Grid tag classification method and device, electronic equipment and storage medium

Abstract

The invention discloses a grid label classification method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a first heterogeneous graph corresponding to job application behavior information and a second heterogeneous graph corresponding to recruitment behavior information, and determining vector information of each node in each heterogeneous graph; determining a grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph; determining a grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph; for each grid label, selecting N target position nodes and M target job seeker nodes; determining an aggregation vector of each grid label according to the vector information of the N target position nodes corresponding to each grid label and the vector information of the M target job seeker nodes; and performing clustering processing on the plurality of aggregation vectors, and determining a target number of target clusters. According to the embodiment of the invention, the flexibility of region division can be improved, and the accuracy of estimating the flow in the divided region is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a grid label classification method, device, electronic device and storage medium. Background Art

[0002] The increasing popularity of the internet and the maturation of data processing technology have brought increasing convenience to users' lives and work. For example, in the recruitment industry, both recruiters and job seekers can use recruitment software to find the candidates they need. Currently, to improve the convenience of recruitment software for both recruiters and job seekers, recruitment software can provide business services such as information promotion, such as increasing the placement and number of impressions of job posts on the homepage.

[0003] Those skilled in the art have discovered that the distribution quantity and density of different positions in different cities, districts, counties and other regions are slightly different. Therefore, different regions use basically the same recommendation strategy when recommending positions, which leads to obvious differences in the promotion effect of the same position in different regions. In related technologies, different regions are often finely divided to facilitate more targeted information promotion. However, since the number of positions covered in the finely divided regions is small, it is more difficult to make accurate traffic estimates for the regions. In other words, it is difficult to estimate the number of potential job seekers available for recruitment. Therefore, the current recommendation strategy is poorly matched with the region, resulting in mediocre recommendation results. Summary of the Invention

[0004] In light of this, embodiments of the present application provide a grid tag classification method, apparatus, electronic device, and storage medium that enhance the flexibility of regional division and improve the accuracy of traffic estimation within the divided regions. Furthermore, based on accurate division results, subsequent sales costs can be reduced and grid sales and recommendations can be integrated, facilitating the subsequent convenience of advertising sales and the effectiveness of advertising recommendations and promotions.

[0005] An embodiment of the present application provides a method for classifying grid labels, comprising: obtaining a first heterogeneous graph corresponding to job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to recruitment behavior information within a preset time period, and determining vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes; obtaining recruitment information within a preset time period, and determining, based on the recruitment information, a grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph; and, based on the job-seeking behavior information and the recruitment information, determining the grid labels associated with each position node included in the first heterogeneous graph and the second heterogeneous graph. a grid label associated with each job seeker node; for each grid label, N target position nodes and M target job seeker nodes are selected from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label; an aggregation vector of each grid label is determined according to the vector information of the N target position nodes and the vector information of the M target job seeker nodes corresponding to each grid label; a clustering process is performed on the multiple aggregation vectors, and a target number of target clustering clusters is determined, wherein the target number is determined according to the classification fineness required by the business application, and different target clustering clusters correspond to different grid categories.

[0006] Optionally, according to the method of an embodiment of the present application, the job-seeking behavior information includes job-viewing behavior and resume-submitting behavior; obtaining a first heterogeneous graph corresponding to the job-seeking behavior information within a preset time period includes: determining the first relationship information between the job seeker node and the job node based on the job-viewing behavior and resume-submitting behavior; generating a first heterogeneous graph based on the first relationship information between the job seeker node and the job node.

[0007] Optionally, according to the method of an embodiment of the present application, the recruitment behavior information includes resume viewing behavior and recruitment communication behavior; obtaining the second heterogeneous graph corresponding to the recruitment behavior information within a preset time period includes: determining the second relationship information between the job seeker node and the position node based on the resume viewing behavior and the recruitment communication behavior; generating a second heterogeneous graph based on the second relationship information between the job seeker node and the position node.

[0008] Optionally, according to the method of an embodiment of the present application, the vector information of each node in the first isomerous graph and the vector information of each node in the second isomerous graph are determined, including: generating a training sample set based on the first isomerous graph and the second isomerous graph; training a pre-constructed vector generation model based on the training sample set, and obtaining a target vector generation model after meeting preset training conditions; and determining the vector information of each node in the first isomerous graph and the vector information of each node in the second isomerous graph through the target vector generation model.

[0009] Optionally, according to the method of an embodiment of the present application, a training sample set is generated according to the first heterogeneous graph and the second heterogeneous graph, including: walking the first heterogeneous graph according to the first meta-path, and walking the second heterogeneous graph according to the second meta-path to obtain a walking sequence set, the walking sequence set including multiple walking sequences corresponding to each meta-path; generating a training sample set according to the order of nodes in each walking sequence in the walking sequence set and K nodes adjacent to the node, where K is a positive integer.

[0010] Optionally, according to the method of an embodiment of the present application, for each grid label, N target position nodes and M target job seeker nodes are selected from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, including: for each grid label, obtaining the first position node associated with the grid label and the first job seeker node associated with the grid label in the first heterogeneous graph to obtain a first node set; and obtaining the second position node associated with the grid label and the second job seeker node associated with the grid label in the second heterogeneous graph to obtain a second node set; for each grid label, N target position nodes are selected from the position nodes included in the first node set and the second node set, and M target job seeker nodes are selected from the job seeker nodes included in the first node set and the second node set.

[0011] Optionally, according to the method of an embodiment of the present application, multiple aggregation vectors are clustered and a target number of target clusters are determined, including: randomly selecting a target number of first aggregation vectors from the multiple aggregation vectors as the initial cluster centers of the target number of clusters; according to a preset clustering algorithm, iteratively updating each cluster and the cluster center of the cluster until a preset stop iteration condition is reached to obtain the target cluster; wherein the second aggregation vector is an aggregation vector other than the first aggregation vector from the multiple aggregation vectors.

[0012] An embodiment of the present application provides a grid label classification device, including: an acquisition module, used to obtain a first heterogeneous graph corresponding to job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to recruitment behavior information within a preset time period, and determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes.

[0013] The acquisition module is also used to obtain recruitment information within a preset time period, and determine the grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph based on the recruitment information; and determine the grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph based on the job-seeking behavior information and the recruitment information.

[0014] The processing module is used to select, for each grid label, N target position nodes and M target job seeker nodes from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label.

[0015] The processing module is further configured to determine an aggregate vector for each grid label based on the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label.

[0016] The processing module is further used to perform clustering processing on multiple aggregation vectors and determine a target number of target clusters, wherein the target number is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid categories.

[0017] An embodiment of the present application provides an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the above method are implemented.

[0018] 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 above method are implemented.

[0019] An embodiment of the present application provides a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, the steps of the above method are implemented.

