Method, system, device and storage medium for constructing a career flow graph

Abstract

The present disclosure provides a method, system, device and storage medium for constructing a career flow chart, relates to the field of data processing, and particularly to the field of big data processing. The specific implementation scheme is: after obtaining two target positions to be determined for flow relationship, for each target position, the first relative representation of each target position in its target adjacent position perspective is calculated, and then the independent representation of the target position is obtained, and then based on the preset flow relationship prediction formula and the independent representation of the two target positions, the prediction result of whether there is a career flow relationship between the two target positions is obtained. By constructing a career flow chart based on a large amount of career data and predicting the career flow relationship based on the career flow chart and the preset flow relationship prediction formula, the present disclosure improves the reliability of career flow prediction while realizing the prediction of the overall career flow from a macro perspective without relying on manual experience analysis.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and particularly relates to the technical field of big data processing. BACKGROUND

[0002] At present, big data processing technology is used in various fields to analyze and predict data. The results of data analysis can be applied to network services in the corresponding field. For example, a recruitment website can analyze labor market data, mainly including data processing and career analysis of big data such as positions and companies in user resumes, and based on the results of data processing and data analysis, providing better data support for enterprises and more suitable career recommendations for job seekers, etc. SUMMARY

[0003] The present disclosure provides a method, system, device and storage medium for constructing a career flow chart for confirming whether there is a flow relationship between different positions.

[0004] According to an aspect of the present disclosure, a method for constructing a career flow chart is provided, comprising:

[0005] obtaining two target positions to be determined for flow relationship;

[0006] based on the pre-stored career flow chart, for each target position, obtaining a target adjacent position having a career flow relationship with the target position, and an average tenure length between the target position and the target adjacent position;

[0007] for each target position, based on the first original feature of the target position, the second original feature of the target adjacent position, and the average tenure length between the target position and the target adjacent position, calculating at least one first relative representation of the target position under the perspective of each target adjacent position;

[0008] for each target position, based on each first relative representation of the target position, obtaining an independent representation of the target position;

[0009] based on a preset flow relationship prediction formula and the independent representations of the two target positions, performing flow prediction on the two target positions to obtain a prediction result of whether there is a career flow relationship between the two target positions.

[0010] According to another aspect of the present disclosure, a system for constructing a career flow chart is provided, comprising: an encoder and a decoder;

[0011] The encoder is configured to: obtain two target positions for which a flow relationship is to be determined; based on a pre-stored occupation flow chart, for each target position, obtain a target adjacent position that has an occupation flow relationship with the target position and an average tenure length between the target position and the target adjacent position; for each target position, based on a first original feature of the target position, a second original feature of the target adjacent position, and the average tenure length between the target position and the target adjacent position, calculate at least one first relative representation of the target position from a perspective of each target adjacent position; and for each target position, based on the at least one first relative representation of the target position, obtain an independent representation of the target position.

[0012] The decoder includes an occupation flow prediction module configured to perform flow prediction on the two target positions based on a preset flow relationship prediction formula and the independent representations of the two target positions, and obtain a prediction result of whether an occupation flow relationship exists between the two target positions.

[0013] According to another aspect of the present disclosure, an electronic device is provided, including:

[0014] at least one processor; and

[0015] a memory in communication connection with the at least one processor; wherein

[0016] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of constructing an occupation flow chart described above.

[0017] According to another aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to perform the method of constructing an occupation flow chart described above.

[0018] According to another aspect of the present disclosure, a computer program product is provided, including a computer program which, when executed by a processor, implements the method of constructing an occupation flow chart described above.

[0019] It should be understood that the contents described in this part are not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0021] Figure 1ais a schematic diagram of a first embodiment of a method for constructing a career flow map according to the present disclosure;

[0022] Figure 1b is a schematic diagram of calculating a Gaussian representation of another post in a post perspective according to an embodiment of the present disclosure;

[0023] Figure 2 is a schematic diagram of a second embodiment of a method for constructing a career flow map according to the present disclosure;

[0024] Figure 3 is a schematic diagram of a third embodiment of a method for constructing a career flow map according to the present disclosure;

[0025] Figure 4 is a schematic diagram of a fourth embodiment of a method for constructing a career flow map according to the present disclosure;

[0026] Figure 5 is a schematic diagram of a first embodiment of a system for constructing a career flow map according to the present disclosure;

[0027] Figure 6 is a schematic diagram of a second embodiment of a system for constructing a career flow map according to the present disclosure;

[0028] Figure 7 is a schematic diagram of a third embodiment of a system for constructing a career flow map according to the present disclosure;

[0029] Figure 8 is a block diagram of an electronic device for implementing a method for constructing a career flow map according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to help in understanding, which should be considered in their context only. Thus, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0031] Career flow, i.e., the transfer of personnel between different posts or companies, in the prior art, is usually analyzed by manpower around small sample flow data. However, due to the limitation of observation and analysis of a small amount of flow data, there is a lack of support of big data, so the prior art cannot model the labor market from a macro perspective based on big data, and cannot determine the transfer of personnel between posts or companies.

[0032] In addition, since the prior art completely relies on manual experience analysis, it is impossible to carry out quantitative research, and it is also impossible to predict career mobility.

[0033] To solve the above problems existing in the prior art, the present disclosure provides a method, system, device and storage medium for constructing a career mobility graph. First, the method for constructing a career mobility graph provided by the present disclosure is introduced.

[0034] Referring to Figure 1a , Figure 1a is a schematic diagram of the first embodiment of the method for constructing a career mobility graph provided by the present disclosure, which can include the following steps:

[0035] Step S110, obtaining two target positions for which the flow relationship is to be determined;

[0036] Step S120, based on the pre-stored career mobility graph, for each target position, obtaining a target adjacent position that has a career mobility relationship with the target position, and the average tenure length between the target position and the target adjacent position.

[0037] In an embodiment of the present disclosure, the above-mentioned career mobility graph can be a directed graph, which can be pre-stored by the following steps:

[0038] Step one, obtaining resume data from an online career website.

[0039] In an embodiment of the present disclosure, the above-mentioned resume data can be collected from a large amount of resume data in the online career network (OPNs) website data from an open source public data set.

