are able to capture the global structural information in the graph, not just the information of the local neighborhood.

[0039] Global heterogeneous graph networks usually adopt a multi-layer structure with hierarchical representation. Each layer can learn representations at different levels to capture structural features at different scales.

[0040] Generative Adversarial Networks (GANs) is a model for generating data samples. The core idea is to learn the distribution of data through competition between two neural network models. GANs learn the distribution of data by training two competing neural network models, the generator and the discriminator.

[0041] Generator: A model used to generate data samples. Its input is usually random noise, and its output is a new sample with a distribution similar to the training data.

[0042] Discriminator: A classifier that distinguishes between real training data samples and fake samples generated by the generator. Its output is a probability value indicating the probability that the input sample is real or fake.

[0043] The person-job matching method based on heterogeneous graph provided in the embodiment of the present application is described in detail below through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0044] Figure 1 It is a flow chart of a heterogeneous graph-based person-job matching method provided in one embodiment of the present application. The execution subject of the heterogeneous graph-based person-job matching method may be a server.

[0045] Exemplarily, the server can be a B-end server or a C-end server of an online recruitment platform. The B-end can be an online talent recruitment system, an online application recruitment system, a training system, etc. used by recruiting users, and the C-end can be a job search system for fresh graduates, an online application job search system, a job recommendation system, etc. used by job seekers.

[0046] The following takes the server as an example of the execution subject of the heterogeneous graph-based person-job matching method to illustrate the heterogeneous graph-based person-job matching method of the present application. It should be noted that the above execution subjects and application scenarios do not constitute a limitation on the present application.

[0047] like Figure 1 As shown, the heterogeneous graph-based person-job matching method provided in the embodiment of the present application may include steps 110 to 160.

[0048] Step 110, obtaining each position information, each resume information, and historical matching behavior data between the position information and the resume information;

[0049] Step 120, taking the position information as the position node and the resume information as the resume node, and establishing a first edge relationship between the position node and the resume node based on the historical matching behavior data;

[0050] Step 130, based on the similarity between each position information, the similarity between each resume information, and the historical matching behavior data, establish a second side relationship between position nodes and position nodes, a third side relationship between resume nodes and resume nodes, and a fourth side relationship between position nodes and resume nodes;

[0051] Step 140, constructing a global heterogeneous graph based on each position node, each resume node and their edge relationships;

[0052] Step 150, training a Global-Local Attention Network (GLAN) model based on the global heterogeneous graph;

[0053] Step 160: convert the to-be-matched job information into heterogeneous graph features based on the GLAN model, and screen matching target resume information for the to-be-matched job information based on the heterogeneous graph features corresponding to the to-be-matched job information.

[0054] The heterogeneous graph-based person-job matching method provided in the embodiment of the present application obtains each position information, each resume information, and historical matching behavior data between the position information and the resume information; takes the position information as the position node and the resume information as the resume node, and establishes a matching relationship between the position node and the resume node, a similar position relationship between the position node and the position node, and a similar resume relationship between the resume node and the resume node based on the similarity between the various position information, the similarity between the various resume information, and the historical matching behavior data, and further mines based on the above three relationships to mine the potential matching relationship between the position node and the resume node, and constructs a global heterogeneous graph based on the various position nodes, the various resume nodes and their edge relationships. In this way, when training the global-local attention network GLAN model based on the global heterogeneous graph, the features learned from the Resume and Job relationship graph constructed using the global heterogeneous graph are used to mine the potential Resume and Job relationship, and the position nodes and resume nodes with potential matching relationships are also used as training samples to expand the number of training samples, overcome the limitation that a large number of real samples must be required in the previous person-job matching, solve the problem that when there are few training samples for a certain position, it is difficult to train and the model effect is not ideal, and improve the training effect and person-job matching performance of the GLAN model. In addition, based on the GLAN model, the position information to be matched is converted into heterogeneous graph features. The heterogeneous graph features can not only reflect the characteristics of the position information to be matched itself, but also retain the complex relationship between the position information to be matched and other position nodes and resume nodes. The feature content of the heterogeneous graph features is relatively rich and comprehensive, so the heterogeneous graph features can provide a reliable basis for subsequent person-job matching. The heterogeneous graph features can be used to screen the target resume information with a higher matching degree for the position information to be matched, improve the matching degree between the target resume information and the position information to be matched, that is, the accuracy of person-job matching, and achieve accurate matching.

[0055] The specific implementation of the above steps will be described in detail below in conjunction with specific embodiments.

[0056] Involving step 110, obtaining each position information, each resume information, and historical matching behavior data between the position information and the resume information.

