CN115409479A

CN115409479A - Information processing method based on neural network, neural network and training method thereof

Info

Publication number: CN115409479A
Application number: CN202211040668.XA
Authority: CN
Inventors: 郭茁宁; 张乐; 刘浩; 秦川; 祝恒书; 熊辉
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-08-29
Filing date: 2022-08-29
Publication date: 2022-11-29

Abstract

The disclosure provides an information processing method based on a neural network, the neural network and a training method thereof, and relates to the field of artificial intelligence, in particular to a machine learning technology and a deep learning technology. The method comprises the following steps: determining company characteristics of a target company and position characteristics of a target position; processing at least one demand value corresponding to at least one time stamp included in the target time series to obtain demand timing characteristics, wherein each demand value indicates the talent demand degree of the target position of the target company at the corresponding time stamp; processing at least one supply value corresponding to the at least one timestamp to obtain supply timing characteristics, each supply value indicating a degree of talent supply by the target company to the target position at the corresponding timestamp; and processing the company characteristics, the job characteristics, the demand time sequence characteristics, and the supply time sequence characteristics to obtain a demand forecast result and a supply forecast result corresponding to both the target company and the target job.

Description

Information processing method based on neural network, neural network and training method thereof

Technical Field

The present disclosure relates to the field of artificial intelligence, and in particular, to a machine learning technique and a deep learning technique, and more particularly, to an information processing method based on a neural network, a training method of the neural network, an electronic device, a computer-readable storage medium, and a computer program product.

Background

Artificial intelligence is the subject of research that makes computers simulate some human mental processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), both at the hardware level and at the software level. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing, and the like; the artificial intelligence software technology mainly comprises a computer vision technology, a voice recognition technology, a natural language processing technology, machine learning/deep learning, a big data processing technology, a knowledge map technology and the like.

In recent years, talent contests have become more intense. Organizations and companies continually review and adjust their recruitment strategies to accommodate the rapidly changing labor market, which urgently requires forecasting of the labor market. As an essential component of the labor market analysis, labor market forecasting is intended to simulate the time-varying patterns of a labor market, including talent demand and supply variation. In fact, timely and accurate prediction of the labor market trend is not only helpful for governments and enterprises to adjust policies and recruitment strategies, but also helpful for job seekers to actively plan career roads.

The approaches described in this section are not necessarily approaches that have been previously conceived or pursued. Unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion in this section. Similarly, unless otherwise indicated, the problems mentioned in this section should not be considered as having been acknowledged in any prior art.

Disclosure of Invention

The present disclosure provides a neural network-based information processing method, a neural network training method, a neural network, an electronic device, a computer-readable storage medium, and a computer program product.

According to an aspect of the present disclosure, there is provided an information processing method based on a neural network including a first feature extraction sub-network, a second feature extraction sub-network, a third feature extraction sub-network, and a prediction sub-network. The method comprises the following steps: determining company features of the target company and position features of the target position by utilizing the first feature extraction sub-network; processing at least one demand value corresponding to at least one time stamp included in the target time sequence by using a second feature extraction sub-network to obtain a demand time sequence feature, wherein each demand value in the at least one demand value indicates the talent demand degree of the target position by the target company at the corresponding time stamp; processing at least one supply value corresponding to the at least one timestamp using a third feature extraction sub-network to obtain supply time-series features, wherein each supply value in the at least one supply value indicates a degree of talent supply of the target job by the target company at the corresponding timestamp; and processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by utilizing the prediction sub-network to obtain a demand prediction result and a supply prediction result of the target position of the target company.

According to another aspect of the present disclosure, there is provided a method of training a neural network, the neural network including a first feature extraction subnetwork, a second feature extraction subnetwork, a third feature extraction subnetwork, and a prediction subnetwork. The training method comprises the following steps: determining a model hyper-parameter of the neural network; obtaining at least one demand value and at least one supply value corresponding to at least one timestamp included in the sample time series, wherein each demand value of the at least one demand value indicates a talent demand level of the sample company for the sample position at the corresponding timestamp, and each supply value of the at least one supply value indicates a talent supply level of the sample company for the sample position at the corresponding timestamp; acquiring a real demand result and a real supply result of a sample company for both sample positions; determining company features of the sample company and position features of the sample position by using the first feature extraction sub-network; processing at least one required value by utilizing a second feature extraction sub-network to obtain a required time sequence feature; processing the at least one supply value by using a third feature extraction sub-network to obtain a supply time sequence feature; processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by using the prediction sub-network to obtain a demand prediction result and a supply prediction result of the sample company on the sample position; and obtaining the trained neural network based on the real demand result, the real supply result, the demand prediction result and the supply prediction result.