[0020] According to an embodiment of the present application, a heterogeneous graph is constructed based on the job-seeking behavior information and published recruitment information acquired within a preset time period. This graph conveniently represents the relationship between job seekers and various positions from the perspective of job seekers, and the relationship between various positions and job seekers from the perspective of recruiters. The published recruitment information includes the grids where the positions are located, thereby conveniently determining the grid labels associated with each position node in each heterogeneous graph. The existing job-seeking behavior information includes the positions that the job seekers have interacted with. Therefore, combined with the recruitment information, the grid labels corresponding to the job seeker nodes in each heterogeneous graph can be conveniently determined. Next, for each grid label, the associated N target position nodes and M target job seeker nodes are found. Combined with the vector information, an aggregation vector for each grid label is determined. Finally, the number of positions required to be covered within each grid category is determined, and the target number of clusters is flexibly determined. Multiple aggregation vectors are automatically processed into a target number of highly differentiated clusters of different categories, thereby improving the accuracy of traffic estimation within the divided area and the efficiency of grid classification processing, achieving more accurate grid classification and merging. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0022] Figure 1 It is a schematic diagram of the system architecture of an embodiment of the present application.

[0023] Figure 2 This is a flowchart of the grid label classification method according to an embodiment of the present application.

[0024] Figure 3 This is a schematic diagram of a first isomeric graph provided in an embodiment of the present application.

[0025] Figure 4 This is a schematic diagram of a second isomeric graph provided in an embodiment of the present application.

[0026] Figure 5 This is a schematic diagram of a walking heterogeneous graph provided in an embodiment of the present application.

[0027] Figure 6 This is a schematic diagram of another first isomeric graph provided in an embodiment of the present application.

[0028] Figure 7 This is a schematic diagram of another second isomeric graph provided in an embodiment of the present application.

[0029] Figure 8 This is a flow chart of another grid label classification method provided in an embodiment of the present application.

[0030] Figure 9 This is a structural block diagram of a grid tag classification device according to an embodiment of the present application.

[0031] Figure 10 Schematic diagram of an electronic device used to implement the grid tag classification method of 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 those of ordinary skill in the art without creative work are within the scope of protection of this application.

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

[0034] The embodiments of the present application relate to terminal devices and / or servers. Those skilled in the art will appreciate that the embodiments of the present application may be implemented as a system, apparatus, device, method, computer-readable storage medium, or computer program product. Therefore, the present disclosure may be specifically implemented in at least one of the following forms: complete hardware, complete software, or a combination of hardware and software. According to the embodiments of the present application, the present application seeks protection for a grid tag classification method, apparatus, electronic device, and storage medium. Figure 1 A schematic diagram of a system architecture of an embodiment of the present application is shown. Figure 1As shown, the system includes a terminal device 102 and a server 104. The terminal device 102 may include at least one of the following: a smartphone, a tablet computer, a laptop computer, a desktop computer, a smart TV, various wearable devices, an augmented reality (AR) device, a virtual reality (VR) device, and the like. The terminal device 102 may utilize recruitment software. For example, a client may be installed on the terminal device 102. This client may be a client dedicated to performing specific functions (such as an application app), or a client embedded with multiple applets (with different functions), or a client logged in through a browser. Job seekers may perform operations on the terminal device 102. For example, they may open the terminal device 102, use the recruitment software, and perform other operations such as viewing positions and posting resumes. Alternatively, recruiters may also perform other operations on the terminal device, such as posting recruitment information, viewing resumes, and communicating with job seekers. Recruitment information may include positions, as well as the job location and job function information for the position. The grid tags corresponding to the positions may include the job location and job function information.

[0035] Server 104 provides storage and retrieval functions, for example, storing job seekers' resumes and recruiters' job postings. After receiving a command from a job seeker or recruiter, terminal device 102 sends a request message containing the command to server 104. Server 104 then performs the corresponding processing and returns the result to terminal device 102. Through a series of data processing and information exchange, user commands are fulfilled. Based on these commands, server 104 can obtain information about job seekers' behavior and the convenience of recruitment software, which also helps recruiters find suitable employees.

[0036] Based on this, the server 104 can regularly obtain job-seeking behavior information, recruitment behavior information and recruitment information, and analyze the relationship between the information, so as to categorize the grid tags, accurately classify and merge the grid tags, and improve the flexibility and accuracy of the grid tag classification process.

[0037] It should be noted that the acquisition, storage, use, and processing of data in the embodiments of this application comply with the relevant provisions of national laws and regulations.

[0038] The following describes the classification method of grid labels provided by the embodiment of the present application with reference to the accompanying drawings. Figure 2 A flow chart of a method for classifying grid labels according to an embodiment of the present application is shown. The method may include the following steps 201 to 205 .

[0039] Step 201: Obtain a first heterogeneous graph corresponding to job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to recruitment behavior information within a preset time period, and determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes.

[0040] Step 202: Obtain recruitment information within a preset time period, and determine, based on the recruitment information, a grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph; and, based on the job-seeking behavior information and the recruitment information, determine a grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph.

[0041] Step 203 : For each grid label, select N target position nodes and M target job seeker nodes from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label.

[0042] Step 204 : Determine an aggregate vector for each grid label based on the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label.

[0043] Step 205 : clustering the multiple aggregated vectors and determining a target number of target clusters, wherein the target number is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid categories.

[0044] The above steps are described in detail below with reference to specific embodiments, as shown below.

[0045] In an embodiment of the present application, the execution subject of the grid label classification method can be an electronic device with data processing capabilities, such as a server. Optionally, the server can be a physical server, a virtual server, or a cloud server. No specific limitation is imposed on the form of the server.

[0046] Regarding step 201 above, the obtained job-seeking behavior information and recruitment behavior information are analyzed to classify the grid tags. Optionally, a preset time period can serve as the time interval between each execution of the grid tag classification method. The preset time period can be set to a desired length based on application requirements. The preset time period can also be set based on actual application requirements, and the preset time period is not specifically limited herein.

[0047] For example, job-seeking behavior information and recruitment behavior information can be obtained from the service equipment of the recruitment software respectively. For example, the job seekers' behaviors of viewing positions and submitting resumes in the recruitment software, and the recruiters' behaviors of viewing resumes for a certain position and communicating with job seekers can all be recorded in the service equipment.

[0048] A heterogeneous graph includes multiple nodes and multiple edges, wherein nodes with associated relationships can be connected through the edges.

[0049] In one example, a job seeker submits a resume to a desired position during the job search process. Job seeker nodes and positions with a job search relationship can then be found using job search behavior information. This job search relationship is also known as first relationship information. Next, connections are established between different job seeker nodes and position nodes based on the job search behavior information, thereby building a first heterogeneous graph. Each resume corresponds to a job seeker node, and each position corresponds to a position node.

[0050] In another example, when a recruiter is looking for a suitable candidate for a position, they may review the resumes of one or more applicants and proactively communicate with them. This proactive communication includes, but is not limited to, asking for basic information, answering questions, and extending interview and job invitations. Thus, using recruitment behavior information, nodes and positions with a recruitment relationship can be found. This recruitment relationship is also the second relationship information. Next, connections are established between different position nodes and job seeker nodes based on the recruitment behavior, thereby building a second heterogeneous graph.

[0051] By establishing the first heterogeneous graph, it is possible to conveniently express the relationship between job seekers and various positions from the perspective of job seekers, and by establishing the second heterogeneous graph, it is possible to conveniently express the relationship between various positions and job seekers from the perspective of recruiters, thereby providing a good data foundation for improving the accuracy of grid label classification processing.