[0057] In step 110, the historical matching behaviors may include but are not limited to: resume submission behavior of job seekers, extending hand to recruiters, recruitment invitation behavior of recruiters to job seekers, resume downloading behavior, extending hand to job seekers, and handshake between recruiters and job seekers.

[0058] Involving step 120, taking the position information as the position node and the resume information as the resume node, based on the historical matching behavior data, a first edge relationship between the position node and the resume node is established.

[0059] In step 120, a first edge relationship is established between the nodes corresponding to the position information and the resume information having historical matching behaviors, and the first edge relationship is used to characterize the matching relationship (between the position and the resume).

[0060] As a specific example, there are the following job information and resume information:

[0061] Job2: Position - AI Engineer; Job requirements - Bachelor degree or above, computer science major; Skill requirements - Familiar with Python and Java, understanding of deep learning algorithms, fine-tuning pre-trained models and experience in model deployment.

[0062] Resume 2: Education: Bachelor’s degree; Major: Artificial Intelligence; Skills: Proficient in Python, familiar with Java, experience in deep learning model training and deployment, and relevant work experience.

[0063] Job3: Position - Algorithm Engineer; Job Requirements - Bachelor degree or above, major in artificial intelligence; Skill Requirements - Familiar with Python and C++, understanding of deep learning algorithms, at least one deep learning framework such as pytorch, and understanding of generative large language models.

[0064] Resume3: Education: Bachelor’s degree; Major: Computer Science and Technology; Skills: Proficient in Python, familiar with C++, familiar with the pytorch framework, understanding of mainstream algorithms, and experience in fine-tuning large open source models to solve problems.

[0065] Among them, the above Job2 and Resume2 have historical matching behaviors, and Job3 and Resume3 have historical matching behaviors, so Figure 2 As shown, a matching relationship is established between Job2 and Resume2, and between Job3 and Resume3, and the matching relationship is represented by a solid line.

[0066] Involving step 130, based on the similarity between each position information, the similarity between each resume information, and the historical matching behavior data, a second side relationship between position nodes and position nodes, a third side relationship between resume nodes and resume nodes, and a fourth side relationship between position nodes and resume nodes are established.

[0067] In step 130, the similarity between each position information and the similarity between each resume information can be calculated in advance using a preset similarity algorithm, such as a cosine similarity algorithm, a Jaccard similarity coefficient, an Euclidean distance, etc., and this application does not make any specific limitations on this.

[0068] The second side relationship is used to characterize the position similarity relationship, such as partial overlap in recruitment skill requirements. The third side relationship is used to characterize the resume similarity relationship, such as similar resume skills and work experience. The fourth side relationship is used to characterize the potential matching relationship, which is the matching relationship between the inferred position node and the resume node.

[0069] Based on the similarity between each piece of job information, the job similarity between every two pieces of job information is determined, and a second edge relationship is established between the job nodes corresponding to two pieces of job information whose job similarity is greater than a preset job similarity threshold.

[0070] As a specific example, Job 1: Position - Network attack and defense security engineer; Job requirements - Bachelor degree or above, computer major; Skill requirements: Familiar with Python language, understanding of network attack and defense security technology, and priority is given to those with work experience in network attack and defense security companies or research institutes.

[0071] If Job1 has similar positions to Job2 and Job3 in the above example, then Figure 2 As shown, a second edge relationship, namely a position similarity relationship, is established between Job1 and Job2, and between Job1 and Job3.

[0072] Based on the similarity between each resume information, the position similarity between each two resume information is determined, and a third edge relationship is established between the resume nodes corresponding to two resume information whose resume similarity is greater than a preset resume similarity threshold.

[0073] As a specific example, Resume 1: Education: Master's degree; Major: Computer technology; Skills: Proficient in Python programming, familiar with mainstream deep learning algorithms, and one year of experience working in a research institute engaged in AI-related work.

[0074] If Resume1 is similar to Resume2 and Resume3 in the above example, you can Figure 2 As shown, a third edge relationship is established between Resume1 and Resume2, and between Resume1 and Resume3, that is, a resume similarity relationship.

[0075] It can be understood that for position nodes and resume nodes with potential matching relationships, there is no matching behavior in actual applications between their corresponding position information and resume information. The potential matching relationship is obtained by deeply mining the first-edge relationship, second-edge relationship, and third-edge relationship between each position node and each resume node.