According to another aspect of the present disclosure, there is provided a neural network, including: a first feature extraction sub-network configured to determine corporate features of a target company and job features of a target job; a second feature extraction sub-network configured to process at least one demand value corresponding to at least one time stamp included in the target time series to obtain a demand time series feature, wherein each demand value of the at least one demand value indicates a talent demand degree of the target position by the target company at the corresponding time stamp; a third feature extraction sub-network configured to process at least one supply value corresponding to the at least one timestamp to obtain supply time-series features, wherein each supply value of the at least one supply value indicates a degree of talent supply of the target position by the target company at the corresponding timestamp; and a forecasting subnetwork configured to process the company characteristics, the position characteristics, the demand timing characteristics, and the supply timing characteristics to obtain a demand forecasting result and a supply forecasting result of the target position with the target company.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the method described above.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the above method.

According to another aspect of the disclosure, a computer program product is provided, comprising a computer program, wherein the computer program realizes the above method when executed by a processor.

According to one or more embodiments of the disclosure, by fusing multiple aspects of information such as characteristics related to a target company, characteristics related to a target position, and talent supply data and talent demand data related to both the target company and the target position, dynamic conditions of a labor market can be modeled more completely, and by processing time series information of supply and demand, prediction performances of the two types of data can be further improved, so that accuracy of a prediction result output by a neural network is improved.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the embodiments and, together with the description, serve to explain the exemplary implementations of the embodiments. The illustrated embodiments are for purposes of illustration only and do not limit the scope of the claims. Throughout the drawings, identical reference numbers designate similar, but not necessarily identical, elements.

Fig. 1 illustrates a schematic diagram of an exemplary system in which various methods described herein may be implemented, in accordance with embodiments of the present disclosure;

fig. 2 shows a flowchart of an information processing method according to an exemplary embodiment of the present disclosure;

FIG. 3 illustrates a flow chart for determining corporate features of a target company and job features of a target job using a first feature extraction subnetwork in accordance with an exemplary embodiment of the present disclosure;

FIG. 4 shows a flowchart of a process for utilizing an embedding subnetwork for processing a plurality of company embedded features and a plurality of job embedded features in accordance with an example embodiment of the present disclosure;

FIG. 5 illustrates a flow chart of processing a company feature, a job feature, a demand timing feature, and a supply timing feature according to an exemplary embodiment of the present disclosure;

FIG. 6 shows a flow chart of a method of training a neural network according to an exemplary embodiment of the present disclosure;

FIG. 7 illustrates a flow chart for determining a model hyper-parameter of a neural network, according to an exemplary embodiment of the present disclosure;

FIG. 8 shows a block diagram of a neural network, according to an example embodiment of the present disclosure;

FIG. 9 shows a block diagram of a first feature extraction subnetwork in accordance with an exemplary embodiment of the present disclosure;

FIG. 10 shows a block diagram of a prediction subnetwork in accordance with an example embodiment of the present disclosure; and

FIG. 11 illustrates a block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, 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 of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

In the present disclosure, unless otherwise specified, the use of the terms "first", "second", etc. to describe various elements is not intended to limit the positional relationship, the timing relationship, or the importance relationship of the elements, and such terms are used only to distinguish one element from another. In some examples, a first element and a second element may refer to the same instance of the element, and in some cases, based on the context, they may also refer to different instances.

The terminology used in the description of the various described examples in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the elements may be one or more. Furthermore, the term "and/or" as used in this disclosure is intended to encompass any and all possible combinations of the listed items.

In the related art, the traditional method relies on classical statistical models and domain expert knowledge, mainly focuses on coarse-grained labor market analysis based on survey data (such as industry-specific demand trends and geographic occupational labor market concentration), but does not consider more complex potential data dependence, while the emerging data-driven method can realize modeling on talent demand or supply through a machine learning technology, but the obtained result has low accuracy.