[0052] Furthermore, in the embodiments of the present application, for the constructed first and second heterogeneous graphs, vector information for each node in the first and second heterogeneous graphs can be determined, respectively. Specifically, vector information for each job seeker node and position node in each heterogeneous graph can be determined. Optionally, a pre-trained vector generation model can be used to output vector information for each node in each heterogeneous graph.

[0053] Next, in step 202, the grid labels associated with each position node and each job seeker node in each heterogeneous graph can be determined as the grid labels to be classified. For example, if a published job posting includes information such as the position, the location of the position, and the job function, the grid labels associated with the corresponding position node can be determined by extracting the location and job function information from the job posting.

[0054] Based on this, by obtaining recruitment information, the grid label associated with each position node in each heterogeneous graph can be easily and accurately determined.

[0055] Existing job search behavior information includes the positions that job seekers have interacted with. Combined with recruitment information, this allows for easy identification of grid labels corresponding to job seeker nodes in heterogeneous graphs. For example, the positions that each job seeker has interacted with can be identified, and the grid labels corresponding to these positions can be assigned to the job seeker nodes.

[0056] Next, referring to the above step 203 , for each grid label, N target position nodes and M target job seeker nodes may be obtained from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph.

[0057] For example, for each grid label, the position nodes associated with the grid label can be determined from the position nodes included in the first and second heterogeneous graphs. All position nodes associated with the grid label are used as a selection range, and N position nodes are obtained from the selection range as target position nodes. N is a preset positive integer, such as 200 or 300. The value of N can be set based on actual application requirements, and there is no specific limitation on the value of N.

[0058] In another example, for each grid label, all job seeker nodes associated with the grid label in the first and second heterogeneous graphs can be selected as another selection range, and M job seeker nodes can be obtained from the range as target job seeker nodes. M is a preset positive integer, such as 200 or 300. The value of M can be set based on actual application requirements, and there is no specific limitation on the value of M.

[0059] Optionally, when selecting the target position node and the target job seeker node, a random selection method or a method of preferentially selecting nodes with higher node degrees may be used. No specific restriction is imposed on the selection method.

[0060] After selecting N target job nodes and M target job seeker nodes, next, in step 204 , the aggregate vector of each grid label can be determined based on the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label.

[0061] For example, vector aggregation refers to fusing multiple vectors into a single vector. Optionally, the aggregation method can be selected based on the specific data distribution, including but not limited to calculating the average, weighted average, maximum or minimum, and geometric mean. By aggregating vectors, the complexity of subsequent grid label clustering can be reduced.

[0062] Taking the calculation of the average value as an example, for each grid label, the average of the vector information of N target position nodes and the vector information of M target job seeker nodes can be calculated to obtain the aggregate vector of the grid label.

[0063] After obtaining the aggregation vector of each grid label, that is, after obtaining the aggregation vectors corresponding to the multiple grid labels to be classified, next, in step 205 , clustering processing can be performed on the multiple aggregation vectors to generate a target number of target clusters.

[0064] Exemplarily, each target cluster includes multiple aggregation vectors, and each aggregation vector corresponds one-to-one to a grid label. Therefore, the aggregation vector included in each target cluster can correspond to a category, and the grid labels corresponding to the multiple aggregation vectors included in the target cluster all belong to this category, which is conducive to achieving accurate division of the grid label categories.

[0065] Based on this, different target clusters can be obtained, which is conducive to accurately classifying and merging grid labels and improving the distinction between grid labels of different categories.

[0066] Optionally, the target number is determined based on the classification fineness required by the business application. For example, the more positions that need to be covered in a grid of a category, the smaller the classification fineness, and accordingly, the smaller the value of the target number can be; the fewer positions that need to be covered in a grid of a category, the greater the classification fineness, and accordingly, the larger the value of the target number can be.

[0067] According to the embodiments of the present application, the number of positions that need to be covered in each category of grid can be determined based on an accurate estimate of the required traffic, the target number of clusters can be flexibly determined, and multiple aggregation vectors can be automatically processed into a target number of clusters with high discrimination and different categories. This not only improves the efficiency of grid classification processing, but also allows for more accurate classification and merging of grids, which is conducive to improving the accurate estimation of traffic in the divided area. Based on this, in the subsequent job promotion business, it is conducive to accurately estimating the promotion effect that can be generated by job promotion, and, when estimating the value of job promotion business, it is conducive to more accurate calculation of promotion value.

[0068] In some embodiments of the present application, job-seeking behavior information includes job-viewing behavior and resume-submitting behavior; based on this, a first heterogeneous graph corresponding to the job-seeking behavior information within a preset time period is obtained, including: determining the first relationship information between the job seeker node and the job node based on the job-viewing behavior and resume-submitting behavior; generating a first heterogeneous graph based on the first relationship information between the job seeker node and the job node.

[0069] For example, based on job posting and resume submission behavior, it can be determined which resumes interact with which jobs. This not only reflects the job seeker's submission history, but also reflects the job seeker's interest in selecting jobs. For example, a job seeker's interest in the job they submitted their resume to may be higher than their interest in the job they only viewed.

[0070] As a specific example, Figure 3 This is a schematic diagram of a first heterogeneous graph provided by an embodiment of the present application. The job seeker's behavior of viewing positions and submitting resumes is associated with the job seeker and the position (J). The job seeker (U) and the position (J) are used as nodes, and viewing and submitting are used as edges to construct the first heterogeneous graph of the job seeker and the position. Figure 3 As shown, U1, U2, U3, U4, and U5 represent different job seeker nodes, and J1, J2, J3, and J4 represent different positions.

[0071] In the embodiment of the present application, the advantage of constructing a heterogeneous graph is that after establishing connections between nodes with associated relationships, the relationship between job seekers and positions can be conveniently expressed, so that relevant feature information of the grid labels can be extracted therefrom, and the grid labels can be accurately classified and merged.

[0072] In some embodiments of the present application, recruitment behavior information includes resume viewing behavior and recruitment communication behavior; based on this, obtaining the second heterogeneous graph corresponding to the recruitment behavior information within a preset time period includes: determining the second relationship information between the job seeker node and the position node based on the resume viewing behavior and the recruitment communication behavior; generating the second heterogeneous graph based on the second relationship information between the job seeker node and the position node.

[0073] For example, based on resume viewing and recruitment communication behaviors, it can be determined which resumes a recruiter has viewed for a position and which job seekers he has interacted with. This not only reflects the recruiter's recruitment history, but also reflects the recruiter's interest in selecting job seekers. For example, a recruiter may be more interested in job seekers he has communicated with than in those whose resumes he has only viewed. As a specific example, Figure 4 This is a schematic diagram of a second heterogeneous graph provided by an embodiment of the present application. The job seeker's behavior of viewing positions and submitting resumes is used to associate the job seeker with the position (J). With the position (J) and the job seeker (U) as nodes and viewing and submitting as edges, the first heterogeneous graph of the job seeker and the position is constructed. Figure 4 As shown, U1, U2, U3, and U4 represent different job seeker nodes, and J1, J2, J3, J4, and J5 represent different positions.