[0076] Specifically, for the second position node in each position node, a set of third resume nodes having a first edge relationship with the second position node is obtained;

[0077] Obtain a fourth resume node set having a third edge relationship with the resume nodes in the third resume node set;

[0078] A fourth edge relationship is established between the second position node and all resume nodes in the fourth resume node set.

[0079] Specifically, the second position node is a position node corresponding to the position information with historical matching behavior, and the resume nodes that have a third edge relationship with any resume node in the third resume node set are obtained to form a fourth resume node set.

[0080] In one exemplary embodiment, Figure 2 As shown in FIG. 1 , Job2 is used as the first position node, then the third resume node set having a first variable relationship with Job2 includes Resume2, and the fourth resume node set having a second edge relationship with Resume2 includes Resume1. Figure 3 As shown, the fourth side relationship “Potential Match” is established between Job2 and Resume1.

[0081] In another exemplary embodiment, Figure 2 As shown in FIG. 1 , Job3 is used as the first position node, then the third resume node set having a first variable relationship with Job3 includes Resume3, and the fourth resume node set having a second edge relationship with Resume3 includes Resume1. Figure 3 As shown, a fourth side relationship “potential match” is established between Job3 and Resume1. In order to distinguish “potential match” from “match”, the fourth side relationship “potential match” is represented by a dotted line in the figure.

[0082] In some embodiments of the present application, in order to mine as many potential matching relationships between position nodes and resume nodes as possible and construct more available training samples, the above step 130 establishes the second side relationship between position nodes and position nodes, the third side relationship between resume nodes and resume nodes, and the fourth side relationship between position nodes and resume nodes, which may specifically include the following steps:

[0083] Based on the similarity between each position information, establish the second side relationship between position nodes;

[0084] Based on the similarity between each resume information, a third-side relationship between resume nodes is established;

[0085] For the first position node of each position node, obtain a set of position nodes that have a second edge relationship with the first position node;

[0086] Obtain a first resume node set having a first side relationship with the position nodes in the position node set, and a second resume node set having a third side relationship with the resume nodes in the first resume node set;

[0087] A fourth edge relationship is established between the first position node and all resume nodes in the first resume node set and the second resume node set.

[0088] Specifically, the first position node is any position node among all the position nodes. For each position node, the above mining process can be used to mine the potential matching relationship between the position node and different resume nodes.

[0089] For example, Figure 2 As shown in the figure, Job1 is the first position node, and the set of position nodes that have a second edge relationship with Job1 includes Job2 and Job3. The set of first resume nodes that have a first edge relationship with Job2 and Job3 includes Resume2 and Resume3, and the set of second resume nodes that have a second edge relationship with Resume2 and Resume3 includes Resume1. Figure 2 As shown, a fourth side relationship “potential match” is established between Job1 and Resume1, Resume2, and Resume3.

[0090] In an embodiment of the present application, after establishing the first side relationship, the second side relationship and the third side relationship between each position node and each resume node, based on the position similarity reflected by the second side relationship, the resume similarity reflected by the third side relationship, and the position-resume matching reflected by the first side relationship, it is possible to effectively retrieve the position nodes and resume nodes that may have a matching relationship, deeply explore the position-resume combinations with potential matching relationships, enrich the training samples of the GLAN model, and thereby improve the training effect and person-job matching performance of the GLAN model.

[0091] Involving step 140, a global heterogeneous graph is constructed based on each position node, each resume node and their edge relationships.

[0092] In step 140, the global heterogeneous graph refers to a graph containing multiple types of nodes and edges. In this application, the nodes are divided into two types: position nodes and resume nodes, and the edges correspond to four types of edge relationships. Therefore, in combination with the first edge relationship, the second edge relationship, the third edge relationship, and the fourth edge relationship between each position node and each resume node, the corresponding position node and the resume node are connected to construct a global heterogeneous graph.

[0093] In some embodiments, a graph data structure, such as an adjacency matrix or an adjacency list, may be used to represent edge relationships between nodes.

[0094] Involving step 150, a global-local attention network GLAN model is trained based on the global heterogeneous graph.

[0095] In step 150, the global heterogeneous graph can reflect the job similarity relationship between job nodes, the resume similarity relationship between resume nodes, and the matching relationship and potential matching relationship between job nodes and resume nodes. Therefore, training the GLAN model based on the global heterogeneous graph can enable the GLAN model to learn similar job features, similar resume features, and job resume features with matching relationships and potential matching relationships, so that the trained GLAN model has the ability to mine matching relationships and potential matching relationships between job resumes.

[0096] Involving step 160, the to-be-matched position information is converted into heterogeneous graph features based on the GLAN model, and matching target resume information is screened for the to-be-matched position information based on the heterogeneous graph features corresponding to the to-be-matched position information.

