Data encoding method and apparatus, electronic device, and medium

Abstract

The application discloses a data coding method and device, electronic equipment and medium, and relates to the technical field of computers. The method comprises the following steps: constructing a first graph network based on a first search path of a job-seeking end and first position information, wherein the first position information is position information in recruitment information clicked by the job-seeking end; constructing a second graph network based on a second search path of a recruitment end and second position information, wherein the second position information is job-seeking information in a resume clicked by the recruitment end; fusing the first graph network and the second graph network to obtain a fused graph network; sampling nodes in the fused graph network based on a random walk strategy to obtain a plurality of node sequences; and performing vector learning on the plurality of node sequences to generate coding vectors of the nodes in the fused graph network. The method combines the double-sided information of the recruitment end and the job-seeking end to construct a graph network, improves the relevance between search words, improves the accuracy of coding, and is beneficial to the improvement of the effect of subsequent task algorithms.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a data encoding method, apparatus, electronic device, and medium. Background Technology

[0002] Embedding refers to the method of representing an object using a numerical vector. Graphs are a fundamental and commonly used structure, and many scenarios can be abstracted into a graph structure, such as the relationship between users and items in e-commerce platforms. Graph Embedding (graph encoding) refers to representing nodes in a graph as low-dimensional dense vectors. It requires that similar nodes in the graph also have similar low-dimensional representations. The resulting representation vectors can be used for downstream tasks such as node classification, link prediction, visualization, or graph reconstruction. In search scenarios, user search terms construct a graph network structure through clicks, submissions, etc. By randomly walking from a node, multiple paths are established, and the search terms on each path form a sentence. Then, the vector representation of the search terms (i.e., the encoded vector) is trained by retrieving contextual information from the search terms. Currently, most search scenarios involve matching people with things. However, recruitment scenarios have their own unique characteristics, involving matching people with people. Existing technical solutions only consider the search terms of job seekers, neglecting those of recruiters. This leads to the loss of semantic information related to recruitment, resulting in a loss of text diversity and low accuracy of the trained vector representation, which is detrimental to downstream data mining tasks. Summary of the Invention

[0003] To solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, embodiments of the present invention provide a data encoding method, apparatus, electronic device, and medium.

[0004] In a first aspect, embodiments of the present invention provide a data encoding method, including:

[0005] Based on the first search path of the job seeker and the first job information related to the job seeker, a first graph network is constructed, wherein the first job information is the job information in the recruitment information clicked by the job seeker.

[0006] A second graph network is constructed based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, wherein the second job information is the job application information in the resume information clicked on the recruitment terminal;

[0007] By fusing the first graph network and the second graph network, a fused graph network is obtained;

[0008] Nodes in the fused graph network are sampled based on a random walk strategy to obtain multiple node sequences;

[0009] Vector learning is performed using the multiple node sequences to generate the encoding vectors of the nodes in the fusion graph network.

[0010] Optionally, constructing a first graph network based on the first search path of the job seeker and the first job information related to the job seeker includes: in response to an input operation of the job seeker, obtaining the search terms input by the job seeker; obtaining the first search path of the job seeker based on the order of the search terms input by the job seeker; in response to a click operation of the job seeker, determining the recruitment information clicked by the job seeker and obtaining the first job information in the recruitment information; using the search terms in the first search path and the first job information as nodes, connecting adjacent search terms in the first search path, and connecting the first job information and the search terms corresponding to the first job information to construct the first graph network.

[0011] Optionally, the method further includes: determining the weight of the connection edge between search terms in the first graph network as a first weight; determining the weight of the connection edge between the first job information and the search term corresponding to the first job information in the first graph network as a second weight, wherein the second weight is less than the first weight.

[0012] Optionally, constructing a second graph network based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal includes: in response to an input operation of the recruitment terminal, obtaining the search terms input by the recruitment terminal; obtaining the second search path of the recruitment terminal based on the order of the search terms input by the recruitment terminal; in response to a click operation of the recruitment terminal, determining the resume information clicked by the recruitment terminal, and obtaining the second job information in the resume information; using the search terms in the second search path and the second job information as nodes, connecting the search terms in the second search path, and connecting the second job information and the search terms corresponding to the second job information to construct a second graph network.

[0013] Optionally, the method further includes: determining the weight of the connection edge between search terms in the second graph network as a third weight; determining the weight of the connection edge between the second job information and the search term corresponding to the second job information in the second graph network as a fourth weight, wherein the fourth weight is less than the third weight.