[0074] In the embodiment of the present application, the advantage of constructing a heterogeneous graph is that after establishing connections between nodes with associated relationships, the relationship between job seekers and positions can be conveniently expressed, so that relevant feature information of the grid labels can be extracted therefrom, and the grid labels can be accurately classified and merged.

[0075] In some embodiments of the present application, the steps of determining the vector information of each node in the first heterogeneous graph and the vector information of each node in the second heterogeneous graph may specifically refer to the following steps 301 to 303.

[0076] Step 301: Generate a training sample set according to the first heterogeneous graph and the second heterogeneous graph.

[0077] Step 302 : training the pre-built vector generation model according to the training sample set, and obtaining the target vector generation model after the preset training conditions are met.

[0078] Step 303: Determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph through the target vector generation model.

[0079] In the embodiment of the present application, the training samples in the training sample set are determined by walking sequences obtained by walking the first heterogeneous graph and the second heterogeneous graph respectively according to different meta paths.

[0080] Exemplarily, a meta-path refers to a specific path pattern between different types of nodes and edges in a heterogeneous graph. For example, the path from the job seeker node to the position node and then to the job seeker node can be used as a meta-path. By setting different meta-paths, multiple random walks can be performed in the heterogeneous graph to generate a walk sequence that reflects the complex relationship between nodes. Each walk sequence actually represents a node relationship under a certain association pattern and can be used as training data for subsequent models. For example, the meta-paths pre-constructed for the first heterogeneous graph are: (1) job seeker → position → job seeker; (2) position → job seeker → position; the meta-paths pre-constructed for the second heterogeneous graph are: (1) job seeker → position → job seeker; (2) position → job seeker → position. Corresponding to each meta-path, multiple walk sequences are found through walks.

[0081] In some embodiments, pre-built vector generation models include, but are not limited to, support for various graph neural network model frameworks such as Deep Graph Library (DGL), PyG, and NeuGraph. Optionally, in embodiments of the present application, a pre-built vector generation model based on the DGL model framework is used to perform graph embedding on the relationships between nodes in a heterogeneous graph.

[0082] For example, the pre-built vector generation model may be a model developed based on a Skip-gram model, where the Skip-gram model is a prediction model in natural language processing that can be used to learn continuous vector representations of words.

[0083] The preset training condition can be that the function value of the loss function tends to a preset value, or that the training cycle (Epoch) of the model reaches a preset number of times, etc. The specific preset training condition can be set according to actual training needs.

[0084] After model training is complete, a target vector generation model is obtained. The target vector generation model can output vector information for each node in the first heterogeneous graph and each node in the second heterogeneous graph, respectively. Specifically, the target vector generation model outputs vector information for each node in the first heterogeneous graph and each node in the second heterogeneous graph, i.e., the vector information for the job seeker node and the position node in the first heterogeneous graph, and the vector information for the job seeker node and the position node in the second heterogeneous graph.

[0085] According to an embodiment of the present application, by using a heterogeneous graph generated based on job-seeking behavior information and recruitment information to generate training samples and train a vector generation model, it is beneficial to deeply explore the potential relationship between job seekers' job-seeking intentions and grid labels, and thus help improve the accuracy of classification processing of grid labels corresponding to positions.

[0086] In addition, by setting different meta-paths and combining them with flexible adjustment of the wandering strategy, the required training samples are generated and the vector generation model is trained. This makes it possible to conveniently and efficiently mine feature information from job-seeking behavior information and recruitment information, and convert it into a reliable vectorized representation to support the subsequent classification processing of grid labels.

[0087] In some embodiments of the present application, generating a training sample set based on a first heterogeneous graph and a second heterogeneous graph includes: navigating the first heterogeneous graph according to a first meta-path, and navigating the second heterogeneous graph according to a second meta-path to obtain a walk sequence set, wherein the walk sequence set includes multiple walk sequences corresponding to each meta-path; and generating a training sample set based on the order of nodes within each walk sequence in the walk sequence set and K adjacent nodes, where K is a positive integer.

[0088] Exemplarily, the meta-paths pre-set for the first heterogeneous graph are described as first meta-paths, for example, meta-path (1) and meta-path (2), and the meta-paths pre-set for the second heterogeneous graph are described as second meta-paths, for example, meta-path (3) and meta-path (4).

[0089] Figure 5 This is a schematic diagram of a walking heterogeneous graph provided in an embodiment of the present application, combined with Figure 5 As shown. Combined Figure 5 As shown in (a), based on the meta-path (1) job seeker → position → job seeker, a walk in the first heterogeneous graph can obtain the walk sequence: U1-J1-U3. Figure 5 As shown in (b), based on the meta-path (3) Position → Applicant → Position, a single walk can be performed on the second heterogeneous graph, resulting in the walk sequence: J1-U1-J3. For each first meta-path, multiple walk sequences can be obtained after walking the first heterogeneous graph. The walk sequences corresponding to each first meta-path are not listed here one by one.

[0090] In some embodiments, the connection lines between the nodes in the first heterogeneous graph and the second heterogeneous graph are undirected edges. Based on this, a wandering direction can be randomly selected at any node.

[0091] Optionally, in the process of walking on each heterogeneous graph, walking parameters can be pre-set, such as the walking depth and the number of times each node in the walking process is used as the walking starting point. Continuing with the first heterogeneous graph and meta-path (1) as an example, if the walking depth is 1, the walking sequence obtained can be U1-J1-U3. If the walking depth is 2, after obtaining U1-J1-U3, continue to use U3 as the starting point and continue to walk in the first heterogeneous graph according to meta-path (1). The final walking sequence is, for example, U1-J1-U3-J4-U4. In the embodiment of the present application, optionally, the walking depth and the number of times each node in the walking process is used as the walking starting point can be set according to application requirements, and no specific restrictions are made here.

[0092] Optionally, in the embodiment of the present application, the weight between the job seeker node and the position node can also be set. Figure 6 As shown in the figure, V represents the job seeker's viewing behavior, A represents the job seeker's submission behavior, and the number after the colon represents the weight of the edge. Figure 7 As shown in the figure, H represents the recruiter's browsing behavior, and C represents the recruiter's communication delivery behavior. The larger the weight, the greater the probability that the node will be selected during the walk. This approach allows for full exploration of the behavioral intentions of both job seekers and recruiters. Using a vector generation model, each node is converted into a vectorized representation, supporting the subsequent classification of grid labels.

[0093] In each walk sequence, assuming a node represents a job seeker, its K neighboring nodes may include positions it has interacted with. By using the sequential relationship between these nodes and their neighbors as training samples, we can further analyze the association between resumes and positions and provide training data for subsequent vector generation models. The value of K can be adjusted based on actual needs. Generally, a larger K allows a node and more of its neighboring nodes to participate in training, generating a richer set of training samples.

[0094] According to the embodiments of the present application, meta-path-based random walks can capture the multi-dimensional relationships between different types of nodes in a heterogeneous graph. This allows the training sample set to be not limited to specific relationships between nodes of a certain type, but to reflect information from multiple dimensions, such as resumes and positions. This facilitates the extraction of useful information from on-the-job behavior information and recruitment information, supporting the subsequent classification of grid labels.