[0040] As a specific embodiment of the present disclosure, after obtaining the above-mentioned resume data, the career path information can be extracted from each resume data, i.e. extracting each career experience in the resume in chronological order. In this embodiment, for each career experience, the company, position and tenure information in the career experience can be saved, for example, for each career experience, the company name, position name and tenure length information in the career experience can be saved, and the tenure length can be in units of years.

[0041] Step two, according to the content required by the preset keyword table, extract the position keywords that meet the requirements of the keyword table from the original records of the resume data, generate a uniform format of the position name as the node (g) of the career mobility graph.

[0042] In an embodiment of the present disclosure, three types of keyword tables can be set, which are respectively: function words, function words, and grade words. Among them, the function words can refer to projects or products, such as image processing or mobile phone shell design, etc.; the function words can refer to the main work content, such as testing or research and development, etc.; and the grade words can refer to the grade of the position, such as junior engineer or senior engineer, etc.

[0043] In the embodiment, the complex and intricate post name representations can be matched and aligned by extracting these keywords from the original records. Specifically, the keywords can be extracted from the post information in the original records of the resume data. Then, the posts with the same three types of keywords are determined as the same post. For example, if the keywords extracted from the post information in multiple resumes are all: image processing, testing, and senior engineer, they belong to the same post. Then, the unified format of the post name can be generated based on the post information and the corresponding company name. For example, the post name can be: company name + post information (for example, post information keywords).

[0044] Based on the above three types of keyword tables, the corresponding post keywords are extracted for each company and post information saved above, so that the complex and diverse post names can be matched in a unified format, and the convenience of subsequent data processing is improved.

[0045] In the embodiment of the present disclosure, a specific post of a specific company can be taken as a node to construct a career flow graph. For example, the a post of A company can be taken as a node, the b post of A company can be taken as a node, and the a post of B company can be taken as a node.

[0046] Step three: extracting the career flow paths between different posts from the resume data as directed edges (e) of the career flow graph, and taking the average tenure length of one post to another post as the weight of the directed edge.

[0047] As described above, in the embodiment of the present disclosure, the above career flow graph can be a directed graph, in which the edge between two nodes represents the career flow relationship between the corresponding two posts, and the direction of the edge represents the direction of the career flow. For example, the edge from node A to node B represents that there is a flow relationship from post A to post B between post A and post B, and the weight of the directed edge can be the average tenure length of each employee in post A when flowing from post A to post B.

[0048] In the embodiments of the present disclosure, after the target post is acquired, the node matched with the target post can be acquired based on the pre-stored career flow chart. As an embodiment, keyword extraction can be performed on the target node based on the keyword table, and the node matched with the target post can be acquired based on the company of the target post, the keywords, and the keywords of the nodes in the career flow chart. The adjacent node of the matched node is the adjacent post of the target post.

[0049] As shown in Figure 1a Step S130, for each target post, at least one first relative representation of the target post in the perspective of each target adjacent post is calculated based on the first original feature of the target post, the second original feature of the target adjacent post, and the average tenure length between the target post and the target adjacent post.

[0050] In an embodiment of the present disclosure, the coding rule can be pre-set for the keywords in the classification keyword table. After keyword extraction is performed on the target post, the extracted target post keywords can be coded according to the coding rule, so as to obtain the one-hot code of the target post as the first original feature of the target post.

[0051] The second original feature of the target adjacent post can be the one-hot code of the target adjacent post acquired from the career flow chart.

[0052] Step S140, for each target post, the independent representation of the target post is obtained based on each first relative representation of the target post.

[0053] In the embodiments of the present disclosure, the first relative representations of the target post can be fused to obtain the independent representation of the target post.

[0054] Step S150, based on the preset flow relationship prediction formula and the independent representations of the two target posts, the flow of the two target posts is predicted to obtain the prediction result of whether there is a career flow relationship between the two target posts.

[0055] It can be seen that the method for constructing a career flow chart provided in the embodiments of the present disclosure, after obtaining two target positions to be determined for a flow relationship, for each target position, based on the pre-stored career flow chart, a target adjacent position having a career flow relationship with the target position is obtained, and the average tenure length between the target position and the target adjacent position is obtained, and based on the first original feature of the target position, the second original feature of the target adjacent position, and the average tenure length between the target position and the target adjacent position, at least one first relative representation of each target position in the perspective of the target adjacent position thereof is calculated, then for each target position, based on the first relative representation of the target position, an independent representation of the target position is obtained, and then based on the preset flow relationship prediction formula and the independent representations of the two target positions, a prediction result of whether there is a career flow relationship between the two target positions is obtained. By constructing a career flow chart based on a large amount of career data and predicting a career flow relationship based on the career flow chart and a preset flow relationship prediction formula, the embodiments of the present disclosure can realize prediction of the overall career flow from a macro perspective, and improve the reliability of career flow prediction without relying on manual experience analysis.

[0056] Further, the method provided in the embodiments of the present disclosure can better support downstream predictive tasks, such as career recommendation, career path planning, and the like.

[0057] In an embodiment of the present disclosure, when obtaining the first relative representation, the following steps can be used to obtain:

[0058] Step 1, for each target position, based on the first original feature of the target position, the second original feature of the target adjacent position, and the average tenure length of the target position to each target adjacent position, the first implicit representation of each target position and the second implicit representation of each target adjacent position are extracted.

[0059] In the embodiments of the present disclosure, GCN network can be used to extract implicit representations of each target position and target adjacent position of each target position.

[0060] GCN (Graph Convolutional networks) can process graph structure data, and can aggregate information of adjacent nodes of each node in the graph to extract features of the node.

[0061] In the embodiments of the present disclosure, the above career flow chart can be input into the GCN network, and the GCN network can aggregate information of each target adjacent position node based on the weight of the target position node corresponding to each target position, the target adjacent position node corresponding to the target adjacent position, and the edge between the target position node and the target adjacent position node, to obtain the implicit representation of the target position node, that is, the first implicit representation of the target position.

[0062] The process of calculating the second implicit representation of each target adjacent post can refer to the process of calculating the first implicit representation of the target post, which will not be described here.

[0063] Step 2, based on the first implicit representation of each target post and the second implicit representation of each target adjacent post, at least one first relative representation of the target post under the perspective of each target adjacent post is calculated.