In order to solve the problems, the dynamic situation of the labor market can be more completely modeled by fusing various information such as characteristics related to a target company, characteristics related to a target position, talent supply data and talent demand data related to both the target company and the target position, and the prediction performance of the two data can be further improved by processing time series information of supply and demand, so that the accuracy of a prediction result output by a neural network is improved.

Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

Fig. 1 illustrates a schematic diagram of an exemplary system 100 in which various methods and apparatus described herein may be implemented in accordance with embodiments of the present disclosure. Referring to fig. 1, the system 100 includes one or

more client devices

101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120.

Client devices

101, 102, 103, 104, 105, and 106 may be configured to execute one or more applications.

In an embodiment of the present disclosure, the server 120 may run one or more services or software applications that enable the neural network-based information processing method and/or the training method of the neural network to be performed.

In some embodiments, the server 120 may also provide other services or software applications that may include non-virtual environments and virtual environments. In certain embodiments, these services may be provided as web-based services or cloud services, such as provided to users of

client devices

101, 102, 103, 104, 105, and/or 106 under a software as a service (SaaS) network.

In the configuration shown in fig. 1, server 120 may include one or more components that implement the functions performed by server 120. These components may include software components, hardware components, or a combination thereof, which may be executed by one or more processors. A user operating a

client device

101, 102, 103, 104, 105, and/or 106 may, in turn, utilize one or more client applications to interact with the server 120 to take advantage of the services provided by these components. It should be understood that a variety of different system configurations are possible, which may differ from system 100. Accordingly, fig. 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

The user may use

client devices

101, 102, 103, 104, 105, and/or 106 to enter demand and supply values for a target position for a target company corresponding to a target time series. The client device may provide an interface that enables a user of the client device to interact with the client device, e.g., the client device may receive information input by the user. The client device may also output information to the user via the interface, e.g., the client may output to the user a demand forecast and a supply forecast generated by an information processing method running on the server. Although fig. 1 depicts only six client devices, those skilled in the art will appreciate that any number of client devices may be supported by the present disclosure.

Client devices

101, 102, 103, 104, 105, and/or 106 may include various types of computer devices, such as portable handheld devices, general purpose computers (such as personal computers and laptop computers), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors or other sensing devices, and so forth. These computer devices may run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, linux, or Linux-like operating systems (e.g., GOOGLE Chrome OS); or include various Mobile operating systems such as MICROSOFT Windows Mobile OS, iOS, windows Phone, android. Portable handheld devices may include cellular telephones, smart phones, tablets, personal Digital Assistants (PDAs), and the like. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. The gaming system may include a variety of handheld gaming devices, internet-enabled gaming devices, and the like. The client device is capable of executing a variety of different applications, such as various Internet-related applications, communication applications (e.g., email applications), short Message Service (SMS) applications, and may use a variety of communication protocols.

Network 110 may be any type of network known to those skilled in the art that may support data communications using any of a variety of available protocols, including but not limited to TCP/IP, SNA, IPX, etc. By way of example only, one or more networks 110 may be a Local Area Network (LAN), an ethernet-based network, a token ring, a Wide Area Network (WAN), the internet, a virtual network, a Virtual Private Network (VPN), an intranet, an extranet, a Public Switched Telephone Network (PSTN), an infrared network, a wireless network (e.g., bluetooth, WIFI), and/or any combination of these and/or other networks.

The server 120 may include one or more general purpose computers, special purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-end servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and/or combination. The server 120 may include one or more virtual machines running a virtual operating system, or other computing architecture involving virtualization (e.g., one or more flexible pools of logical storage that may be virtualized to maintain virtual storage for the server). In various embodiments, the server 120 may run one or more services or software applications that provide the functionality described below.

The computing units in server 120 may run one or more operating systems including any of the operating systems described above, as well as any commercially available server operating systems. The server 120 may also run any of a variety of additional server applications and/or middle tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, and the like.