[0014] Optionally, the first weight is equal to the third weight.

[0015] Optionally, the fusion of the first graph network and the second graph network includes: for a first target search term connected to the first job information in the first graph network, taking the job seeker who inputs the first target search term as the target job seeker, determining whether the recruiter who published the first job information clicked on the resume information of the target job seeker; if the recruiter who published the first job information clicked on the resume information of the target job seeker, then taking the search term input by the recruiter who published the first job information as the second target search term; connecting the first target search term in the first graph network and the second target search term in the second graph network to fuse the first graph network and the second graph network.

[0016] Optionally, the method further includes: determining the weight of the connecting edge connecting the first graph network and the second graph network as a fifth weight, wherein the fifth weight is greater than the first weight and the fifth weight is greater than the third weight.

[0017] Optionally, the step of sampling nodes in the fused graph network based on a random walk strategy to obtain multiple node sequences includes: normalizing the weights of the connecting edges in the fused graph network; and sampling nodes in the fused graph network based on the random walk strategy and the normalized weights to obtain multiple node sequences.

[0018] Optionally, normalizing the weights of the connecting edges in the fused graph network includes: determining the starting node and the adjacent nodes of the starting node; normalizing the weights of the connecting edges between the starting node and the adjacent nodes, and determining that the direction of the connecting edges is from the starting node to the adjacent node;

[0019] The step of sampling nodes in the fused graph network based on the random walk strategy and normalized weights to obtain multiple node sequences includes: sampling nodes in the fused graph network based on the random walk strategy, the direction of the connecting edges, and the normalized weights to obtain multiple node sequences.

[0020] In a second aspect, embodiments of the present invention provide a data encoding apparatus, comprising:

[0021] The first construction module is used to construct a first graph network based on the first search path of the job seeker and the first job information related to the job seeker, wherein the first job information is the job information in the recruitment information clicked by the job seeker.

[0022] The second construction module is used to construct a second graph network based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, wherein the second job information is the job application information in the resume information clicked by the recruitment terminal;

[0023] The fusion module is used to fuse the first graph network and the second graph network to obtain a fused graph network;

[0024] The sampling module is used to sample nodes in the fused graph network based on a random walk strategy to obtain multiple node sequences;

[0025] The encoding module is used to perform vector learning using the multiple node sequences to generate the encoding vectors of the nodes in the fused graph network.

[0026] Optionally, the first construction module is configured to: in response to an input operation from the job seeker, obtain the search terms input by the job seeker; based on the order of the search terms input by the job seeker, obtain a first search path of the job seeker; in response to a click operation from the job seeker, determine the job posting clicked by the job seeker, and obtain the first job information in the job posting; use the search terms and the first job information in the first search path as nodes, connect adjacent search terms in the first search path, and connect the first job information and the search terms corresponding to the first job information to construct a first graph network.

[0027] Optionally, the first construction module is further configured to: determine the weight of the connection edge between search terms in the first graph network as a first weight; determine the weight of the connection edge between the first job information and the search term corresponding to the first job information in the first graph network as a second weight, wherein the second weight is less than the first weight.

[0028] Optionally, the second construction module is configured to: in response to an input operation from the recruitment terminal, obtain the search terms input by the recruitment terminal; based on the order of the search terms input by the recruitment terminal, obtain the second search path of the recruitment terminal; in response to a click operation from the recruitment terminal, determine the resume information clicked by the recruitment terminal, and obtain the second job information in the resume information; use the search terms in the second search path and the second job information as nodes, connect the search terms in the second search path, and connect the second job information and the search terms corresponding to the second job information to construct a second graph network.

[0029] Optionally, the second construction module is further configured to: determine the weight of the connection edge between search terms in the second graph network as a third weight; determine the weight of the connection edge between the second job information and the search term corresponding to the second job information in the second graph network as a fourth weight, wherein the fourth weight is less than the third weight.

[0030] Optionally, the fusion module is further configured to: for a first target search term connected to the first job information in the first graph network, take the job seeker who inputs the first target search term as the target job seeker, determine whether the recruiter who published the first job information clicked on the resume information of the target job seeker; if the recruiter who published the first job information clicked on the resume information of the target job seeker, then take the search term input by the recruiter who published the first job information as the second target search term; connect the first target search term in the first graph network and the second target search term in the second graph network to fuse the first graph network and the second graph network.