[0095] In some embodiments of the present application, optionally, the training sample set includes multiple positive training samples and multiple negative training samples. Optionally, a training sample set is generated based on the order of the nodes in each walk sequence and the K nodes in the neighborhood of each node. The positive training samples and negative training samples can be obtained by referring to the following steps. Specifically, for each walk sequence, the current node is combined with each of the K nodes in the neighborhood of the current node to obtain a positive training sample, the current node is any node in the walk sequence, and in each positive training sample, the current node is the input sample node and the other node is the output sample node.

[0096] Exemplarily, the walk sequence set includes multiple walk sequences corresponding to each first-element path and multiple walk sequences corresponding to each second-element path. For each walk sequence, a positive training sample is obtained by combining the current node with each of the K nodes in the neighborhood of the current node.

[0097] For each input sample node in the positive training sample, L nodes are randomly obtained from the multiple walk sequences included in the walk sequence set as output negative training sample nodes, and the L output negative training sample nodes are respectively combined with the input sample nodes to obtain L negative training samples corresponding to each positive training sample.

[0098] As a specific example, let's take the walk sequence U1-J1-U3-J4-U4 as an example. U1, J1, U3, J4, and U4 are each used as the center point, and each node in their K neighborhood is combined. For example, with K = 2, U1 is the center point, and U1's neighborhood nodes are J1 and U3. Thus, two positive training samples can be obtained: U1-J1 and U1-U3. R2 is the input sample node in the positive training sample.

[0099] J1 is the center point, and its neighboring nodes are U1, U3, and J4. Thus, three positive training samples can be obtained: J1-U1, J1-U3, and J1-J4. Among them, J1 is the input sample node in the positive training sample.

[0100] U3 is the center point, and its neighboring nodes are U1, J1, J4, and U4. Thus, four positive training samples can be obtained: U3-U1, U3-J1, U3-J4, and U3-U4. Among them, U3 is the input sample node in the positive training sample.

[0101] Based on the same positive training sample generation logic, three positive training samples can be obtained when J4 is the center point, and two positive training samples can be obtained when U4 is the center point. They are not listed here one by one.

[0102] Usually, there are many nodes in heterogeneous graphs. In order to speed up model training, negative training samples can be generated by negative sampling. Taking meta-path (1) as an example, U1~U5 are used as starting points respectively. According to the preset depth, the wandering paths are obtained, and a total of 5 paths are obtained. Among all the nodes included in the 5 paths, nodes are selected to generate negative training samples. For example, for the positive training sample U1-J1, 5 nodes are randomly selected from all U and J as negative training samples according to probability: U1-F5, U1-U2, U1-U3, U1-J5 and U1-J4. It can be seen that in the process of selecting negative training samples, the frequency of node appearance is used as the probability of being selected as a negative training sample. That is, the higher the frequency of node appearance, the greater the probability of being selected as a negative training sample. The advantage of this setting is to strengthen the training of high-frequency nodes and reduce the risk of the model being biased by high-frequency nodes.

[0103] It is understandable that the number of negative training samples corresponding to each positive training sample is set according to actual application requirements and performance, and is not specifically limited here.

[0104] According to an embodiment of the present application, vectors of each node are generated based on a heterogeneous graph and a target vector generation model, which can effectively mine useful information from job-seeking behavior information and recruitment information and convert it into a vectorized representation to support the subsequent steps of classifying and processing the grid labels.

[0105] In some embodiments of the present application, before selecting N target job nodes and M target job seeker nodes for each grid label, the grid labels and nodes may be screened. For example, nodes with low node degrees in each heterogeneous graph may be screened out, and the grid labels to be classified are determined based on the nodes included in the first and second heterogeneous graphs after the node screening process.

[0106] For example, the degree of a node refers to the number of edges directly connected to a node. In this embodiment of the present application, the edges in the first and second heterogeneous graphs are all undirected edges, and the degree of a node refers to the total number of edges connected to the node. For example, if a node is connected to three other nodes, its degree is 3.

[0107] For example, the preset screening condition may be: screening out nodes with node degrees less than or equal to 9 in each heterogeneous graph. Therefore, the processing for the first heterogeneous graph is specifically to retain only nodes with node degrees greater than 10 and screen out nodes with node degrees less than or equal to 9. After completing the screening process, the grid labels associated with the remaining job seeker nodes and the grid labels associated with the position nodes in the first heterogeneous graph are determined as grid labels to be classified, and then the target job seeker nodes or target position nodes are selected for the grid labels. Based on the same processing method, the grid labels corresponding to the remaining job seeker nodes and the grid labels corresponding to the position nodes in the second heterogeneous graph can be determined as grid labels to be classified, and then the target job seeker nodes or target position nodes are selected for the grid labels, which will not be repeated here.

[0108] Since a larger node degree indicates more connections between that node and other nodes (e.g., job seeker nodes, job position nodes), filtering out nodes with smaller degrees makes the relationships between nodes in the heterogeneous graphs more informative, helping to extract more useful information. Therefore, using target job seeker nodes and target job position nodes with larger node degrees as the basis for grid label classification improves the accuracy of subsequent grid label classification processing.

[0109] In an embodiment of the present application, for each grid label, N target position nodes and M target job seeker nodes are selected from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph. Specifically, the following steps may be included: for each grid label, the first position node associated with the grid label and the first job seeker node associated with the grid label in the first heterogeneous graph are obtained to obtain a first node set; and the second position node associated with the grid label and the second job seeker node associated with the grid label in the second heterogeneous graph are obtained to obtain a second node set; for each grid label, N target position nodes are selected from the position nodes included in the first node set and the second node set, and M target job seeker nodes are selected from the job seeker nodes included in the first node set and the second node set.

[0110] For example, in the first and second heterogeneous graphs, each job seeker node and each position node are associated with a grid label, that is, each node is associated with a grid label, wherein the grid labels associated with different nodes can be different or the same. Thus, for each grid label, the job seeker node and position node associated with the grid label can be easily determined, wherein the node set consisting of the first position node associated with the grid label and the first job seeker node associated with the grid label in the first heterogeneous graph can be described as a first node set; and the node set consisting of the second position node associated with the grid label and the second job seeker node associated with the grid label in the second heterogeneous graph can be described as a second node set.

[0111] Alternatively, the number of target position nodes to be selected from the first and second node sets can be pre-set. For example, the first and second node sets can each select N / 2 nodes. Alternatively, the number of target position nodes to be selected from the first and second node sets can be determined based on the ratio of the number of position nodes included in each of the first and second node sets. For example, the greater the number of position nodes, the greater the number of target position nodes that can be selected. This is not specifically limited here. Alternatively, the target job seeker nodes can be selected based on the same selection logic as the target position nodes. This is not further described here.

[0112] In an embodiment of the present application, by selecting a certain number of target job seeker nodes and target position nodes for each grid label to be classified and used to determine the aggregation vector of the grid label, it helps to retain key information for grid classification processing and improve the efficiency of clustering processing.