[0097] In step 160, when the job information to be matched is input to the GLAN model, similar job information, matching resume information and potential matching resume information of the job information to be matched are obtained through the GLAN model, and the similar job information, matching resume information and potential matching resume information are combined to output heterogeneous graph features, which can be used to characterize the job similarity between the job information to be matched and other job information, as well as the matching and potential matching with other resume information. Therefore, based on the richness of the feature content represented by the heterogeneous graph feature, the target resume information can be accurately screened to improve the matching degree between the target resume information and the job information.

[0098] In some embodiments of the present application, in order to further improve the accuracy of person-job matching, the above step 160, based on the heterogeneous graph features corresponding to the job information to be matched, screens the target resume information that matches the job information to be matched, which may specifically include:

[0099] Based on the text feature learning model LSTM, determine the text features corresponding to the job information to be matched;

[0100] Fuse the heterogeneous graph features with the text features to obtain the fused features corresponding to the information to be matched;

[0101] Based on the fusion features corresponding to the information to be matched, the target resume information matching the position information to be matched is determined.

[0102] Specifically, the heterogeneous graph feature and the text feature can be concatenated to obtain the fused feature corresponding to the information to be matched.

[0103] As a specific example, the graph feature is represented by the matrix [103, 210, 190, 220, 660], and the text feature is represented by the matrix [102, 201, 180, 230, 550]. By fusing the two features, we can get the fused feature [102, 201, 180, 230, 550, 103, 210, 190, 220, 660].

[0104] In the embodiment of the present application, the text feature learning model is used to preliminarily understand some semantic information in the job requirements and project experience of the job information to be matched, determine the text features corresponding to the job information to be matched, and obtain the text representation of the job information to be matched at the word level. Therefore, the heterogeneous graph features and text features are fused, and the fused features can contain more semantic information than the heterogeneous graph features. The fused features can be used to match people and jobs in combination with the context information of the job information to be matched, and select target resume information with a higher matching degree for the job information to be matched.

[0105] In some embodiments of the present application, in order to improve the richness of training samples of the text feature learning model, before inputting the to-be-matched position information into the text feature learning model, the following steps may also be included:

[0106] From the global heterogeneous graph, select the first position resume node combination with the first edge relationship, the second position resume node combination with the fourth edge relationship, and the third position resume node combination without the edge relationship;

[0107] The word vectors corresponding to the first position resume node combination and the second position resume node combination are taken as positive samples, and the word vectors corresponding to the third position resume node combination and the position information and resume information in different technical fields are taken as negative samples and input into the text feature learning model for model training.

[0108] Among them, the first position resume node combination is a combination of position nodes and resume nodes with historical matching behavior, the second position resume node combination is a combination of position nodes and resume nodes with potential matching relationships, and the third position resume node combination is a combination of position nodes and resume nodes that have neither historical matching behavior (i.e., the first edge relationship) nor potential matching relationships (i.e., the fourth edge relationship), and the edge weight of the first edge relationship is greater than the edge weight of the fourth edge relationship. The word vector corresponding to the third position resume node combination, as well as the word vectors corresponding to position information and resume information in different technical fields are used as negative samples.

[0109] For example, the positive sample job is: AI engineer, bachelor degree or above, computer science major; skill requirements: 1. Familiar with Python and Java; 2. Understand deep learning algorithms and fine-tune pre-trained models; 3. Have experience in model deployment.

[0110] Positive sample Resume: Education: Bachelor's degree; Major: Artificial Intelligence; Skills: 1. Proficient in Python and familiar with Java; 2. Experience in deep learning model training and deployment; 3. Related work experience.

[0111] Negative sample Job: AI engineer, bachelor degree or above, computer science major; Skill requirements: 1. Familiar with Python and Java; 2. Understand deep learning algorithms and fine-tune pre-trained models; 3. Have experience in model deployment.

[0112] Negative sample Resume: Education: Bachelor's degree; Major: Civil Engineering; Skills: 1. Assist the tutor to complete scientific research projects and participate in the design and analysis of bridge structures; 2. Use MATLAB to conduct numerical simulation experiments; 3. Write research reports and technical documents.

[0113] In this way, the potential matching relationship between job information and resume information can be mined based on the global heterogeneous graph. By using the second job resume node combination obtained by mining the potential relationship as a training positive sample, the richness of the training samples of the text feature learning model can be improved. In addition, the third job resume node combination and the job information and resume information corresponding to different technical fields are directly combined to obtain negative samples, which can quickly construct a large amount of negative sample training data, while not affecting the model training effect, reducing the difficulty of obtaining negative sample training data and improving the acquisition efficiency.