[0031] Optionally, the fusion module is further configured to: determine the weight of the connecting edge connecting the first graph network and the second graph network as a fifth weight, wherein the fifth weight is greater than the first weight and the fifth weight is greater than the third weight.

[0032] Optionally, the apparatus further includes a normalization module for: normalizing the weights of the connection edges in the fused graph network;

[0033] The sampling module is used to sample nodes in the fused graph network based on a random walk strategy and normalized weights to obtain multiple node sequences.

[0034] Optionally, the sampling module is configured to: determine the starting node and the adjacent nodes of the starting node; normalize the weight of the connecting edge between the starting node and the adjacent nodes, and determine that the direction of the connecting edge is from the starting node to the adjacent node;

[0035] The sampling module is used to sample nodes in the fused graph network based on the random walk strategy, the direction of the connecting edges, and the normalized weights, to obtain multiple node sequences.

[0036] Thirdly, embodiments of the present invention provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the data encoding method of any embodiment of the present invention.

[0037] Fourthly, embodiments of the present invention provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the data encoding method of any embodiment of the present invention.

[0038] One embodiment of the above invention has the following advantages or beneficial effects:

[0039] The data encoding method of this invention constructs a first graph network based on a first search path from a job seeker and the first job information clicked by the job seeker, and constructs a second graph network based on a second search path from a recruiter and the second job information in the resume information clicked by the recruiter. Then, the first and second graph networks are merged to obtain a fused graph network. A random walk strategy is used to sample nodes in the fused graph network to obtain multiple node sequences. These multiple node sequences are then used for vector learning to generate the encoding vectors of the nodes in the fused graph network. This method combines bilateral information from both the recruiter and job seeker sides to construct the graph network, improving the relevance between search terms, increasing encoding accuracy, and contributing to the improvement of subsequent task algorithms.

[0040] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0041] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0042] Figure 1 A schematic flowchart of a data encoding method according to an embodiment of the present invention is shown;

[0043] Figure 2 A schematic diagram of the data encoding method for constructing a first graph network according to an embodiment of the present invention is shown;

[0044] Figure 3 A schematic diagram of a first graph network according to an embodiment of the present invention is shown;

[0045] Figure 4 A schematic diagram illustrating the process of constructing a second graph network using the data encoding method of an embodiment of the present invention is shown;

[0046] Figure 5 A schematic diagram of a second graph network according to an embodiment of the present invention is shown;

[0047] Figure 6 A schematic diagram of a fusion graph network according to an embodiment of the present invention is shown;

[0048] Figure 7A A schematic diagram of a data encoding method according to another embodiment of the present invention is shown;

[0049] Figure 7B and Figure 7C The diagrams show the connection edge weights of the fused graph network before and after weight normalization.

[0050] Figure 8 A schematic diagram of the structure of a data encoding device according to an embodiment of the present invention is shown;

[0051] Figure 9A schematic diagram of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0052] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0053] Figure 1 A flowchart of a data encoding method according to an embodiment of the present invention is shown, as follows: Figure 1 As shown, the method includes:

[0054] Step S101: Based on the first search path of the job seeker and the first job information related to the job seeker, construct the first graph network. The first job information is the job information in the recruitment information clicked by the job seeker.

[0055] The primary search path on the job-seeking platform can include one or more search terms entered by the job seeker on the job-seeking software. The primary job information relevant to the job seeker refers to job information in job postings that the job seeker clicks to view or submits an application to.

[0056] Figure 2 A flowchart illustrating the construction of a first graph network according to an embodiment of the present invention is shown. Figure 2 As shown, step S101, which involves constructing the first graph network, may include:

[0057] Step S201: In response to the input operation of the job seeker, obtain the search terms entered by the job seeker;

[0058] Step S202: Based on the order of the search terms entered by the job seeker, obtain the first search path of the job seeker;

[0059] Step S203: In response to the click operation on the job seeker's end, determine the job posting clicked on the job seeker's end, and obtain the first job information in the job posting;

[0060] Step S204: Using the search terms and first job information in the first search path as nodes, connect adjacent search terms in the first search path, and connect the first job information and the search terms corresponding to the first job information to construct a first graph network.

[0061] refer to Figure 2If job seeker A enters search term "query 1" on a job search app and clicks on a job posting (the job information in that posting is post 1), and then enters new search terms "query 2," "query 3," and "query 4," and after entering search term "query 4," the user applies to another job posting (post 2), then job seeker A's search path is: query 1, query 2, query 3, query 4. The job information related to job seeker A includes post 1 and post 2. Then, if... Figure 3 As shown, connect query 1, query 2, query 3, and query 4 in sequence, connect query 1 and post 1, and connect query 4 and post 2 to obtain the first graph network.