[0113] In some embodiments of the present application, after selecting N target position nodes and M target job seeker nodes, an aggregate vector for each grid label can be determined based on the vector information of the N target position nodes and the vector information of the M target job seeker nodes corresponding to each grid label. Compared to directly calculating the vectors corresponding to the N target position nodes and the M target job seeker nodes corresponding to the grid label, using the aggregate vector to represent the grid label can reduce the computational difficulty and can also more accurately summarize the features included in the grid label in a multi-dimensional manner.

[0114] After obtaining the aggregation vector for each grid label, that is, after obtaining the aggregation vectors corresponding to the multiple grid labels to be classified, the multiple aggregation vectors are clustered and a target number of target clusters are determined. Specifically, the following steps can be referred to: randomly selecting a target number of first aggregation vectors from the multiple aggregation vectors as the initial cluster centers of the target number of clusters; iteratively updating each cluster and the cluster center of the cluster according to a preset clustering algorithm until a preset stop iteration condition is met to obtain the target cluster; wherein the second aggregation vector is the aggregation vector other than the first aggregation vector from the multiple aggregation vectors.

[0115] Optionally, a preset clustering algorithm may be used to cluster the multiple aggregation vectors to generate a target number of target clusters. Exemplarily, the preset clustering algorithm may include, but is not limited to, any one of a variety of clustering methods such as Mini Batch K-means clustering, hierarchical clustering, and density clustering.

[0116] For example, the clustering process is introduced by taking Mini Batch K-means clustering as an example to generate a target number of clusters.

[0117] The target number is determined based on the classification precision required by the business application. For example, for target number Z, Z clustering vectors are randomly selected from multiple clustering vectors, and the Z first clustering vectors are respectively determined as Z cluster centers. At this point, the Z cluster centers are the initial cluster centers.

[0118] For the convenience of description, Z randomly selected aggregation vectors are described as first aggregation vectors, and the remaining aggregation vectors in the multiple aggregation vectors are described as second aggregation vectors.

[0119] Next, during each iterative calculation, Y second clustering vectors are randomly selected from all second clustering vectors. For each of the Y second clustering vectors, the cluster center associated with each second clustering vector is determined. The cluster center associated with each second clustering vector is the cluster center closest to the second clustering vector among the Z cluster centers. The position of the cluster center associated with the second clustering vector is then updated based on the second clustering vector, thus completing one iterative calculation.

[0120] It is understandable that in the entire clustering process, each second aggregation vector is selected only once, and each iterative calculation only selects the second aggregation vector that has not been calculated. After the preset stopping iteration condition is reached, Z target cluster centers are obtained.

[0121] Exemplarily, the preset condition for stopping iteration may be that all second aggregation vectors participate in the iterative calculation, or that the number of iterative calculations reaches a preset number of iterations, or that the position change of each cluster center is less than a threshold, etc.

[0122] Based on this, the grid labels can be accurately classified and merged to obtain Z different target clusters with high discrimination. Different target clusters correspond to different grid label categories, which can flexibly respond to different classification requirements.

[0123] In an embodiment of the present application, the target number Z can be determined based on business application requirements. For example, the target number can be determined with reference to the following business application requirements: (1) the proportion of grids in the cluster that are delivered within the cluster; (2) the proportion of correlations between grids within the cluster; and (3) the balance of traffic distribution between clusters.

[0124] For example, business application requirement (1) can be calculated from the perspective of job seekers. Each job seeker node and position node has an associated grid tag. Specifically, when each position is posted, the recruitment information may include not only position information but also job function information. By analyzing the recruitment information, the association between positions and grid tags can be found. For example, different positions may correspond to different grid tags, or they may correspond to the same grid tag.

[0125] During the job search process, job seekers can submit resumes to desired positions. This can be simplified into a grid-to-grid delivery relationship. Based on this, the calculation method can be: for the grid labels associated with the current job seeker and the grid labels associated with the position for which the job seeker submitted their resume, if they are still within the current cluster, then count them. Finally, divide this count by the total number of submissions, which is the proportion of grids in the cluster that have submitted to the grid within the cluster.

[0126] For example, the business application requirement (2) can be calculated based on the job seeker's perspective.

[0127] The calculation formula can be shown as formula (1).

[0128]

[0129] Where C is all the grids in the current cluster and G is the grid.

[0130] For the grids in each cluster, assuming that the cluster includes n grids, there are n*n combinations. Within a certain time range (such as within 30 days), statistics show that there are delivery relationships between grids in the n*n combinations in each cluster. The number of possible combinations / (n*n) is the proportion of correlation between grids in the cluster.

[0131] In another example, business application requirements can also be calculated from the perspective of recruiters. For example, the proportion of job seekers in the cluster who chat about positions in the cluster grid is calculated. Each job seeker node and position node has an associated grid tag, as well as a grid tag. Recruiters have resumes they like for a certain position and can actively communicate with the job seekers corresponding to the resumes they like. Therefore, based on the grid tags associated with the above-mentioned positions and the grid tags associated with the job seekers corresponding to the resumes, the communication behavior of the recruiters can be simplified into a recruitment relationship between grids. Based on this, the calculation method can be: for the grid tags associated with the current position and the grid tags associated with the job seekers communicated based on the position are still in the current cluster, then count them, and finally use the count / all submissions to get the proportion of job seekers in the cluster who chat about positions in the cluster grid.

[0132] In addition, from the perspective of the recruiter, we can also calculate the proportion of chat correlations between grids within a cluster. The calculation method is the same as the calculation logic for calculating the proportion of correlations between grids within a cluster from the perspective of the job seeker, so we will not repeat it here.

[0133] In some embodiments, the business application requirement (3) can be to focus on recommending a certain grid to job seekers and increase the exposure of the grid, and then the grid size can be determined from the perspective of job seekers; when it is necessary to increase the amount of communication between the recruiter and the job seeker, the grid size can be determined from the perspective of the recruiter.

[0134] According to the embodiments of the present application, based on the accurate estimation of the required traffic, the number of positions that need to be covered in the grid of each category can be determined, the target number of clusters can be flexibly determined, and multiple aggregation vectors can be automatically processed into a target number of clusters with high discrimination and different categories. This not only improves the efficiency of grid classification processing, but also allows for more accurate classification and merging of grids, which is conducive to improving the accurate estimation of traffic in the divided area.

[0135] The above describes the implementation of the embodiments of the present application and the advantages brought by them through multiple embodiments. In order to more clearly introduce the embodiments of the present application, the specific processing process of the embodiments of the present application is described in detail below with reference to specific examples.

[0136] Figure 8 This is a flow chart of another grid label classification method provided by an embodiment of the present application, combined with Figure 8 As shown, the solution may include the following steps 801 to 810.

[0137] Step 801: Obtain job-seeking behavior data and recruitment behavior information, and construct a first heterogeneous graph and a second heterogeneous graph.

[0138] Step 802: Obtain the meta-path.

[0139] Step 803 , traverse the job seeker nodes and position nodes in the first heterogeneous graph and the second heterogeneous graph respectively, and walk in the first heterogeneous graph and the second heterogeneous graph respectively according to a preset meta-path, and generate multiple walk sequences corresponding to each meta-path.