In some implementations, the server 120 may include one or more applications to analyze and consolidate data feeds and/or event updates received from users of the

client devices

101, 102, 103, 104, 105, and 106. Server 120 may also include one or more applications to display data feeds and/or real-time events via one or more display devices of

client devices

101, 102, 103, 104, 105, and 106.

In some embodiments, the server 120 may be a server of a distributed system, or a server incorporating a blockchain. The server 120 may also be a cloud server, or a smart cloud computing server or a smart cloud host with artificial intelligence technology. The cloud Server is a host product in a cloud computing service system, and is used for solving the defects of high management difficulty and weak service expansibility in the traditional physical host and Virtual Private Server (VPS) service.

The system 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as audio files and video files. The database 130 may reside in various locations. For example, the database used by the server 120 may be local to the server 120, or may be remote from the server 120 and may communicate with the server 120 via a network-based or dedicated connection. The database 130 may be of different types. In certain embodiments, the database used by the server 120 may be a database, such as a relational database. One or more of these databases may store, update, and retrieve data to and from the database in response to the command.

In some embodiments, one or more of the databases 130 may also be used by applications to store application data. The databases used by the application may be different types of databases, such as key-value stores, object stores, or conventional stores supported by a file system.

The system 100 of fig. 1 may be configured and operated in various ways to enable application of the various methods and apparatus described in accordance with the present disclosure.

As talents become an important competitive power among enterprises, labor trend analysis is more and more emphasized. The disclosure describes a fine-grained talent demand and supply prediction task as a joint prediction problem, and provides an information processing method based on a neural network, a training method of the neural network and the neural network on the basis of in-depth data-driven analysis.

According to an aspect of the present disclosure, there is provided a neural network-based information processing method. The neural network includes a first feature extraction sub-network, a second feature extraction sub-network, a third feature extraction sub-network, and a prediction sub-network. As shown in fig. 2, the method includes: step S201, utilizing the first feature extraction sub-network to determine company features of a target company and position features of a target position; step S202, processing at least one demand value corresponding to at least one time stamp included in the target time sequence by utilizing a second feature extraction sub-network to obtain a demand time sequence feature, wherein each demand value in the at least one demand value indicates the talent demand degree of a target position by a target company at the corresponding time stamp; step S203, processing at least one supply value corresponding to at least one timestamp by using a third feature extraction sub-network to obtain supply time sequence features, wherein each supply value in the at least one supply value indicates the talent supply degree of a target position for a target company at the corresponding timestamp; and step S204, processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by utilizing the forecast sub-network to obtain a demand forecast result and a supply forecast result of the target company for the target position.

Therefore, by fusing various information such as characteristics related to the target company, characteristics related to the target position, talent supply data and talent demand data related to both the target company and the target position, the dynamic situation of the labor market can be more completely modeled, and by processing the time series information of supply and demand, the prediction performance of the two data can be further improved, so that the accuracy of the prediction result output by the neural network is improved.

Given a target time series (t) comprising at least one time stamp _s ,t _e ) Talent demand sequence for company-position pairs (c, p) for target company and target position

(i.e., at least one demand value corresponding to at least one time stamp included in the target time series) and talent supply series

(i.e., at least one supply value corresponding to at least one timestamp), the goal of the information processing method is to predict a demand forecast result and a supply forecast result for the target position by the target company. It will be understood that t _s Is the first time stamp in the target time series, t _e Is the most in the target time seriesThe latter time stamp.

According to some embodiments, the demand forecast may characterize the talent demand level of the target position by the target company at a later time stamp of the target time series, and the supply forecast may characterize the talent supply level of the target position by the target company at the later time stamp. It is to be understood that the demand and supply forecasts can also characterize other information related to the talent demand and talent supply for the target position by the target company, and are not limited thereto.

The goal of the talent supply and demand joint prediction problem may be to simultaneously predict the demand and supply values for the target position (c, p) for the target company for the next timestamp:

wherein the content of the first and second substances,

a dynamic corporate-job profile, as will be described in more detail below.

Before modeling the supply and demand relationship of the target position of the target company, the company characteristics of the target company and the position characteristics of the target position can be obtained based on prior knowledge.

In some embodiments, various data, related information, features about the target company and target position may be embedded to obtain embedded features corresponding to the target company and target position, respectively. The neural network may also be trained using samples associated with the target company and the target position to enable the neural network to learn the embedded features corresponding to the target company and the target position.