[0062] Step S102: Based on the second search path of the recruitment end and the second job information related to the recruitment end, construct a second graph network. The second job information is the job application information in the resume information clicked on the recruitment end.

[0063] The second search path on the recruitment platform can include one or more search terms entered by the recruiter on the job search software. Second job information related to the recruitment platform refers to job information that the recruiter clicks to view in the process of creating a job or job information in a resume that they click to invite.

[0064] Figure 4 A flowchart illustrating the construction of a second graph network according to an embodiment of the present invention is shown. Figure 4 As shown, step S102, which involves constructing the second graph network, includes:

[0065] Step S401: In response to the input operation on the recruitment end, obtain the search terms entered on the recruitment end;

[0066] Step S402: Based on the order of the search terms entered on the recruitment platform, obtain the second search path on the recruitment platform;

[0067] Step S403: In response to the click operation on the recruitment terminal, determine the resume information clicked on the recruitment terminal, and obtain the second job information in the resume information;

[0068] Step S404: Using the search terms in the second search path and the second job information as nodes, connect the search terms in the second search path, and connect the second job information and the search terms corresponding to the second job information to construct a second graph network.

[0069] refer to Figure 5If recruiter B enters search term "query 5" on a job search app and clicks on a resume (the job posting is post 3), and then enters new search terms "query 6" and "query 7," and the user, after entering search term "query 7," requests another resume with the job posting is post 4, then recruiter B's search path is: query 5, query 6, query 7. The job information related to recruiter B includes post 3 and post 4. Then, if... Figure 5 As shown, connect query 5, query 6, and query 7 in sequence, connect query 5 and post 3, and connect query 7 and post 4 to obtain the second graph network.

[0070] Step S103: Merge the first graph network and the second graph network to obtain the merged graph network.

[0071] The fusion of the first and second graph networks aims to integrate relevant information from both graphs, specifically the correlation between search terms in the first and second graph networks. In one scenario, if job seeker user1 searches and clicks on a job posting by recruiter user2, while recruiter user2 searches and clicks on user1's resume, then the search terms entered by user1 and user2 are strongly correlated. Therefore, the search terms can be connected in the graph, thus merging the first and second graph networks to obtain a fused graph network.

[0072] Therefore, the first graph network and the second graph network can be merged according to the following process:

[0073] For the first target search term connected to the first job information in the first graph network, the job seeker who enters the first target search term is taken as the target job seeker, and it is determined whether the recruiter who posted the first job information clicked on the resume information of the target job seeker.

[0074] If the recruiter who posted the first job information clicks on the resume information of the target job seeker, the search term entered by the recruiter who posted the first job information will be used as the second target search term.

[0075] Connect the first target search term in the first graph network and the second target search term in the second graph network to merge the first graph network and the second graph network.

[0076] refer to Figure 3 and Figure 5 In the first and second graph networks shown, job seeker A submitted the first job posting (post2) after entering the search term "query4". Recruiter B, who posted post2, then invited job seeker A after entering the search term "query7". Then, as follows... Figure 6 As shown, search terms query4 and query7 are connected, and the first graph network and the second graph network are merged.

[0077] In a fused graph network, the weights of connecting edges can be the same or different; this embodiment does not impose any restrictions on this.

[0078] Step S104: Sample the nodes in the fusion graph network based on the random walk strategy to obtain multiple node sequences.

[0079] Step S105: Use multiple node sequences to perform vector learning to generate the encoding vectors of nodes in the fusion graph network.

[0080] In steps S104-S105, vector representations of search terms in the fused graph network can be obtained based on network embedding methods such as the Deepwalk algorithm. The Deepwalk algorithm employs a random walk approach when constructing sentence context, using the co-occurrence relationships between nodes in the graph to learn node vector representations. This algorithm mainly consists of two parts: random walk and vector generation. First, a random walk algorithm is used to extract vertex sequences from the graph; then, using natural language processing principles, the generated vertex sequences are treated as sentences composed of words, and all sequences can be considered as a large corpus. Finally, natural language processing tools are used to represent each vertex as a vector of dimension d. As an example, the word2vec open-source model Gensim can be used for vector learning. Word2vec is a language model that learns semantic knowledge from large text corpora in an unsupervised manner and is widely used in natural language processing. Gensim is an open-source third-party toolkit used to learn topic vector representations of text hidden layers from raw, unstructured text in an unsupervised manner.