[0064] It can be seen that in the embodiments of the present disclosure, when calculating the first relative representation of each target post, the GCN network is used to learn the representation of the adjacent post, so that more information can be contained in the first representation.

[0065] In the embodiments of the present disclosure, each representation described above can be in the form of a vector.

[0066] In the embodiments of the present disclosure, the first relative representation of the target post can be a Gaussian representation. Specifically, the first relative representation of the target post can be calculated by the following steps:

[0067] Step 1, for each target post, the first implicit representation thereof is spliced with the second implicit representation of each target adjacent post respectively, to obtain a first spliced representation which is the same as the number of target adjacent posts.

[0068] In the embodiments of the present disclosure, when calculating the first relative representation of the target post under the perspective of each target adjacent post, the first implicit representation of the target post can be sequentially spliced with the second implicit representation of each target adjacent post to obtain the first spliced representation.

[0069] For example, if the first implicit representation of the target post A is h a , the second implicit representation of a target adjacent post B thereof is h b (h a , h b are vectors), when calculating the first relative representation of the target post A under the perspective of the target adjacent post B, h a , h b can be sequentially spliced as the first spliced representation, that is, h a h b is the first spliced representation.

[0070] Step 2, each first spliced representation is multiplied with a preset Gaussian distribution weight matrix respectively, to obtain a Gaussian representation which is the same as the number of target adjacent posts, as the first relative representation.

[0071] Specifically, the aforementioned first relative representation can be the mean and variance of the Gaussian distribution of the aforementioned first splicing representation, which is the head node representation from the perspective of the tail node during splicing.

[0072] Accordingly, in this embodiment of the disclosure, the Gaussian distribution weight matrix may include: a mean weight matrix and a variance weight matrix.

[0073] In this embodiment of the disclosure, both the mean weight matrix and the variance weight matrix are learnable weight matrices.

[0074] Therefore, step ② above can specifically be:

[0075] Each first spliced ​​representation is multiplied by the mean weight matrix and the variance weight matrix respectively to obtain the same number of Gaussian representations as the target adjacent positions, which are used as the first relative representations.

[0076] like Figure 1b As shown, based on the example in step ① above, when calculating the first relative representation of target position A from the perspective of target adjacent position B, the first splicing matrix h can be used as a reference. a h b , respectively with the mean weight matrix W μ And variance weight matrix W σ Multiplying these values ​​yields the mean μa|b and variance σa|b of the Gaussian distribution of the first concatenated matrix. The mean and variance are then concatenated to obtain the first relative representation Z of target position A from the perspective of adjacent position B. a|b .

[0077] In one embodiment of this disclosure, steps ① to ② above can be implemented by a pre-trained Gaussian representation acquisition module, which may include the learnable mean weight matrix and variance weight matrix.

[0078] As described above, the first relative representation of the target position is a Gaussian representation of the first concatenated representation. The Gaussian distribution can effectively describe the uncertainty of the representation. In other words, in this embodiment, when learning the representation for each target position, the interference caused by various uncertainties during career mobility is considered, making subsequent career mobility predictions more accurate. Furthermore, using the more refined modeling method of the Gaussian distribution to model the aforementioned Gaussian representation acquisition module can capture the uncertainties in career mobility, making the model prediction more accurate and robust.

[0079] Accordingly, step S140 above can be further refined as follows:

[0080] For each target position, the same number of Gaussian representations as the adjacent positions are fused together to obtain an independent representation for that target position.

[0081] As described above, in the embodiments of the present disclosure, the first relative representation can be in the form of a vector, and therefore, when fusing the Gaussian representations (first relative representations) of the target post, the numbers at the same positions in the Gaussian representation vectors can be averaged or weighted averaged to obtain the independent representation of the target post. When using the weighted average method to fuse the Gaussian representation vectors, the weight of each Gaussian representation vector can be the weight of the edge between the target post node and the corresponding target adjacent post node.

[0082] By fusing the Gaussian representations of the target post to obtain the independent representation of the target post, the target post can learn the representations of its adjacent posts, reduce the influence of uncertainty in career mobility on the independent representation of the target post, and thus the independent representation can more accurately describe the target post.

[0083] In an embodiment of the present disclosure, as shown in Figure 2 the step S150 shown in Figure 1a may be refined as follows:

[0084] Step S151, converting the independent representations of the two target posts into vector representations.

[0085] For example, the independent representation of each target post can be a 64-dimensional vector.

[0086] Step S152, for each target post vector, extracting a scalar from it as the first quality parameter of the target post, and the remaining part of the vector as the first position parameter of the target post, to obtain two first quality parameters and two first position parameters.

[0087] In the embodiments of the present disclosure, a scalar can be arbitrarily selected from the independent representation vector of the target post as the first quality parameter, and the remaining part as the first position parameter.

[0088] Based on the example in step S151, the last data in the 64-dimensional vector can be selected as the first quality parameter of the target post, and the remaining 63-dimensional vector as the first position parameter.

[0089] Step S153, inputting the two first quality parameters and the two first position parameters into the flow relationship prediction formula based on the idea of universal gravitation, to predict the flow of the two target posts, and obtain the prediction result of whether there is a career flow relationship between the two target posts.

[0090] In the embodiments of the present disclosure, when predicting whether there is a career flow relationship between two target posts, it can be specifically predicted whether there is a career flow relationship between the two target posts and the direction of the career flow, so that the prediction of the career flow is more comprehensive.

[0091] In the embodiments of the present disclosure, the career flow between posts can be analogized to the universal gravitation in physics, that is, the mutual attraction between objects is used to simulate the mutual attraction between posts.

[0092] Specifically, in the embodiments of the present disclosure, the flow relationship prediction formula based on the idea of universal gravitation (gravity) can be:

[0093]

[0094] In the formula, i and j are the numbers of the two target posts, e ij is the flow relationship from the target post i to the target post j (that is, the edge between the two target post nodes), p(e ij |vi,vj) can represent the probability of the existence of the flow relationship from the target post i to the target post j, is the first mass parameter of the target post j, λ g is a constant with a value range of 0-1, in the embodiments of the present disclosure, λ g The specific value of λ respectively represent the first position parameters of the target post i and the target post j, represents the square of the distance between the two first position parameters, and σ represents the sigmold function, which can limit the value range of the input data to 0-1.