[0114] In some embodiments of the present application, the above-mentioned text feature learning model LSTM is used to determine the text features corresponding to the job information to be matched, which may specifically include:

[0115] By inputting the job information to be matched into the word vector model BERT, the job information to be matched is converted into a word vector;

[0116] By inputting word vectors into LSTM, the word vectors are converted into text features corresponding to the job information to be matched.

[0117] In this way, before inputting into the LSTM model, by inputting the to-be-matched position information into the word vector model BERT, the word vector can be obtained by combining the context information of the to-be-matched position information, and then by inputting into the LSTM, a good semantic representation can be obtained.

[0118] In some embodiments of the present application, the above-mentioned determination of target resume information matching the to-be-matched position information based on the fusion features corresponding to the to-be-matched information may specifically include:

[0119] The fused features are input into the detector and the discriminator respectively, and the matching degree between the job information to be matched and the resume information in the resume database is evaluated by combining the detector and the discriminator, and multiple target resume information are output in the order of matching scores from high to low.

[0120] Specifically, after the text vector enhanced by the global heterogeneous graph represents the fusion feature, the fusion feature is sent to the designed detector and discriminator respectively. The detector and the discriminator jointly complete the person-job matching classification task.

[0121] The detector uses a fully connected softmax classifier to classify the learned results into two categories: matching and mismatching. For example, when the model input resume2 and job2 pass through the softmax classifier, the probabilities of matching and mismatching are obtained, for example, 90% matching and 10% mismatching.

[0122] In the embodiment of the present application, the features specific to a certain job function can be removed by the discriminator, thereby obtaining shared features between the jobs, for example, when recruiting for the position of Python development engineer, all candidates are from computer science majors, then when the model is learning, it is likely to associate "Python development engineer" with "computer science major", and think that computer science major is a feature specific to the position of Python development engineer, so as long as the resume of computer science major is seen later, it will be judged that it is suitable for the position of Python development engineer, which will lead to errors in matching people to positions. Therefore, shared features are features after similar specific features are removed, such as "mastering at least one programming language (not limited to Python)" and "understanding data structures and algorithms". By learning these shared features, the robustness of the model can be well enhanced, and the accuracy of matching people to positions can be stabilized.

[0123] In some embodiments of the present application, the target person-job matching model may include BERT, LSTM, GLAN model, detector and discriminator. The above step 160 converts the to-be-matched job information into heterogeneous graph features based on the GLAN model, and screens the target resume information matching the to-be-matched job information based on the heterogeneous graph features corresponding to the to-be-matched job information, which may specifically include:

[0124] The target person-job matching model is input with the job information to be matched, so that the target person-job matching model can screen the target resume information that matches the job information to be matched based on the heterogeneous graph features output by the GLAN model and the text features output by the LSTM.

[0125] In one example, if Figure 4 As shown in the figure, the target person-job matching model can include BERT, LSTM, GLAN model, detector and discriminator. First, the job information to be matched is input into the "GLAN" model and "BERT+LSTM" respectively. After the two models output heterogeneous graph features and text features respectively, feature fusion is performed. Then, the fused features are input into the Detector and Discriminator respectively, and the two models jointly output the matching score of the target resume information.

[0126] Specifically, the target person-job matching model is obtained by introducing an adversarial training mechanism. During the training process of the target person-job matching model, the generator generates a matching score between the job information and the resume information, and the discriminator evaluates the authenticity of the match. An adversarial loss function can be set to optimize the matching result, including the generator loss and the discriminator loss, to ensure that the matching score generated by the target person-job matching model is authentic and effective.

[0127] In this way, inspired by adversarial networks, this application proposes for the first time to use the idea of ​​adversarial networks to solve the problem of person-job matching. By introducing adversarial networks to continuously enhance the identification ability of the model, the robustness of the model is made stronger than previous person-job matching methods, thereby improving the accuracy of person-job matching.

[0128] As a specific example, the present application can be applied to Figure 5 In the job matching scenario shown, after the HR user logs in to the recruitment website system, he selects a published job and clicks on the smart recommendation. The system requests the smart job matching model service, which requests resources from the resource server. In response to the resource request, the resource server filters the resume data from the resume database and returns the resume data with a matching score. The resume database may contain resume data Resume1, Resume1…Resume n. After completing the request, the service obtains the resume data with a matching score and returns the resume data to the system platform page in descending order according to the matching value. The matching value represents the degree of fit between the resume and the job.