[0081] The data encoding method of this invention constructs a first graph network based on the first search path of the job seeker and the first job information clicked by the job seeker, and constructs a second graph network based on the second search path of the recruiter and the second job information in the resume information clicked by the recruiter. Then, the first graph network and the second graph network are fused to obtain a fused graph network. The nodes in the fused graph network are sampled based on a random walk strategy to obtain multiple node sequences. Vector learning is performed using these multiple node sequences to generate the encoding vectors of the nodes in the fused graph network. This method combines bilateral information from both the recruiter and job seeker sides to construct the graph network, improves the relevance between search terms, enhances the accuracy of encoding, and is beneficial to improving the performance of subsequent task algorithms.

[0082] Figure 7A A schematic diagram of a data encoding method according to another embodiment of the present invention is shown. Figure 7A As shown, the method includes:

[0083] Step S701: Based on the first search path of the job seeker and the first job information related to the job seeker, construct the first graph network, where the first job information is the job information in the recruitment information clicked by the job seeker.

[0084] Step S702: Determine the weights of the connecting edges between nodes in the first graph network;

[0085] Step S703: Based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, construct a second graph network. The second job information is the job application information in the resume information clicked on the recruitment terminal.

[0086] Step S704: Determine the weights of the connecting edges between nodes in the second graph network;

[0087] Step S705: Merge the first graph network and the second graph network to obtain the fused graph network;

[0088] Step S706: Determine the weights of the connecting edges that link the first graph network and the second graph network;

[0089] Step S707: Normalize the weights of the connection edges between all nodes in the fusion graph network and determine the direction of the connection edges;

[0090] Step S708: Based on the random walk strategy, the direction of the connecting edges, and the normalized weights, sample the nodes in the fusion graph network to obtain multiple node sequences. Use the multiple node sequences to perform vector learning to generate the encoding vectors of the nodes in the fusion graph network.

[0091] For step S701, the first search path on the job-seeking end can include one or more search terms entered by the job seeker on the job-seeking software. This first search path can be mined from the job seeker's search session and search logs. A session refers to an interaction process within a certain period of time, including actions such as "opening the App, browsing multiple pages, clicking on multiple functions, and submitting applications." The first job information related to the job seeker refers to the job information in the recruitment posts clicked or submitted by the job seeker. After obtaining the first search path and the first job information, the search terms in the first search path and the first job information are connected to generate the first graph network.

[0092] For step S702, the search terms in the job seeker's first search path are search terms that the job seeker changed themselves. These search terms are highly correlated, so the weight of the edges connecting search terms will be relatively high. The job postings submitted by the job seeker may contain one or multiple job postings, leading to a higher probability of semantic drift noise in the mined job information. Therefore, the weight of this part will be relatively low. Based on this principle, the weight of the edges connecting search terms in the first graph network can be determined as the first weight, and the weight of the edges connecting the first job posting and the corresponding search terms in the first graph network can be determined as the second weight. The second weight is less than the first weight. The larger the weight of an edge, the greater the correlation between the nodes at both ends of the edge. The larger the weight, the greater the probability of being sampled during random walk sampling.

[0093] For step S703, the second search path on the recruitment end can include one or more search terms entered by the recruiter on the job search software. This second search path can be mined from the recruiter's search session and search logs. A session refers to an interaction process within a certain period of time, including actions such as "opening the App, browsing multiple pages, clicking on multiple functions, and submitting applications." The second job information related to the recruitment end refers to the job information in the resumes clicked or submitted by the recruiter. After obtaining the second search path and the second job information, the search terms in the second search path and the second job information are connected to generate a second graph network.

[0094] Step S704 can refer to step S702. The weight of the connection edge between the search terms in the second graph network can be determined as the third weight; the weight of the connection edge between the second job information and the search term corresponding to the second job information in the second graph network can be determined as the fourth weight, and the fourth weight is less than the third weight.

[0095] Optionally, the first weight and the third weight are equal. In some other optional embodiments, the first weight and the third weight may also be unequal.

[0096] For step S705, please refer to Figure 1 The embodiments shown are not described in detail here.