[0095] In the embodiments of the present disclosure, the probability of the existence of the flow relationship between the two posts can be obtained through the above formula.

[0096] As can be seen from the above formula, in a single calculation, only the probability of the existence of the flow relationship from the target post i to the target post j can be calculated, and if the probability of the existence of the flow relationship from the target post j to the target post i is to be calculated, i and j in the formula can be exchanged.

[0097] In the embodiments of the present disclosure, a probability threshold can be set, if the calculated probability of the existence of the flow relationship from the target post i to the target post j is greater than the probability threshold, it can be determined that there is a flow relationship from the target post i to the target post j, otherwise, there is no flow relationship.

[0098] In the embodiments of the present disclosure, by analogizing the career flow behavior to some common physical phenomena, the method for constructing a career flow graph provided by the present disclosure has better interpretability.

[0099] In an embodiment of the present disclosure, the method for constructing the career flow graph can further include: Figure 1a as shown in Figure 3 , the method for constructing the career flow graph can further include:

[0100] In step S360, if the two target posts have a career flow relationship, the two target posts are adjacent posts, and the second relative representation of the two target posts in the perspective of the other target post is calculated based on the first original feature of the two target posts.

[0101] The two target posts have a career flow relationship, that is, there is an edge between the two target post nodes, so the two target posts are adjacent posts.

[0102] In an embodiment of the present disclosure, the second relative representation can be calculated by the following steps:

[0103] The second relative representation of the two target posts in the perspective of the other target post is calculated based on the first implicit representation of the two target posts.

[0104] The execution process of this step can refer to the process of calculating the first relative representation described above. Here, no longer be described.

[0105] Specifically, in calculating the second relative representation of the two target posts in the perspective of the other target post, the following steps can be used to calculate:

[0106] Step 1: The first implicit representation of the two target posts is spliced in the first way and the second way respectively to obtain two second spliced representations;

[0107] Step 2: Each second spliced representation is multiplied by a preset Gaussian distribution weight matrix to obtain two Gaussian representations as the second relative representation respectively.

[0108] Similar to the first relative representation, the second relative representation is the Gaussian representation of the current target post in the perspective of the other target post. Therefore, in an embodiment of the present disclosure, when the second relative representation of the two target posts is obtained, the first implicit of the two target posts can be spliced from two directions, so that the Gaussian representation of the current post in the perspective of the other target post can be obtained for the two target posts respectively.

[0109] For example, as shown in Figure 1b , if the first implicit representation of the two target posts is h a and h b , then the two second spliced representations obtained can be h a h b and h b h a .

[0110] As described above, in the embodiments of the present disclosure, the Gaussian distribution weight matrix can include a mean weight matrix and a variance weight matrix. Therefore, in the embodiments of the present disclosure, when calculating the second relative representation, it can be:

[0111] Each second spliced representation is multiplied by the mean weight matrix and the variance weight matrix respectively to obtain two Gaussian representations as the second relative representation.

[0112] Based on the example in step 2, see Figure 1b , the second spliced representation h a h b and h b h a are multiplied by the mean weight matrix W μ and the variance weight matrix W σ respectively, and the obtained two Gaussian representations can be Z a|b and Z b|a .

[0113] The specific execution process of this step can refer to the specific process of calculating the first relative representation described above, which will not be described here.

[0114] In step S370, based on the preset flow tenure prediction formula and the second relative representation of the two target posts in the perspective of another target post, tenure prediction is performed on the two target posts to obtain a prediction result of the flow tenure between the two target posts.

[0115] As can be seen, in the embodiments of the present disclosure, not only can the existence of career flow and the direction of career flow between posts be predicted, but also the tenure length of the possible career flow can be further predicted, further improving the comprehensiveness of career flow prediction.

[0116] In an embodiment of the present disclosure, as shown in Figure 4 , the step S370 shown in Figure 3 may be refined as follows:

[0117] In step S371, the second relative representation of the two target posts in the perspective of another target post is converted into a vector representation respectively.

[0118] In step S372, for the vector of each second relative representation, a scalar is extracted therefrom as a second quality parameter of the target post, and the remaining part of the vector is taken as a second position parameter of the target post, to obtain two second quality parameters and two second position parameters.

[0119] In step S373, the two second mass parameters and the second position parameter are input into a flow tenure prediction formula preset based on the object work idea, tenure prediction is performed on the two target posts, and a prediction result of the flow tenure between the two target posts is obtained.

[0120] In the embodiments of the present disclosure, the time length required for the flow between the target posts can be analogized to the work required to move the object at the current target post node to another target post node position.

[0121] Specifically, the flow tenure prediction formula preset based on the object work idea can be as follows:

[0122]

[0123] In the formula, i and j are the numbers of the two target posts, W ij may represent the tenure time length of the career flow from the target post i to the target post j, is the second mass parameter of the target post i, and λ e is a constant with a value range of 0-1, in the embodiments of the present disclosure, λ e may be artificially selected; and respectively represent the second position parameters of the target post i and the target post j, represents the square of the distance between the two second position parameters.

[0124] In the embodiments of the present disclosure, the unit of the flow tenure time length obtained based on the above flow tenure prediction formula can be years.

[0125] In the embodiments of the present disclosure, the flow tenure prediction formula is determined based on the object work idea, so that the formula has good interpretability.

[0126] According to the embodiments of the present disclosure, the present disclosure further provides a system for constructing a career flow graph, as shown in FIG. 5, the system can include an encoder 510 and a decoder 520. Figure 5

[0127] The encoder 510 can be used to obtain two target posts to be determined for a flow relationship; based on a pre-stored career flow graph, for each target post, a target adjacent post having a career flow relationship with the target post and an average tenure time length between the target post and the target adjacent post are obtained; for each target post, at least one first relative representation of the target post under each target adjacent post perspective is calculated based on a first original feature of the target post, a second original feature of the target adjacent post, and the average tenure time length between the target post and the target adjacent post; and for each target post, an independent representation of the target post is obtained based on each first relative representation of the target post. ​

[0128] The decoder 520 can include a career flow prediction module 521.