According to some embodiments, the first feature extraction sub-network may include an embedding sub-network, a company timing feature extraction sub-network, and a position timing feature extraction sub-network. As shown in fig. 3, the step S201 of determining the company feature of the target company and the position feature of the target position by using the first feature extraction subnetwork includes: step S301, determining company embedding characteristics of the target company corresponding to each time stamp in the at least one time stamp and position embedding characteristics of the target position corresponding to each time stamp by utilizing the embedding sub-network; step S302, fusing at least one company embedded feature of the target company corresponding to at least one time stamp by utilizing a company time sequence feature extraction sub-network to obtain a company feature of the target company corresponding to a target time sequence; and step S303, fusing at least one position embedding feature of the target position corresponding to at least one timestamp by utilizing the position time sequence feature extraction sub-network to obtain a position feature corresponding to the target time sequence and aiming at the target position.

Therefore, the embedded characteristics of the target company and the target position corresponding to each time stamp are determined, and then the company characteristics and the position characteristics corresponding to the target time sequence are obtained based on the time stamp characteristics, so that the obtained characteristics can include information corresponding to each time stamp, and the subsequent prediction accuracy of demand and supply is improved.

In some embodiments, the target company and the target position may have different embedding characteristics at different time stamps, and those skilled in the art may determine the embedding characteristics of the target company and the target position at each time stamp in various ways, for example, the target company and the target position may be modeled by a time series model, and the time stamps may be directly input into a neural network for embedding the target company and the target position, which is not limited herein.

In some embodiments, a company/job embedded signature sequence H of timestamps for a target company and target job over a longer historical period (fully encompassing all timestamps in the target time sequence) may be determined, using a 0-1 vector

As a mask, and use H and

as a product with the target time series (t) _s ,t _e ) Corresponding toAt least one company/job embedding feature. In some embodiments, on this basis, a learnable attention-based mechanism vector a may be further multiplied for the company-embedded feature and the position-embedded feature, respectively, wherein the attention value corresponding to each timestamp in the target time series is included, so as to improve the prediction performance of the subsequently obtained company (sequence) feature and position (sequence) feature. It will be appreciated that a person skilled in the art may determine the attention-based mechanism vector A in various ways, and embed the feature vector A for a company _c And vector A for job embedding features _p May be the same or different, and is not limited herein.

The above process can be expressed as:

wherein

May be a company character

Or job characteristics

Is a 0-1 vector and is at t _s And t _e The value in between 1,A is a learnable attention-based vector, H is a sequence of company/job embedded features that include historical timestamps.

In modeling a priori information about a company and a position, a Graph (Graph) can be used to characterize relationships about the company and the position, and a Graph neural network can be used to learn the embedding of the company and the position. In accordance with some embodiments of the present invention,

the embedded sub-networks are graphical neural networks based on company-job bit maps. The company-job map may characterize relationships between a plurality of companies including the target company and relationships between a plurality of jobs including the target job, and/or the company-job map may characterize relationships between a plurality of companies and a plurality of jobs. In some embodiments, the company-job bitmap (hereinafter simply referred to as a map) may be an anomaly map, which includes company nodes and job nodes. Company-to-company edges (e.g., corresponding slot-skipping relationships) and job-to-job edges (e.g., corresponding job-translation relationships) may be included in the graph, thereby enabling the graph neural network to update the embedded features of the target company based on the embedded features of other companies related to the target company and to update the embedded features of the target job based on the embedded features of other jobs related to the target job. Company-job edges (e.g., corresponding supply-demand relationships) may also be included in the graph, thereby enabling the graph neural network to update the embedded features of the target company based on the embedded features of other jobs related to the target company and update the embedded features of the target job based on the embedded features of other companies related to the target job. It will be appreciated that a variety of the relationships described above may also be included in the graph, thereby enabling more complex modeling.