[0129] Corresponding to the method embodiment of the present application, the present application also provides a person-job matching device based on a heterogeneous graph.

[0130] Figure 6 Schematic diagram of a person-job matching device based on a heterogeneous graph provided in an embodiment of the present application. Figure 6 As shown, the heterogeneous graph-based person-job matching device 600 may include: an acquisition module 610 , a node relationship establishment module 620 , a graph construction module 630 , a model training module 640 , and a person-job matching module 650 .

[0131] Among them, the acquisition module 610 is used to obtain each position information, each resume information, and the historical matching behavior data between the position information and the resume information; the node relationship establishment module 620 is used to use the position information as the position node and the resume information as the resume node, and establish a first side relationship between the position node and the resume node based on the historical matching behavior data, wherein the first side relationship is used to characterize the matching relationship; the node relationship establishment module 620 is also used to establish a second side relationship between the position node and the position node, and a third side relationship between the resume node and the resume node based on the similarity between the various position information, the similarity between the various resume information, and the historical matching behavior data. and a fourth edge relationship between a position node and a resume node, wherein the second edge relationship is used to characterize a position similarity relationship, the third edge relationship is used to characterize a resume similarity relationship, and the fourth edge relationship is used to characterize a potential matching relationship; a graph construction module 630, used to construct a global heterogeneous graph based on each position node, each resume node and their edge relationships; a model training module 640, used to train a global-local attention network GLAN model based on the global heterogeneous graph; a person-job matching module 650, used to convert the position information to be matched into heterogeneous graph features based on the GLAN model, and based on the heterogeneous graph features corresponding to the position information to be matched, screen matching target resume information for the position information to be matched.

[0132] The heterogeneous graph-based person-job matching device provided in the embodiment of the present application obtains each position information, each resume information, and historical matching behavior data between the position information and the resume information; takes the position information as the position node and the resume information as the resume node, and establishes a matching relationship between the position node and the resume node, a similar position relationship between the position node and the position node, and a similar resume relationship between the resume node and the resume node based on the similarity between the various position information, the similarity between the various resume information, and the historical matching behavior data, and further mines based on the above three relationships to mine out the potential matching relationship between the position node and the resume node, and constructs a global heterogeneous graph based on the various position nodes, the various resume nodes and their edge relationships. In this way, when training the global-local attention network GLAN model based on the global heterogeneous graph, the features learned from the Resume and Job relationship graph constructed using the global heterogeneous graph are used to mine the potential Resume and Job relationship, and the position nodes and resume nodes with potential matching relationships are also used as training samples to expand the number of training samples, overcome the limitation that a large number of real samples must be required in the previous person-job matching, solve the problem that when there are few training samples for a certain position, it is difficult to train and the model effect is not ideal, and improve the training effect and person-job matching performance of the GLAN model. In addition, after converting the to-be-matched position information into heterogeneous graph features based on the GLAN model, the heterogeneous graph features can not only reflect the characteristics of the to-be-matched position information itself, but also retain the complex relationship between the to-be-matched position information and other position nodes and resume nodes. The feature content of the heterogeneous graph features is relatively rich and comprehensive, so the heterogeneous graph features can provide a reliable basis for subsequent person-job matching. Based on the heterogeneous graph features, the target resume information with a higher matching degree can be screened for the to-be-matched position information, and the matching degree between the target resume information and the to-be-matched position information, that is, the accuracy of person-job matching, can be improved to achieve accurate matching.

[0133] In some embodiments of the present application, the node relationship module is specifically used to: establish a second side relationship between position nodes and position nodes based on the similarity between each position information; establish a third side relationship between resume nodes and resume nodes based on the similarity between each resume information; for the first position node of each position node, obtain a position node set that has a second side relationship with the first position node; obtain a first resume node set that has a first side relationship with the position nodes in the position node set, and a second resume node set that has a third side relationship with the resume nodes in the first resume node set; establish a fourth side relationship between the first position node and all resume nodes in the first resume node set and the second resume node set.

[0134] In some embodiments of the present application, the person-job matching module includes: a determination unit, used to determine the text features corresponding to the job information to be matched based on the text feature learning model LSTM; a feature fusion unit, used to fuse the heterogeneous graph features with the text features to obtain the fused features corresponding to the information to be matched; the determination unit is also used to determine the target resume information that matches the job information to be matched based on the fused features corresponding to the information to be matched.