[0097] For step S706, since the search terms at both ends of the connecting edge between the first graph network and the second graph network are strongly correlated, the weight of the connecting edge between the first graph network and the second graph network can be determined as the fifth weight, which is greater than the first weight and greater than the third weight.

[0098] For step S707, the sum of the weights of all connecting edges of a node in the fused graph network may be greater than 1 or less than 1. To ensure that the weights of the connecting edges of all nodes are scaled consistently, the weights of the connecting edges need to be normalized. When normalizing the weights, undirected connecting edges need to be converted into directed connecting edges. Therefore, when normalizing the weights, it is necessary to determine the starting node and its neighboring nodes, normalize the weights of the connecting edges between the starting node and its neighboring nodes, and determine the direction of the connecting edges to be from the starting node to the neighboring node. As a specific example, such as... Figure 7B and 7C As shown, where, Figure 7B A schematic diagram of the fusion network graph before weight normalization is shown. Figure 7C A schematic diagram of the fused network graph after normalization of weights is shown. Figure 7B In the diagram, the weight of the edge connecting query 1, query 2, query 3, and query 4 is a1; the weight of the edge connecting query 1 and post 1 is a2; and the weight of the edge connecting query 4 and post 2 is a2. Similarly, the weight of the edge connecting query 5, query 6, and query 7 is a1; the weight of the edge connecting query 5 and post 3 is a3; and the weight of the edge connecting query 7 and post 4 is a3. Here, a1 > a3 > a2. Taking node query1 as an example, query1 has two edges, so the weights of these two edges are normalized, as follows: Figure 7C As shown, the weight of the connection edge starting with query1 and ending with post1 is... The weight of the connecting edge with query1 as the starting node and query2 as the ending node is... If node post1 has only one connection edge, then the weight of that connection edge is normalized. The normalized connection edge starts at post1 and ends at query1, with a weight of a2.

[0099] For step S708, when performing a random walk on nodes, the weight and direction of the connecting edges need to be considered. The greater the weight, the greater the correlation between the two nodes. Therefore, when using a random walk strategy for sampling, nodes at both ends of connecting edges with greater weights are more likely to be sampled. Thus, node sampling is not completely random but introduces prior knowledge, improving the correlation between search terms, enhancing the accuracy and effectiveness of encoding, and improving the performance of downstream applications (such as search recall, search term rewriting, and synonym retrieval).

[0100] The data encoding method of this invention combines bilateral information from the job seeker and the recruiter to construct a graph network. At the same time, it assigns different weights to the connecting edges of the graph network according to the relevance between search terms, and performs random walk sampling based on the weights of the connecting edges. This introduces prior knowledge when generating node sequences, which can improve the graph network's ability to represent the textual relationships between search terms, achieve effective information extraction, and improve the accuracy and effectiveness of encoding.

[0101] Figure 8 A schematic diagram of the structure of a data encoding device according to an embodiment of the present invention is shown, as follows: Figure 8 As shown, the data encoding device 800 includes:

[0102] The first construction module 801 is used to construct a first graph network based on the first search path of the job seeker and the first job information related to the job seeker. The first job information is the job information in the recruitment information clicked by the job seeker.

[0103] The second construction module 802 is used to construct a second graph network based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal. The second job information is the job information in the resume information clicked on the recruitment terminal.

[0104] The fusion module 803 is used to fuse the first graph network and the second graph network to obtain a fused graph network.

[0105] The sampling module 804 is used to sample nodes in the fused graph network based on a random walk strategy to obtain multiple node sequences;

[0106] The encoding module 805 is used to learn vectors from multiple node sequences to generate encoded vectors for nodes in the fused graph network.

[0107] The data encoding device of this invention constructs a first graph network based on a first search path from a job seeker and the first job information clicked by the job seeker, and constructs a second graph network based on a second search path from a recruiter and the second job information in the resume information clicked by the recruiter. Then, the first and second graph networks are merged to obtain a fused graph network. Nodes in the fused graph network are sampled using a random walk strategy to obtain multiple node sequences. Vector learning is then performed using these multiple node sequences to generate the encoding vectors for the nodes in the fused graph network. This technique combines bilateral information from both the recruiter and job seeker sides to construct the graph network, improving the relevance between search terms, increasing encoding accuracy, and contributing to the improvement of subsequent task algorithms.