[0129] The career flow prediction module 521 can be configured to perform flow prediction on the two target posts based on a preset flow relationship prediction formula and the independent representations of the two target posts, and obtain a prediction result of whether there is a career flow relationship between the two target posts.

[0130] In the embodiments of the present disclosure, the encoder 510 can be an encoder based on uncertainty representation, the decoder 520 can be a decoder based on physical heuristic, and the career flow prediction module 521 can be a decoder based on gravity heuristic.

[0131] It can be seen that the system for constructing a career flow graph provided in the embodiments of the present disclosure, after the encoder obtains two target posts for which a flow relationship needs to be determined, for each target post, the encoder obtains a target adjacent post having a career flow relationship with the target post and an average tenure length between the target post and the target adjacent post based on a pre-stored career flow graph, and calculates at least one first relative representation of each target post from the perspective of the target adjacent post of the target post based on a first original feature of the target post, a second original feature of the target adjacent post, and the average tenure length between the target post and the target adjacent post. Then, for each target post, the decoder obtains an independent representation of the target post based on each first relative representation of the target post. The career flow prediction module in the decoder obtains a prediction result of whether there is a career flow relationship between the two target posts based on a preset flow relationship prediction formula and the independent representations of the two target posts. By constructing a career flow graph based on a large amount of career data and predicting career flow relationships based on the career flow graph and the preset flow relationship prediction formula, the embodiments of the present disclosure do not rely on manual experience analysis, and can achieve prediction of global career flow from a macro perspective while improving the reliability of career flow prediction.

[0132] In an embodiment of the present disclosure, the career flow prediction module 521 can be configured to convert the independent representations of the two target posts into vector representations, extract a scalar from the vector representation of each target post as a first quality parameter of the target post, and extract the remaining part of the vector representation as a first position parameter of the target post, to obtain two first quality parameters and two first position parameters. The two first quality parameters and the two first position parameters are input into a flow relationship prediction formula based on the idea of universal gravitation to perform flow prediction on the two target posts, and obtain a prediction result of whether there is a career flow relationship between the two target posts.

[0133] In the embodiments of the present disclosure, the career flow prediction module can include the following flow relationship prediction formula based on the idea of universal gravitation:

[0134]

[0135] In the formula, i and j are the numbers of the two target positions, e ij is the flow relationship between target position i and target position j (i.e., the edge between the two target position nodes), p(e ij |vi,vj) can represent the probability of the existence of a flow relationship between target position i and target position j, is the first quality parameter of target position j, λ g is a constant with a value ranging from 0 to 1, and in the embodiments of the present disclosure, λ g The specific value of λ respectively represent the first position parameters of target position i and target position j, represents the square of the distance between the two first position parameters, and σ represents the sigmold function, which can limit the value range of the input data to 0-1.

[0136] In an embodiment of the present disclosure, on the basis of Figure 5 as shown in Figure 6 the decoder 520 can further include a flow tenure prediction module 522;

[0137] The flow tenure prediction module 522 can be configured to, if there is a career flow relationship between the two target positions, the two target positions are adjacent positions, calculate second relative representations of the two target positions from the perspective of another target position based on the first original features of the two target positions; and perform tenure prediction on the two target positions based on a preset flow tenure prediction formula and the second relative representations of the two target positions from the perspective of another target position, to obtain a prediction result of the flow tenure between the two target positions.

[0138] In an embodiment of the present disclosure, the flow tenure prediction module 522, based on a preset flow tenure prediction formula and second relative representations of the two target positions from the perspective of another target position, can perform tenure prediction on the two target positions to obtain a prediction result of the flow tenure between the two target positions, can include:

[0139] The independent representations of the two target positions are converted into vector representations. For each target position's vector, a scalar is extracted as the first mass parameter of that target position, and the remaining part of the vector is used as the first position parameter of that target position, resulting in two first mass parameters and two first position parameters. The two first mass parameters and two first position parameters are then input into a flow relationship prediction formula based on the concept of universal gravitation to predict the flow between the two target positions and obtain the prediction result of whether there is a job mobility relationship between the two target positions.

[0140] In this embodiment of the disclosure, the above-mentioned flow tenure prediction module may include the following flow tenure prediction formula based on the idea of ​​work done by an object:

[0141]

[0142] In this formula, i and j are the numbers of the two target positions, respectively, and W ij This can represent the length of tenure for a career move from target position i to target position j. Let λ be the second quality parameter for target position i. e As a constant with a value range of 0 to 1, in this embodiment of the disclosure, λ e The specific value can be selected manually; Then, these represent the second positional parameters for target position i and target position j, respectively. This represents the square of the distance between the two second positional parameters.

[0143] In this embodiment of the disclosure, the aforementioned flow term prediction module can be a decoder based on work-inspired principles.

[0144] In one embodiment of this disclosure, the occupational mobility graph is a directed graph, the nodes of the occupational mobility graph are the names of various positions, the directed edges of the occupational mobility graph are the mobility relationships from one position to another, and the weight of the edge is the average term of office.

[0145] like Figure 6 As shown, the encoder 510 may include: a pre-stored occupational flow map 511, a hidden layer representation acquisition module 512, and a Gaussian representation acquisition module 513;

[0146] The hidden layer representation acquisition module 512 can be used to extract the first implicit representation of each target position and the second implicit representation of each target adjacent position based on the first original feature of the target position, the second original feature of the target adjacent positions, and the average tenure of the target position to each target adjacent position, using a pre-trained graph convolutional network.

[0147] In this embodiment of the disclosure, the hidden layer representation acquisition module 512 can acquire the target adjacent positions of the target position and the average tenure of the target position to each target adjacent position based on the pre-stored occupational mobility map 511.

[0148] The Gaussian representation acquisition module 513 can be used to calculate at least one first relative representation of the target position from the perspective of each adjacent target position, based on the first implicit representation of each target position and the second implicit representation of each target adjacent position; and / or,

[0149] Based on the first implicit representation of the two target positions, the second relative representation of the two target positions from the perspective of the other target position is calculated respectively.