[0135] In some embodiments of the present application, it also includes: a screening module, which is used to screen, from the global heterogeneous graph, a first position resume node combination with a first edge relationship, a second position resume node combination with a fourth edge relationship, and a third position resume node combination without an edge relationship, before inputting the to-be-matched position information into the text feature learning model, wherein the edge weight of the first edge relationship is greater than the edge weight of the fourth edge relationship; a model training module, which is used to use the word vectors corresponding to the first position resume node combination and the second position resume node combination as positive samples, and use the word vectors corresponding to the third position resume node combination and position information and resume information in different technical fields as negative samples, and input them into the text feature learning model for model training.

[0136] In some embodiments of the present application, the determination unit is specifically used to: convert the to-be-matched position information into a word vector by inputting the to-be-matched position information into the word vector model BERT; and convert the word vector into a text feature corresponding to the to-be-matched position information by inputting the word vector into the LSTM.

[0137] In some embodiments of the present application, the determination unit is also used to: input fusion features to the detector and the discriminator respectively, evaluate the matching degree between the position information to be matched and the resume information in the resume library in combination with the detector and the discriminator, and output multiple target resume information in order of the matching scores from high to low.

[0138] In some embodiments of the present application, the target person-job matching model includes BERT, LSTM, GLAN model, detector and discriminator, and the person-job matching module is specifically used to: input the job information to be matched into the target person-job matching model, so that the target person-job matching model can screen the matching target resume information for the job information to be matched based on the heterogeneous graph features output by the GLAN model and the text features output by the LSTM; wherein the target person-job matching model is obtained by introducing an adversarial training mechanism.

[0139] The heterogeneous graph-based person-job matching device provided in the embodiment of the present application can achieve Figure 1-5 The various processes implemented by the service platform in the method embodiment can achieve the same technical effect. To avoid repetition, they will not be described here.

[0140] Figure 7It is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application.

[0141] like Figure 7 As shown, the electronic device 700 includes a memory 701 , a processor 702 , and a computer program stored in the memory 701 and executable on the processor 702 .

[0142] In an example, the processor 702 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0143] The memory 701 may include a read-only memory (ROM), a random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer executable instructions, and when the software is executed (e.g., by one or more processors), it can be operated to perform the operations described with reference to the card opening method in the embodiment according to the first aspect of the present application.

[0144] The processor 702 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 701, so as to implement the card activation method in the embodiment of the first aspect.

[0145] In some examples, the electronic device 700 may further include a communication interface 703 and a bus 710. Figure 7 As shown, the memory 701, the processor 702, and the communication interface 703 are connected via a bus 710 and communicate with each other.

[0146] The communication interface 703 is mainly used to realize the communication between the modules, devices, units and / or equipment in the embodiment of the present application. The communication interface 703 can also be used to access input devices and / or output devices.

[0147] The bus 710 includes hardware, software, or both, coupling the components of the electronic device 700 to each other. By way of example and not limitation, the bus 710 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a Memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 710 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0148] The electronic device provided in the embodiment of the present application can realize Figure 1-5 The various processes implemented by the electronic device in the method embodiment can achieve the same technical effect, and to avoid repetition, they will not be described here.

[0149] In combination with the heterogeneous graph-based person-job matching method in the above embodiment, the present application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, the steps of any one of the heterogeneous graph-based person-job matching methods in the above embodiment are implemented.

[0150] In combination with the heterogeneous graph-based person-job matching method in the above embodiment, the present application embodiment can provide a computer program product for implementation. The (computer) program product is stored in a non-volatile storage medium, and when the program product is executed by at least one processor, it implements the steps of any one of the heterogeneous graph-based person-job matching methods in the above embodiment.

[0151] An embodiment of the present application further provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned heterogeneous graph-based person-job matching method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0152] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0153] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0154] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), appropriate firmware, plug-in, function card, etc. When implemented in software, the elements of the present application are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0155] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0156] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0157] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A person-job matching method based on heterogeneous graph, characterized in that: include: Obtaining each position information, each resume information, and historical matching behavior data between the position information and the resume information; Taking the position information as a position node and the resume information as a resume node, and establishing a first edge relationship between the position node and the resume node based on the historical matching behavior data, wherein the first edge relationship is used to characterize the matching relationship; Based on the similarities between the various job information, the similarities between the various resume information, and the historical matching behavior data, a second side relationship between job nodes and job nodes, a third side relationship between resume nodes and resume nodes, and a fourth side relationship between job nodes and resume nodes are established, wherein the second side relationship is used to characterize the job similarity relationship, the third side relationship is used to characterize the resume similarity relationship, and the fourth side relationship is used to characterize the potential matching relationship; Construct a global heterogeneous graph based on each position node, each resume node and their edge relationships; Training a global-local attention network GLAN model based on the global heterogeneous graph; The to-be-matched job information is converted into heterogeneous graph features based on the GLAN model, and matching target resume information is screened for the to-be-matched job information based on the heterogeneous graph features corresponding to the to-be-matched job information.