[0108] In an optional embodiment, the first construction module is configured to: in response to an input operation from the job seeker, obtain the search terms entered by the job seeker; based on the order of the search terms entered by the job seeker, obtain a first search path from the job seeker; in response to a click operation from the job seeker, determine the job posting clicked by the job seeker, and obtain the first job information in the job posting; use the search terms and the first job information in the first search path as nodes, connect adjacent search terms in the first search path, and connect the first job information and the search terms corresponding to the first job information to construct a first graph network.

[0109] In an optional embodiment, the first construction module is further configured to: determine the weight of the connection edge between search terms in the first graph network as a first weight; determine the weight of the connection edge between the first job information and the search term corresponding to the first job information in the first graph network as a second weight, wherein the second weight is less than the first weight.

[0110] In an optional embodiment, the second construction module is configured to: in response to an input operation from the recruitment terminal, obtain the search terms entered by the recruitment terminal; based on the order of the search terms entered by the recruitment terminal, obtain the second search path of the recruitment terminal; in response to a click operation from the recruitment terminal, determine the resume information clicked by the recruitment terminal, and obtain the second job information in the resume information; use the search terms in the second search path and the second job information as nodes, connect the search terms in the second search path, and connect the second job information and the search terms corresponding to the second job information to construct a second graph network.

[0111] In an optional embodiment, the second construction module is further configured to: determine the weight of the connection edge between search terms in the second graph network as a third weight; and determine the weight of the connection edge between the second job information and the search term corresponding to the second job information in the second graph network as a fourth weight, wherein the fourth weight is less than the third weight.

[0112] In an optional embodiment, the fusion module is further configured to: for a first target search term connected to the first job information in the first graph network, take the job seeker who inputs the first target search term as the target job seeker, determine whether the recruiter who published the first job information clicked on the resume of the target job seeker; if the recruiter who published the first job information clicked on the resume of the target job seeker, then take the search term input by the recruiter who published the first job information as the second target search term; connect the first target search term in the first graph network and the second target search term in the second graph network to fuse the first graph network and the second graph network.

[0113] In an optional embodiment, the fusion module is further configured to: determine the weight of the connection edge connecting the first graph network and the second graph network as the fifth weight, wherein the fifth weight is greater than the first weight and the fifth weight is greater than the third weight.

[0114] In an optional embodiment, the apparatus further includes a normalization module for: normalizing the weights of the connection edges in the fused graph network;

[0115] The sampling module is used to sample nodes in the fused graph network based on a random walk strategy and normalized weights to obtain multiple node sequences.

[0116] In an optional embodiment, the sampling module is used to: determine the starting node and the adjacent nodes of the starting node; normalize the weights of the connecting edges between the starting node and the adjacent nodes, and determine that the direction of the connecting edges is from the starting node to the adjacent nodes;

[0117] The sampling module is used to sample nodes in the fused graph network based on the random walk strategy, the direction of the connecting edges, and the normalized weights, to obtain multiple node sequences.

[0118] The above-described apparatus can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0119] This invention also provides an electronic device, such as... Figure 9 As shown, it includes a processor 901, a communication interface 902, a memory 903, and a communication bus 904. The processor 901, communication interface 902, and memory 903 communicate with each other via the communication bus 904.

[0120] Memory 903 is used to store computer programs;

[0121] When processor 901 executes a program stored in memory 903, it performs the following steps:

[0122] Based on the first search path of the job seeker and the first job information related to the job seeker, a first graph network is constructed, where the first job information is the job information in the recruitment information clicked by the job seeker.

[0123] Based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, a second graph network is constructed. The second job information is the job information in the resume information clicked on the recruitment terminal.

[0124] By merging the first graph network and the second graph network, a fused graph network is obtained.

[0125] Multiple node sequences are obtained by sampling nodes in the fusion graph network based on a random walk strategy;

[0126] Vector learning is performed using multiple node sequences to generate the encoded vectors of nodes in the fused graph network.

[0127] The communication bus 904 mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 904 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0128] Communication interface 902 is used for communication between the aforementioned terminal and other devices.

[0129] The memory 903 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor 901.

[0130] The processor 901 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0131] In another embodiment of the present invention, a computer-readable medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform any of the data encoding methods described in the above embodiments.

[0132] In another embodiment of the present invention, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the data encoding methods described in the above embodiments.