In some embodiments, step S301, utilizing an embedding subnetwork to determine company embedding characteristics (h) of a target company corresponding to each of at least one timestamp _c ) And job embedding features (h) of the target job corresponding to each timestamp _p ) The method comprises the following steps: for each of the at least one timestamp, processing, with an embedding sub-network, a plurality of company embedding features corresponding to the plurality of companies with respect to a last timestamp and a plurality of position embedding features corresponding to the plurality of positions with respect to the last timestamp based on a company-position graph to obtain company embedding features for the timestamp and position embedding features for the timestamp for the plurality of positions for each of the plurality of companies.

Thus, by using the graph neural network, more complex modeling of company-position can be realized, and accurate company embedding features and position embedding features can be obtained based on the relationship between company-position.

According to some embodiments, the company embedded features and the job embedded features may be further modeled based on the temporal characteristics:

H _C ,H _P ＝φ(G；H′ _C ；H′ _P ),

wherein H' _C And H' _P Is a company-embedded feature sequence and a job-embedded feature sequence input into an embedded subnetwork (graph neural network), G is a company-job graph, H _C And H _P Is the output company embedded characteristic sequence and the position embedded characteristic sequence, phi (-) represents the neural network of the graph.

The graph neural network φ (-) can comprise a total of three steps:

first, to resolve the heterogeneity, the company-job graph may be split into its three included subgraphs according to the type of the edge, namely the first subgraph G (V) that characterizes the relationship between multiple companies _c ,E _c,c ) A second sub-graph G (V) characterizing relationships between the plurality of positions _p ,E _p,p ) And a third sub-graph G (V, E) characterizing relationships between a plurality of companies and a plurality of positions _c,p )。

Second, for supply and demand type edges (E) in the three subgraphs _c,p ) Company jump type edge (E) _c,c ) And job transformation type edge (E) _p,p ) Three graph neural networks may be used to aggregate neighbor information. In one embodiment, three graph convolution operations, CPConv (), CCConv (), and PPConv (), may be used to generate point tokens based on aggregated neighbor information. Three convolution operations can be defined uniformly as:

in which h can be marked _u For embedding of point u, σ is the ReLU activation function, W is the learnable parameter, | N _v And | is a neighbor of point v. It will be appreciated that other graph neural networks besides the graph convolution network may be used.

Third, consider that CPConv (-) and CCConv (-) process the embedded feature H on behalf of the company _C To obtain two components, at CPConv (-) and PPConv (-)Manages the embedded characteristic H of the representative position _P To obtain two components, these components can be fused in various ways to obtain the embedded features for output. In one embodiment, an average update operation is applied to the company and job embedding, resulting in an output embedding feature of φ (-).

To continuously learn the chronological pattern of company and job embedding and generate a representation of the company and job on each timestamp, the learned embedding of the last timestamp can be used as input to the rotation unit (i.e., graph neural network/embedding sub-network) and output a new embedding of the current timestamp t. Based on a single unit function φ (·), a round-robin operation is defined at the time stamp t as

In this way, the time t can be obtained sequentially _s To t _e The company position of (1) embedding sequence, i.e.

And

in some embodiments, the embedding sub-network includes a first graph neural network based on a first sub-graph, a second graph neural network based on a second sub-graph, and a third graph neural network based on a third sub-graph, as shown in fig. 4, the processing, with the embedding sub-network, the plurality of company embedded features of a last timestamp corresponding to the plurality of companies and the plurality of position embedded features corresponding to the plurality of positions for each of the at least one timestamp, step S301, includes: step S401, aiming at each time stamp in at least one time stamp, processing at least one first related company embedded feature related to the last time stamp corresponding to at least one first related company adjacent to a target company in a first subgraph by using a first graph neural network to obtain a company embedded feature first component related to the time stamp corresponding to the target company; step S402, processing at least one first relevant position embedding feature related to a previous time stamp and corresponding to at least one first relevant position adjacent to the target position in the second sub-graph by utilizing a second graph neural network to obtain a position embedding feature first component related to the time stamp and corresponding to the target position; step S403, processing at least one second relevant position embedding feature which corresponds to at least one second relevant position adjacent to the target company and is relevant to the last timestamp in the third sub-graph by using a third graph neural network to obtain a company embedding feature second component which corresponds to the target company and is relevant to the timestamp; step S404, processing at least one second related company embedded feature related to a previous time stamp and corresponding to at least one second related company adjacent to the target position in the third sub-graph by using a third graph neural network to obtain a position embedded feature second component related to the time stamp and corresponding to the target position; step S405, obtaining company embedded characteristics corresponding to the target company and related to the time stamp based on the first company embedded characteristic component and the second company embedded characteristic component; and step S406, obtaining the position embedding characteristics corresponding to the target position and related to the time stamp based on the position embedding characteristic first component and the position embedding characteristic second component.