2. The method according to claim 1, characterized in that The establishing of the second side relationship between the position nodes and the position nodes, the third side relationship between the resume nodes and the resume nodes, and the fourth side relationship between the position nodes and the resume nodes includes: Based on the similarity between the various job information, a second edge relationship between job nodes is established; Based on the similarity between each resume information, a third-side relationship between resume nodes is established; For a first position node of each position node, obtaining a set of position nodes having a second edge relationship with the first position node; Acquire a first resume node set having a first edge relationship with the position nodes in the position node set, and a second resume node set having a third edge relationship with the resume nodes in the first resume node set; A fourth edge relationship is established between the first position node and all resume nodes in the first resume node set and the second resume node set.

3. The method according to claim 1, characterized in that The step of screening matching target resume information for the job information to be matched based on the heterogeneous graph features corresponding to the job information to be matched includes: Determine the text features corresponding to the job information to be matched based on the text feature learning model LSTM; Performing feature fusion on the heterogeneous graph features and the text features to obtain fused features corresponding to the information to be matched; Based on the fusion features corresponding to the information to be matched, target resume information matching the position information to be matched is determined.

4. The method according to claim 3, characterized in that Before inputting the to-be-matched job information into the text feature learning model, the method further includes: From the global heterogeneous graph, a first position resume node combination having the first edge relationship, a second position resume node combination having the fourth edge relationship, and a third position resume node combination having no edge relationship are screened, wherein the edge weight of the first edge relationship is greater than the edge weight of the fourth edge relationship; The word vectors corresponding to the first position resume node combination and the second position resume node combination are used as positive samples, and the word vectors corresponding to the third position resume node combination and position information and resume information in different technical fields are used as negative samples, and are input into the text feature learning model for model training.

5. The method according to claim 3, characterized in that: The LSTM-based text feature learning model determines the text features corresponding to the to-be-matched job information, including: By inputting the to-be-matched job information into the word vector model BERT, the to-be-matched job information is converted into a word vector; By inputting the word vector into LSTM, the word vector is converted into text features corresponding to the job information to be matched.

6. The method according to claim 3, characterized in that The determining, based on the fusion features corresponding to the information to be matched, target resume information matching the position information to be matched comprises: The fusion features are input into the detector and the discriminator respectively, and the matching degree between the to-be-matched position information and the resume information in the resume database is evaluated by combining the detector and the discriminator, and a plurality of target resume information are output in descending order of matching degree scores.

7. The method according to claim 1, characterized in that The target person-job matching model includes BERT, LSTM, GLAN model, detector and discriminator. The GLAN model is used to convert the job information to be matched into heterogeneous graph features, and based on the heterogeneous graph features corresponding to the job information to be matched, matching target resume information is screened for the job information to be matched, including: Inputting the to-be-matched job information into the target person-job matching model, so that the target person-job matching model screens matching target resume information for the to-be-matched job information based on the heterogeneous graph features output by the GLAN model and the text features output by the LSTM; Among them, the target person-job matching model is obtained by introducing an adversarial training mechanism.

8. A person-job matching device based on heterogeneous graph, characterized in that: include: An acquisition module is used to acquire each position information, each resume information, and historical matching behavior data between the position information and the resume information; A node relationship establishment module, used to use the position information as a position node and the resume information as a resume node, and to establish a first edge relationship between the position node and the resume node based on the historical matching behavior data, wherein the first edge relationship is used to characterize the matching relationship; The node relationship establishment module is further used to establish a second side relationship between position nodes, a third side relationship between resume nodes, and a fourth side relationship between position nodes and resume nodes based on the similarities between the various position information, the similarities between the various resume information, and the historical matching behavior data, wherein the second side relationship is used to characterize the position similarity relationship, the third side relationship is used to characterize the resume similarity relationship, and the fourth side relationship is used to characterize the potential matching relationship; Graph construction module, used to build a global heterogeneous graph based on each position node, each resume node and their edge relationships; A model training module, used for training a global-local attention network GLAN model based on the global heterogeneous graph; The person-job matching module is used to convert the job information to be matched into heterogeneous graph features based on the GLAN model, and screen matching target resume information for the job information to be matched based on the heterogeneous graph features corresponding to the job information to be matched.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the electronic device executes the computer program instructions, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images