[0150] In one embodiment of this disclosure, the Gaussian representation acquisition module 513 calculates at least one first relative representation of the target position from the perspective of each target adjacent position, based on the first implicit representation of each target position and the second implicit representation of each target adjacent position, including:

[0151] For each target position, its first implicit representation is concatenated with the second implicit representations of each adjacent target position to obtain the same number of first concatenated representations as the number of adjacent target positions; each first concatenated representation is then multiplied by a pre-trained Gaussian weight matrix to obtain the same number of Gaussian representations as the number of adjacent target positions, which are used as the first relative representations; and / or,

[0152] Based on the first implicit representation of the two target positions, the second relative representation of the two target positions from the perspective of the other target position is calculated, including: concatenating the first implicit representation of the two target positions in the first and second ways respectively to obtain two second concatenated representations; multiplying each second concatenated representation with a preset Gaussian distribution weight matrix to obtain two Gaussian representations, which are used as the second relative representations respectively.

[0153] See Figure 7 , Figure 7 This is a schematic diagram of a specific example of a system for constructing occupational mobility maps in this disclosure:

[0154] like Figure 7 As shown, this system can utilize an encoder based on uncertainty representation and a decoder based on physics. The figure illustrates the process of using this system to predict career mobility for target positions a and b, specifically including the following steps:

[0155] Step ①: Using the GCN network (the implicit representation acquisition module in this disclosure), neighbor information of the target job node is aggregated based on the occupational mobility graph.

[0156] likeFigure 7 As shown, the target adjacent post nodes of the target post node a are post nodes c, d, e, and the GCN network is used to perform feature extraction on the target post node a and the target adjacent post nodes c, d, and e, so that the first implicit representation h a of the target post a and the second implicit representations h c , h d , h e of the target adjacent post nodes c, d, and e can be obtained. b Similarly, the first implicit representation h e of the target post b and the second implicit representations h f , h g of the target adjacent post nodes e, f, and g can be obtained. b e f g .

[0157] Step 2: Using the Gaussian representation module (i.e., the Gaussian representation acquisition module described above), based on the first implicit representation of each target post and the second implicit representation of its target adjacent post, the implicit representation of each target post is calculated.

[0158] In this embodiment, the Gaussian representation of each target post under the perspective of each target adjacent post can be obtained based on the first implicit representation of each target post and the second implicit representation of its target adjacent post, and then the Gaussian representations are fused to obtain the independent representation of the target post.

[0159] As shown in the left part of the Gaussian representation module, taking posts a and b as examples, when calculating the representation of post a under the perspective of post b, the implicit representation vectors of the two posts can be spliced to obtain the spliced representation h a h b , and then h a h b is multiplied by the mean weight matrix W μ and the variance weight matrix W σ , respectively, to obtain the mean μ a|b and the variance σ a|b of the Gaussian distribution of the spliced representation, and μ a|b and σ a|b are spliced to obtain the Gaussian representation Z a|b of post a under the perspective of post b.

[0160] The above steps are used for calculating the target post a and its target adjacent posts c, d, and e, and the Gaussian representations μ a|c σ a|c , μ a|d σ a|d , μ a|e σ a|e of the target post a under the perspective of the target adjacent posts c, d, and e are obtained, respectively, and the Gaussian representations are fused to obtain the independent representation Z of the target post a.a (μ a σ a ). Similarly, the independent representation Z b of the target post b can be obtained.

[0161] Step 3, the independent representation of the target post a and b is respectively split into the first quality parameter and the first position parameter The gravity-inspired decoder (i.e., the career flow prediction module described above) is used to calculate the first quality parameter and the first position parameter based on the career flow prediction formula, and the sigmold function is used to obtain the career flow prediction result p(e ab |va,vb) of the target post a to the target post b.

[0162] The result shows that the probability of the career flow relationship between the target post a and the target post b exists. If the probability of the career flow relationship between the target post b and the target post a is to be obtained, the order of a and b in the career flow prediction formula can be exchanged, i.e., the post b is taken as i in the career flow prediction formula, and the post a is taken as j in the career flow prediction formula.

[0163] Step 4, if the result in step 3 shows that there is a career flow relationship between the target post a and the target post b, the first implicit representation of the target post a and b is spliced from two directions by using the Gaussian representation module to obtain two spliced representations, and the Gaussian representation of the target post a under the target post b Z a|b and the Gaussian representation of the target post b under the target post a Z b|a are obtained based on the two spliced representations (as shown in the left part of the Gaussian representation module). Figure 7

[0164] The specific calculation process has been described in step 2, which will not be repeated here.

[0165] Step 5, Z a|b and Z b|a are respectively split into the second quality parameter and the second position parameter The work-inspired encoder (the flow tenure prediction module in the present disclosure) is used to calculate the two second quality parameters and the second position parameters based on the flow tenure prediction formula to obtain the flow tenure prediction result W ab for the target post a and b.

[0166] In the present embodiment, the learning and prediction tasks of the two modules can be optimized synchronously in an end-to-end manner during the training process of the whole system.

[0167] ​In an embodiment of the present disclosure, the encoder can be trained based on the pre-stored career flow graph. For example, the career flow graph can be input into the encoder to be trained, the Gaussian representation of the current node from the perspective of each adjacent node and the independent representation of each node output by the encoder to be trained are obtained, and the independent representation of each node is input into a preset career flow prediction module (including a career flow prediction formula), and the Gaussian representation of the current node from the perspective of each adjacent node is input into a preset flow tenure prediction module (including the flow tenure prediction formula). The probability of the existence of a career flow relationship between two nodes and the flow tenure length output by the career flow prediction module and the flow tenure prediction module are obtained respectively, and the encoder is trained according to the results until the output results of the preset career flow prediction module and the flow tenure prediction module are consistent with the relationship between the nodes in the career flow graph, or the error is less than a preset error threshold.

[0168] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the technical solution comply with relevant laws and regulations and do not violate public order and good customs.

[0169] According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0170] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0171] As shown in Figure 8 The device 800 includes a computing unit 801 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0172] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices over a computer network, such as the Internet, and / or various telecommunication networks.

[0173] The computing unit 801 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the method of constructing a career flow graph. For example, in some embodiments, the method of constructing a career flow graph can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded onto the RAM 803 and executed by the computing unit 801, one or more steps of the method of constructing a career flow graph described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the method of constructing a career flow graph by any other appropriate means, such as by means of firmware.