Therefore, through the mode, information among companies, positions and positions can be fully learned and utilized, and changes of company embedded features and position embedded features on a time scale are considered, so that more effective vector representation is obtained, and the accuracy of a prediction result output by a neural network is improved.

In some embodiments, the plurality of company-embedded features for initial timestamps corresponding to a plurality of companies and the plurality of position-embedded features for initial timestamps corresponding to a plurality of positions are derived by random initialization, the initial timestamp being a previous timestamp of the target time series. That is to say that the position of the first electrode,

and

is obtained by random initialization. Thus, the problem that the first time stamp of the target time series does not correspond to the previous adjacent time stamp can be solved. It is understood that the company embedded feature and the job embedded feature corresponding to the initial timestamp may be determined in other manners by those skilled in the art, and are not limited herein.

After obtaining the company characteristics of the target company and the job characteristics of the target job, the demand sequence and the supply sequence may be modeled to obtain information directly related to the historical supply and the historical demand of the target company for the target job.

According to some embodiments, the second feature extraction subnetwork comprises a demand timestamp feature extraction subnetwork and a demand timing feature extraction subnetwork. Step S202, processing at least one required value corresponding to at least one timestamp included in the target time series by using the second feature extraction subnetwork includes: for each timestamp of the at least one timestamp, determining a demand timestamp feature corresponding to the timestamp by using a demand timestamp feature extraction sub-network based on the demand value corresponding to the timestamp, the company feature of the target company, and the position feature of the target position; and processing at least one demand timestamp feature corresponding to the at least one timestamp by using the demand timing feature extraction sub-network to obtain a demand timing feature.

Therefore, embedded characteristics (namely, demand timestamp characteristics) related to demands corresponding to the timestamps are determined on the basis of the company characteristics, the job characteristics and the demand values of the target jobs of the target companies in each timestamp, and then the time series is modeled on the basis of the demand timestamp characteristics of each timestamp to obtain the demand time series characteristics, so that the prediction capability of the feature vector of the related information for representing the historical demand values is improved.

According to some embodiments, the third feature extraction sub-network comprises a supply time stamp feature extraction sub-network and a supply timing feature extraction sub-network, and the processing of the at least one supply value corresponding to the at least one time stamp with the third feature extraction sub-network of step S203 comprises: for each of the at least one timestamp, determining, with a supply timestamp feature extraction sub-network, a supply timestamp feature corresponding to the timestamp based on the supply value corresponding to the timestamp, the company feature of the target company, and the job feature of the target job; and processing at least one provisioning timestamp feature corresponding to the at least one timestamp using the provisioning timing feature extraction subnetwork to obtain the provisioning timing feature.

Therefore, embedded characteristics (namely, supply timestamp characteristics) related to supply corresponding to each timestamp are determined according to the company characteristics, the position characteristics and the supply value of the target position of the target company in each timestamp, and then the time series is modeled according to the supply timestamp characteristics of each timestamp to obtain supply time sequence characteristics, so that the prediction capability of a feature vector representing the related information of historical supply values is improved.

In some embodiments, for example, the company characteristic, the job characteristic, and the corresponding demand/supply value may be fused in various manners such as splicing, weighted summation, embedding, or any combination thereof to obtain the corresponding demand/supply timestamp characteristic, which is not limited herein.

In some embodiments, modeling the demand/supply-related time series may be accomplished, for example, by fusing at least one demand/supply timestamp signature in a self-attention-based mechanism (Transformer) manner to obtain a corresponding demand/supply timing signature. The above process can be expressed as:

where e may represent a requirement e _D Or supply e _S Time stamp feature of e ^t Can express the required time sequence characteristics

Or supply toOn-time-sequence feature

It is understood that other ways of modeling the time series may be used by one skilled in the art to derive the corresponding demand/supply timing characteristics based on at least one demand/supply timestamp characteristic, and is not limited herein.