[0140] Step 804: Sampling each walk sequence to generate multiple positive training samples and multiple negative training samples associated with each positive training sample.

[0141] Step 805: construct a vector generation model.

[0142] Step 806: Train the vector generation model based on the positive training samples and the negative training samples.

[0143] Step 807: Train for multiple cycles. When the loss of the vector generation model no longer decreases, stop training to obtain the target vector generation model.

[0144] After stopping training, the model parameters of the current model are saved to obtain the target vector generation model.

[0145] Step 808 , filtering out nodes whose node degrees in the first and second heterogeneous graphs are less than or equal to a preset threshold, and determining grid labels for the nodes included in the first and second heterogeneous graphs after the node filtering process.

[0146] Step 809 : For each grid label, obtain N target position nodes and M target job seeker nodes according to the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph.

[0147] Step 810 : Determine an aggregate vector for each grid label based on the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label.

[0148] Step 811 : performing clustering processing on the multiple aggregation vectors to generate a target number of target clusters.

[0149] Optionally, the number of targets is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid label categories.

[0150] Exemplarily, among multiple aggregation vectors, a target number of first aggregation vectors are randomly selected as the initial cluster centers of the target number of clusters; according to a preset clustering algorithm, each cluster and the cluster center of the cluster are iteratively updated until a preset stop iteration condition is reached to obtain a target cluster; wherein, the second aggregation vector is an aggregation vector other than the first aggregation vector among the multiple aggregation vectors.

[0151] According to the embodiments of the present application, based on the accurate estimation of the required traffic, the number of positions that need to be covered in the grid of each category can be determined, the target number of clusters can be flexibly determined, and multiple aggregation vectors can be automatically processed into a target number of clusters with high discrimination and different categories. This not only improves the efficiency of grid classification processing, but also allows for more accurate classification and merging of grids, which is conducive to improving the accurate estimation of traffic in the divided area.

[0152] Corresponding to the method embodiment of the present application, the present application also provides a classification device for grid labels, such as Figure 9 As shown, the grid label classification device 900 includes: an acquisition module 901 and a processing module 902.

[0153] Acquisition module 901 is used to obtain a first heterogeneous graph corresponding to job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to recruitment behavior information within a preset time period, and determine the vector information of each node in the first heterogeneous graph and the vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes.

[0154] The acquisition module 901 is also used to obtain recruitment information within a preset time period, and determine the grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph based on the recruitment information; and determine the grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph based on the job-seeking behavior information and the recruitment information.

[0155] Processing module 902 is used to select, for each grid label, N target position nodes and M target job seeker nodes from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label.

[0156] The processing module 902 is further configured to determine an aggregate vector for each grid label based on the vector information of the N target position nodes and the vector information of the M target job seeker nodes corresponding to each grid label.

[0157] The processing module 902 is further configured to perform clustering processing on the multiple aggregation vectors and determine a target number of target clusters, wherein the target number is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid categories.

[0158] In some embodiments of the present application, optionally, job-seeking behavior information includes job posting behavior and resume submission behavior.

[0159] The processing module 902 is further configured to determine first relationship information between the job seeker node and the job position node based on the job viewing behavior and the resume submission behavior.

[0160] The processing module 902 is further configured to generate a first heterogeneous graph according to the first relationship information between the job seeker nodes and the position nodes.

[0161] In some embodiments of the present application, optionally, the recruitment behavior information includes resume viewing behavior and recruitment communication behavior.

[0162] The processing module 902 is further configured to determine second relationship information between the job seeker node and the position node based on the resume viewing behavior and the recruitment communication behavior.

[0163] The processing module 902 is further configured to generate a second heterogeneous graph according to the second relationship information between the job seeker nodes and the position nodes.

[0164] In some embodiments of the present application, optionally, the processing module 902 is further configured to generate a training sample set according to the first heterogeneous graph and the second heterogeneous graph.

[0165] The processing module 902 is further configured to train the pre-built vector generation model according to the training sample set, and obtain a target vector generation model after a preset training condition is met.

[0166] The processing module 902 is further configured to determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph by using a target vector generation model.

[0167] In some embodiments of the present application, optionally, the processing module 902 is further configured to navigate the first heterogeneous graph according to the first meta-path, and to navigate the second heterogeneous graph according to the second meta-path, to obtain a walk sequence set, where the walk sequence set includes multiple walk sequences corresponding to each meta-path.

[0168] The processing module 902 is further configured to generate a training sample set according to the order of nodes in each walking sequence in the walking sequence set and K nodes adjacent to the node, where K is a positive integer.

[0169] In some embodiments of the present application, the acquisition module 901 is optionally further configured to, for each grid label, acquire the first position node and the first job seeker node associated with the grid label in the first heterogeneous graph to obtain a first node set. Furthermore, the acquisition module 901 is configured to acquire the second position node and the second job seeker node associated with the grid label in the second heterogeneous graph to obtain a second node set.

[0170] The processing module 902 is further configured to select, for each grid label, N target job nodes from the job nodes included in the first node set and the second node set, and M target job seeker nodes from the job seeker nodes included in the first node set and the second node set.

[0171] In some embodiments of the present application, optionally, the processing module 902 is further configured to randomly select a target number of first aggregation vectors from a plurality of aggregation vectors as initial cluster centers of a target number of clusters.

[0172] The processing module 902 is further configured to iteratively update each cluster and its cluster center according to a preset clustering algorithm until a preset stopping condition is reached to obtain a target cluster.

[0173] The second aggregation vector is an aggregation vector other than the first aggregation vector among the multiple aggregation vectors.

[0174] It can be understood that the grid tag classification device of the embodiment of the present application can correspond to the execution entity of the grid tag classification method provided by the embodiment of the present application. The specific details of the operation and / or function of each module / unit of the grid tag classification device can be found in the description of the corresponding parts of the grid tag classification method provided by the above-mentioned embodiment of the present application. For the sake of brevity, they will not be repeated here.

[0175] The electronic device in the embodiment of the present application can be a user terminal device, a server, other computing devices, or a cloud server. Figure 10 A schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application is shown. The electronic device may include a processor 1001 and a memory 1002 storing computer program instructions. When the processor 1001 executes the computer program instructions, the process or function of any of the above-mentioned embodiments is implemented.

[0176] Specifically, the processor 1001 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. The memory 1002 may include a large-capacity memory for data or instructions. For example, the memory 1002 may be at least one of the following: a hard disk drive (HDD), a read-only memory (ROM), a random access memory (RAM), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, a universal serial bus (USB) drive, or other physical / tangible memory storage device. For another example, the memory 1002 may include a removable or non-removable (or fixed) medium. For another example, the memory 1002 may be inside or outside the integrated gateway disaster recovery device. The memory 1002 may be a non-volatile solid-state memory. In other words, the memory 1002 typically includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with computer-executable instructions, and when the software is executed (such as by one or more processors), the operations described in the method of the embodiments of the present application may be performed. The processor 1001 implements the process or function of any method in the above embodiments by reading and executing computer program instructions stored in the memory 1002.