[0174] The various implementations of the systems and techniques described above herein can be realized in a digital electronic circuit system, an integrated circuit system, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an application specific standard product (ASSP), a system on a chip system (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0175] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package, or entirely on a remote machine or server.

[0176] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0177] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0178] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0179] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions between them occurring over a communication network. The relationship between client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, servers of a distributed system, or servers incorporating blockchain.

[0180] It should be understood that the steps shown in the various forms above can be reordered, added to, or removed. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, without limitation herein, so long as the desired results of the technology disclosed in the present disclosure are achieved.

[0181] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and principles of the disclosure. Any further modifications, equivalents or alternatives within the spirit and principles of the disclosure are to be considered as falling within the scope of the disclosure.

Claims

1. A method for constructing a career flow graph, applied to an electronic device, comprising: obtaining two target positions for which a flow relationship is to be determined; based on a pre-stored career flow graph, obtaining, for each target position, target neighboring positions that have a career flow relationship with the target position, and an average tenure length between the target position and the target neighboring positions; for each target position, based on a first original feature of the target position, a second original feature of the target neighboring positions, and the average tenure length from the target position to each target neighboring position, extracting a first implicit representation of each target position and a second implicit representation of each target neighboring position, and based on the first implicit representation of each target position and the second implicit representation of each target neighboring position, calculating at least one first relative representation of the target position from the perspective of each target neighboring position, wherein the first original feature comprises a one-hot code obtained by encoding keywords of the target position, and the second original feature comprises a one-hot code obtained by encoding keywords of the target neighboring positions; for each target position, based on each first relative representation of the target position, obtaining an independent representation of the target position, wherein the independent representation is obtained by fusing each first relative representation of the target position; converting the independent representations of the two target positions into vector representations; for each target position, extracting a scalar from the vector representation of the target position as a first quality parameter of the target position, and the remaining part of the vector representation as a first position parameter of the target position, to obtain two first quality parameters and two first position parameters; inputting the two first quality parameters and the two first position parameters into a pre-set flow relationship prediction formula based on the idea of universal gravitation to predict the flow of the two target positions, and obtaining a prediction result of whether there is a career flow relationship between the two target positions, wherein the prediction result is used to provide career recommendations or career path planning for job seekers; wherein the pre-set flow relationship prediction formula is used to calculate a probability of a career flow relationship between a target position i and a target position j, wherein the probability is positively correlated with the first quality parameter of the target position j, and the probability is negatively correlated with the logarithm of the distance between the first position parameters of the target position i and the target position j; the career flow graph is a directed graph, and the following steps are used to pre-store: obtaining resume data from an online career website; extracting job keywords that meet the requirements of a pre-set keyword table from the original records of the resume data to generate job names in a unified format as nodes of the career flow graph, wherein the job keywords include three types: function words, function words, and level words, the function words represent projects or products, the function words represent work content, and the level words represent the level of the position, and each node of the career flow graph represents a job, and the three types of job keywords of the same job are consistent; extracting career flow paths between different jobs from the resume data as directed edges of the career flow graph, and taking the average tenure length from one job to another job as the weight of the directed edge. 2.The method of claim 1, wherein The flow relationship prediction formula is: where i, j are the numbers of the two target positions, e ij is the flow relationship from target position i to target position j, p(e ij |vi,vj) is the probability of the existence of a career flow relationship from target position i to target position j, is the first quality parameter of target position j, λ g is a constant with a value ranging from 0 to 1; respectively represent the first position parameter of target position i and target position j, and σ represents the sigmold function.

3. The method of claim 1, further comprising: if there is a career flow relationship between the two target posts, the two target posts are adjacent posts, and based on the first original features of the two target posts, second relative representations of the two target posts from the perspective of the other target post are calculated respectively; based on the preset flow tenure prediction formula and the second relative representations of the two target posts from the perspective of the other target post, tenure prediction is performed on the two target posts to obtain a prediction result of the flow tenure between the two target posts.

4. The method of claim 3, wherein, the step of performing tenure prediction on the two target posts based on the preset flow tenure prediction formula and the second relative representations of the two target posts from the perspective of the other target post to obtain a prediction result of the flow tenure between the two target posts comprises: converting the second relative representations of the two target posts from the perspective of the other target post into vector representations respectively; for the vector of each second relative representation, a scalar is extracted from the vector as a second quality parameter of the target post, and the remaining part of the vector is taken as a second position parameter of the target post, to obtain two second quality parameters and two second position parameters; inputting the two second quality parameters and second position parameters into the flow tenure prediction formula preset based on the object work idea to perform tenure prediction on the two target posts to obtain a prediction result of the flow tenure between the two target posts.

5. The method of claim 4, wherein, the flow tenure prediction formula is: where i, j are the indices of two target positions, W ij denotes the tenure length of the career flow from target position i to target position j, is the second quality parameter of target position i, λ e is a constant with a value range of 0-1, denote the second location parameters of target position i and target position j, respectively.

6. The method of claim 3, wherein, the second relative representation is obtained by the following steps: based on the first implicit representation of the two target posts, second relative representations of the two target posts from the perspective of the other target post are calculated respectively.

7. The method of claim 6, wherein, the step of calculating at least one first relative representation of the target post from the perspective of each target adjacent post based on the first implicit representation of each target post and the second implicit representation of each target adjacent post comprises: for each target post, its first implicit representation is spliced with the second implicit representation of each target adjacent post to obtain a same number of first spliced representations as the number of target adjacent posts; each first spliced representation is multiplied by a preset Gaussian distribution weight matrix to obtain a same number of Gaussian representations as the number of target adjacent posts, which are taken as first relative representations respectively; and / or the step of calculating second relative representations of the two target posts from the perspective of the other target post based on the first implicit representation of the two target posts comprises: the first implicit representations of the two target posts are spliced in the first way and the second way to obtain two second spliced representations; each second spliced representation is multiplied by a preset Gaussian distribution weight matrix to obtain two Gaussian representations, which are taken as second relative representations respectively.