[0133] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0134] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0135] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0136] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A data encoding method, characterized in that, include: Based on the first search path of the job seeker and the first job information related to the job seeker, a first graph network is constructed, wherein the first job information is the job information in the recruitment information clicked by the job seeker. A second graph network is constructed based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, wherein the second job information is the job application information in the resume information clicked on the recruitment terminal; By fusing the first graph network and the second graph network, a fused graph network is obtained; Normalizing the weights of the connection edges in the fused graph network includes: Determine the starting node and its neighboring nodes; normalize the weights of the connecting edges between the starting node and its neighboring nodes, and determine the direction of the connecting edges as from the starting node to the neighboring node; Sampling of nodes in the fused graph network based on a random walk strategy to obtain multiple node sequences includes: sampling of nodes in the fused graph network based on the random walk strategy, the direction of the connecting edges, and the normalized weights to obtain multiple node sequences; Vector learning is performed using the multiple node sequences to generate the encoding vectors of the nodes in the fusion graph network.

2. The method according to claim 1, characterized in that, The first graph network is constructed based on the first search path from the job seeker and the first job information related to the job seeker, including: In response to the input operation of the job seeker, obtain the search terms entered by the job seeker; Based on the order of the search terms entered on the job search platform, the first search path of the job search platform is obtained; In response to a click operation on the job seeker's terminal, determine the job posting clicked on the job seeker's terminal and obtain the first job information in the job posting; Using the search terms and the first job information in the first search path as nodes, connect adjacent search terms in the first search path, and connect the first job information and the search terms corresponding to the first job information to construct a first graph network.

3. The method according to claim 2, characterized in that, The method further includes: The weight of the connection edge between search terms in the first graph network is determined as the first weight; The weight of the connection edge between the first job information and the search term corresponding to the first job information in the first graph network is determined as the second weight, which is less than the first weight.

4. The method according to claim 3, characterized in that, The second graph network is constructed based on the second search path from the recruitment platform and the second job information related to the recruitment platform, including: In response to input operations on the recruitment platform, obtain the search terms entered on the recruitment platform; Based on the order of the search terms entered on the recruitment platform, a second search path for the recruitment platform is obtained; In response to a click operation on the recruitment terminal, determine the resume information clicked on the recruitment terminal, and obtain the second job information from the resume information; By using the search terms in the second search path and the second job information as nodes, the search terms in the second search path are connected, and the second job information and the search terms corresponding to the second job information are connected, to construct a second graph network.

5. The method according to claim 4, characterized in that, The method further includes: The weights of the edges connecting search terms in the second graph network are determined as the third weights; The weight of the connection edge between the second job information and the search term corresponding to the second job information in the second graph network is determined as the fourth weight, which is less than the third weight.

6. The method according to claim 5, characterized in that, The first weight is equal to the third weight.

7. The method according to claim 5 or 6, characterized in that, The fusion of the first graph network and the second graph network includes: For the first target search term connected to the first job information in the first graph network, the job seeker who inputs the first target search term is taken as the target job seeker, and it is determined whether the recruitment end that published the first job information clicked on the resume information of the target job seeker. If the recruitment terminal that posted the first job information clicks on the resume information of the target job seeker, then the search term entered by the recruitment terminal that posted the first job information will be used as the second target search term. Connect the first target search term in the first graph network and the second target search term in the second graph network to merge the first graph network and the second graph network.

8. The method according to claim 7, characterized in that, The method further includes: The weight of the connecting edge between the first graph network and the second graph network is determined as the fifth weight, which is greater than the first weight and greater than the third weight.

9. A data encoding device, characterized in that, include: The first construction module is used to construct a first graph network based on the first search path of the job seeker and the first job information related to the job seeker, wherein the first job information is the job information in the recruitment information clicked by the job seeker. The second construction module is used to construct a second graph network based on the second search path of the recruitment terminal and the second job information related to the recruitment terminal, wherein the second job information is the job application information in the resume information clicked by the recruitment terminal; The fusion module is used to fuse the first graph network and the second graph network to obtain a fused graph network; The normalization module is used to normalize the weights of the connection edges in the fused graph network; The sampling module is used to determine the starting node and its adjacent nodes; The weights of the connecting edges between the starting node and adjacent nodes are normalized, and the direction of the connecting edges is determined to be from the starting node to the adjacent node. The sampling module is further configured to sample nodes in the fused graph network based on a random walk strategy to obtain multiple node sequences, including: sampling nodes in the fused graph network based on the random walk strategy, the direction of the connecting edges, and the normalized weights to obtain multiple node sequences; The encoding module is used to perform vector learning using the multiple node sequences to generate the encoding vectors of the nodes in the fusion graph network.

10. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-8.

11. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-8.

Images