[0177] In one example, Figure 10The electronic device shown may also include a communication interface 1003 and a bus 1010. The processor 1001, memory 1002, and communication interface 1003 are connected via bus 1010 and communicate with each other. The communication interface 1003 is primarily used to implement communication between the various modules, devices, units, and / or equipment in the embodiments of the present application. The bus 1010 includes hardware, software, or both, and can couple the components of the online data traffic metering device to each other. For example, the bus may include at least one of the following: an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industrial Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industrial 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-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Area Bus (VLB) bus, or other suitable buses. The bus 1010 may include one or more buses. Although the embodiments of the present application describe or illustrate a specific bus, the embodiments of the present application may consider any suitable bus or interconnection method.

[0178] In combination with the method in the above embodiments, an embodiment of the present application also provides a computer-readable storage medium, which stores computer program instructions. When the computer program instructions are executed by a processor, they implement the process or function of any method in the above embodiments.

[0179] In addition, an embodiment of the present application further provides a computer program product, which stores computer program instructions. When the computer program instructions are executed by a processor, the process or function of any one of the methods in the above embodiments is implemented.

[0180] The flowcharts and / or block diagrams of the methods, devices, systems and computer program products of the embodiments of the present application are described above by way of example, and various aspects thereof are described. It should be understood that each box in the flowchart and / or block diagram or a combination thereof may be implemented by computer program instructions, or may be implemented by dedicated hardware that performs a specified function or action, or may be implemented by a combination of dedicated hardware and computer instructions. For example, these computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine that enables these instructions executed by such a processor to enable the implementation of the functions / actions specified in each box in the flowchart and / or block diagram or a combination thereof. Such a processor may be a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.

[0181] The functional blocks shown in the block diagram of the embodiment of the present application 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 (ASIC), appropriate firmware, a plug-in, a function card, etc.; when implemented in software, it is a program or code segment used to perform the required task. The program or code segment can be stored in a memory or transmitted on a transmission medium or a communication link via a data signal carried in a carrier. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0182] It should be noted that the present application is not limited to the specific configurations and processes described above or shown in the figures. The above is only a specific implementation method of the present application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the described system, device, module or unit can refer to the corresponding process in the method embodiment without further description. It should be understood that the scope of protection of the present application is not limited to this. Any technician familiar with the technical field can 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 scope of protection of the present application.

Claims

1. A grid label classification method, characterized in that: include: Obtaining a first heterogeneous graph corresponding to the job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to the recruitment behavior information within the preset time period, and determining vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes; Acquire recruitment information within the preset time period, and determine, based on the recruitment information, a grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph; and, based on the job-seeking behavior information and the recruitment information, determine a grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph; For each of the grid labels, N target position nodes and M target job seeker nodes are selected from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label; Determine an aggregate vector of each grid label according to the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label; Clustering is performed on the plurality of aggregation vectors, and a target number of target clusters is determined, wherein the target number is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid categories.

2. The method according to claim 1, characterized in that The job-seeking behavior information includes job-viewing behavior and resume-submitting behavior; The step of obtaining a first heterogeneous graph corresponding to the job-seeking behavior information within a preset time period includes: Determine first relationship information between the job seeker node and the job position node according to the job viewing behavior and the resume submitting behavior; The first heterogeneous graph is generated according to the first relationship information between the job seeker nodes and the position nodes.

3. The method according to claim 1, characterized in that: The recruitment behavior information includes resume viewing behavior and recruitment communication behavior; Obtaining the second heterogeneous graph corresponding to the recruitment behavior information within the preset time period includes: Determining second relationship information between the job seeker node and the position node according to the resume viewing behavior and the recruitment communication behavior; The second heterogeneous graph is generated according to the second relationship information between the job seeker nodes and the position nodes.

4. The method according to claim 1, characterized in that: The determining of the vector information of each node in the first heterogeneous graph and the vector information of each node in the second heterogeneous graph includes: Generate a training sample set according to the first heterogeneous graph and the second heterogeneous graph; Training the pre-constructed vector generation model according to the training sample set, and obtaining the target vector generation model after reaching a preset training condition; The target vector generation model is used to determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph.

5. The method according to claim 4, characterized in that The step of generating a training sample set according to the first heterogeneous graph and the second heterogeneous graph includes: According to the first meta-path, wander the first heterogeneous graph, and according to the second meta-path, wander the second heterogeneous graph to obtain a wandering sequence set, wherein the wandering sequence set includes a plurality of wandering sequences corresponding to each meta-path; The training sample set is generated according to the order of nodes in each walking sequence in the walking sequence set and K nodes adjacent to the node, where K is a positive integer.

6. The method according to claim 1, characterized in that For each of the grid labels, N target position nodes and M target job seeker nodes are selected from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, including: For each of the grid labels, obtaining a first position node associated with the grid label and a first job seeker node associated with the grid label in the first heterogeneous graph to obtain a first node set; and obtaining a second position node associated with the grid label and a second job seeker node associated with the grid label in the second heterogeneous graph to obtain a second node set; For each grid label, N target job nodes are selected from the job nodes included in the first node set and the second node set, and M target job seeker nodes are selected from the job seeker nodes included in the first node set and the second node set.

7. The method according to claim 1, characterized in that The clustering process is performed on the plurality of the aggregated vectors, and determining a target number of target clusters, including: Randomly selecting a target number of first aggregation vectors from the multiple aggregation vectors as initial cluster centers of the target number of cluster clusters; According to a preset clustering algorithm, each of the clusters and the cluster centers of the clusters are iteratively updated until a preset stop iteration condition is reached to obtain the target cluster; The second aggregation vector is an aggregation vector among the multiple aggregation vectors except the first aggregation vector.

8. A grid label classification device, characterized in that: include: An acquisition module, used to acquire a first heterogeneous graph corresponding to the job-seeking behavior information within a preset time period and a second heterogeneous graph corresponding to the recruitment behavior information within the preset time period, and determine vector information of each node in the first heterogeneous graph and vector information of each node in the second heterogeneous graph, wherein the first heterogeneous graph includes first relationship information between job seeker nodes and position nodes, and the second heterogeneous graph includes second relationship information between job seeker nodes and position nodes; The acquisition module is further used to acquire recruitment information within the preset time period, and determine the grid label associated with each position node included in the first heterogeneous graph and the second heterogeneous graph according to the recruitment information; and determine the grid label associated with each job seeker node included in the first heterogeneous graph and the second heterogeneous graph according to the job-seeking behavior information and the recruitment information; A processing module, configured to select, for each of the grid labels, N target position nodes and M target job seeker nodes from the nodes respectively included in the first heterogeneous graph and the second heterogeneous graph, wherein the N target position nodes include the position nodes associated with the grid label, and the M target job seeker nodes include the job seeker nodes associated with the grid label; The processing module is further used to determine the aggregate vector of each grid label according to the vector information of the N target job nodes and the vector information of the M target job seeker nodes corresponding to each grid label; The processing module is further used to perform clustering processing on the multiple aggregation vectors and determine a target number of target clusters, wherein the target number is determined according to the classification fineness required by the business application, and different target clusters correspond to different grid categories.

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.

11. A computer program product, characterized in that It comprises computer program instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 7.

Images