8. The method of claim 7, wherein, the Gaussian distribution weight matrix comprises a mean weight matrix and a variance weight matrix; The step of respectively multiplying each first spliced representation with a preset Gaussian distribution weight matrix to obtain two Gaussian representations as second relative representations, comprises: The step of respectively multiplying each first spliced representation with the mean weight matrix and the variance weight matrix to obtain two Gaussian representations as first relative representations, comprises: The step of respectively multiplying each second spliced representation with a preset Gaussian distribution weight matrix to obtain two Gaussian representations as second relative representations, comprises: The step of respectively multiplying each second spliced representation with the mean weight matrix and the variance weight matrix to obtain two Gaussian representations as first relative representations, comprises:

9. The method of claim 7, wherein, The step of obtaining, for each target post, an independent representation of the target post based on the respective first relative representations of the target post, comprises: The step of obtaining, for each target post, an independent representation of the target post based on the respective first relative representations of the target post, comprises:

10. A system for constructing a career flow graph, applied to an electronic device, comprising: An encoder and a decoder; The encoder is configured to obtain two target posts for which a career flow relationship is to be determined; Based on a pre-stored career flow graph, for each target post, obtain target neighboring posts that have a career flow relationship with the target post, and an average tenure length between the target post and the target neighboring posts; For each target post, based on the first original features of the target post, the second original features of the target neighboring posts, and the average tenure lengths from the target post to the target neighboring posts, extract a first implicit representation of the target post and a second implicit representation of each target neighboring post, and based on the first implicit representation of the target post and the second implicit representation of each target neighboring post, calculate at least one first relative representation of the target post from the perspective of each target neighboring post; For each target post, based on the respective first relative representations of the target post, obtain an independent representation of the target post, the independent representation being obtained by fusing the respective first relative representations of the target post, the first original features comprising one-hot codes obtained by encoding keywords of the target post, and the second original features comprising one-hot codes obtained by encoding keywords of the target neighboring posts; The decoder comprises a career flow prediction module, the career flow prediction module being configured to convert the independent representations of the two target posts into vector representations, for each target post, extract a scalar from the vector representation of the target post as a first quality parameter of the target post, and the remaining part of the vector representation as a first position parameter of the target post, to obtain two first quality parameters and two first position parameters, and input the two first quality parameters and the two first position parameters into a flow relationship prediction formula preset based on the idea of universal gravitation to perform flow prediction on the two target posts, and obtain a prediction result of whether a career flow relationship exists between the two target posts, the prediction result being used to provide career recommendation or career path planning for job seekers. The preset flow relationship prediction formula is used to calculate a probability that the target post i and the target post j have a career flow relationship, the probability is positively correlated with a first quality parameter of the target post j, and the probability is negatively correlated with a logarithm of a distance between the first position parameters of the target post i and the target post j. The career flow graph is a directed graph, and the system further includes a construction module configured to: obtain resume data from an online career website; extract post keywords meeting the requirements of a preset keyword table from original records of the resume data according to content required by the preset keyword table, generate post names in a unified format as nodes of the career flow graph, wherein the post keywords include three types: function words, function words and grade words, the function words represent projects or products, the function words represent work contents, and the grade words represent the grade of the position, and each node of the career flow graph represents a post, and the three types of post keywords of the same post are consistent; extract career flow paths between different posts from the resume data as directed edges of the career flow graph, and take an average tenure length of one post to another post as a weight of the directed edge.

11. The system of claim 10, wherein the decoder further includes a flow tenure prediction module; the flow tenure prediction module is configured to, if the two target posts have a career flow relationship, the two target posts are adjacent posts, calculate second relative representations of the two target posts in a perspective of another target post based on first original features of the two target posts, and perform tenure prediction on the two target posts based on a preset flow tenure prediction formula and the second relative representations of the two target posts in the perspective of the another target post to obtain a prediction result of a flow tenure between the two target posts.

12. The system of claim 11, wherein the flow tenure prediction module performs tenure prediction on the two target posts based on a preset flow tenure prediction formula and the second relative representations of the two target posts in the perspective of the another target post to obtain a prediction result of a flow tenure between the two target posts, including: converting independent representations of the two target posts into vector representations, extracting a scalar from the vector representation of each target post as a first quality parameter of the target post, and taking the remaining part of the vector representation as a first position parameter of the target post to obtain two first quality parameters and two first position parameters; and inputting the two first quality parameters and the two first position parameters into a preset flow relationship prediction formula based on the idea of universal gravitation to perform flow prediction on the two target posts and obtain a prediction result of whether the two target posts have a career flow relationship.

13. The system of claim 10, wherein the nodes of the career flow graph are post names, the directed edges of the career flow graph are flow relationships from one post to another post, and the weights of the edges are average tenure lengths; the encoder includes a pre-stored career flow graph, an implicit representation acquisition module and a Gaussian representation acquisition module. The hidden layer representation obtaining module is configured to extract, based on the first original feature of the target post, the second original feature of the target adjacent post, and the average tenure length from the target post to each target adjacent post, a first hidden representation of each target post and a second hidden representation of each target adjacent post by using a pre-trained graph convolution network. The Gaussian representation obtaining module is configured to calculate, based on the first hidden representations of the two target posts, second relative representations of the two target posts from the perspective of another target post.

14. The system of claim 13, wherein, The Gaussian representation obtaining module is configured to calculate, based on the first hidden representation of each target post and the second hidden representation of each target adjacent post, at least one first relative representation of the target post from the perspective of each target adjacent post, including: for each target post, concatenating the first hidden representation of the target post with the second hidden representation of each target adjacent post to obtain a same number of first concatenated representations as the number of target adjacent posts; and multiplying each first concatenated representation with a pre-trained Gaussian distribution weight matrix to obtain a same number of Gaussian representations as the number of target adjacent posts, as the first relative representation. The Gaussian representation obtaining module is configured to calculate, based on the first hidden representations of the two target posts, second relative representations of the two target posts from the perspective of another target post, including: concatenating the first hidden representations of the two target posts in the first manner and the second manner to obtain two second concatenated representations; and multiplying each second concatenated representation with a pre-set Gaussian distribution weight matrix to obtain two Gaussian representations, as the second relative representation.

15. An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

16. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method of any one of claims 1-9.

17. A computer program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1-9.

Images