According to some embodiments, the prediction subnetwork comprises a synthetic feature computation subnetwork, a demand prediction subnetwork, and a supply prediction subnetwork. As shown in fig. 5, the processing of the company characteristics, the job characteristics, the demand timing characteristics, and the supply timing characteristics by using the forecast sub-network in step S204 includes: s501, utilizing a comprehensive characteristic calculation sub-network to perform fusion processing on the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics to obtain supply and demand comprehensive characteristics; step S502, a demand forecasting sub-network is utilized to process the characteristics after the demand time sequence characteristics and the supply and demand comprehensive characteristics are further fused so as to obtain a demand forecasting result; and step S503, processing the feature obtained by further fusing the supply time sequence feature and the supply and demand comprehensive feature by using the supply forecasting sub-network to obtain a supply forecasting result.

Therefore, the information obtained by respectively modeling the company-position dynamic relation, the time sequence related to the demand and the time sequence related to the supply is fully utilized to obtain an accurate prediction result by fusing the relevant information about the target company embedding, the target position embedding, the historical demand value sequence and the historical supply value sequence to obtain the supply and demand comprehensive characteristic and further fusing the comprehensive characteristic with the demand time sequence characteristic and the supply time sequence characteristic respectively to obtain the final demand prediction result and supply prediction result.

According to some embodiments, the step S501 of performing a fusion process on the company feature, the position feature, the demand timing feature, and the supply timing feature by using the comprehensive feature calculation sub-network to obtain the supply and demand comprehensive feature includes: fusing the company characteristics and the position characteristics to obtain prior knowledge characteristics; and fusing the prior knowledge characteristic, the demand time sequence characteristic and the supply time sequence characteristic to obtain a supply and demand comprehensive characteristic. Therefore, the feature vector obtained based on the priori knowledge can be distinguished from the feature vector obtained based on the posterior knowledge, and the prediction performance of the obtained comprehensive feature vector is improved.

In some embodiments, the demand timing characteristic e may be fused in the following manner using the following manner _D Supply timing characteristics e _S Company characteristics

Job position feature

To obtain supply and demand comprehensive characteristics ζ:

where MLP (·) represents a multi-layered perceptron and | represents a stitching operation. It is understood that the above four features can be combined in other ways, and are not limited herein.

Further, in some embodiments, to achieve information sharing, two attention-based mechanisms may be used to couple ζ and e, respectively _D And e _E Fusion as a novel feature

And

the operation is defined as:

wherein e represents e _D Or e _S ，

To represent

Or

w represents a learnable parameter.

In some embodiments, the method will comprise

And

processing the output as eta by two independent multi-layer perceptrons using LogSoftMax _D And η _S . Specifically, the dimensionality of the output vector is equal to the number of trend types, and the ith element is the probability of predicting the ith trend type. Furthermore, argMax conversion η may be used _D And η _S Is a trend type y _D And y _S Is composed of

Where y is y _D Or y _S Eta is eta _D Or η _S ，η ⁱ Is the ith element of eta, N _y Is the number of trend types. In one exemplary embodiment, the trend types may include the following five items: large rise, large fall, small rise, small fall, and substantially unchanged.

In some embodiments, other pairs of approaches may be used

And

processing to obtain other forecasted results related to talent demand and talent supply by the target company for the target job。

According to another aspect of the present disclosure, a method of training a neural network is provided. The neural network includes a first feature extraction sub-network, a second feature extraction sub-network, a third feature extraction sub-network, and a prediction sub-network. As shown in fig. 6, the training method includes: s601, determining a model hyper-parameter of a neural network; step S602, obtaining at least one demand value and at least one supply value corresponding to at least one time stamp included in the sample time sequence, wherein each demand value in the at least one demand value indicates the talent supply degree of the sample company to the sample position at the corresponding time stamp, and each supply value in the at least one supply value indicates the talent supply degree of the sample company to the sample position at the corresponding time stamp; step S603, acquiring a real demand result and a real supply result of the sample company for the sample position; step S604, determining company features of a sample company and position features of a sample position by utilizing the first feature extraction sub-network; step S605, processing at least one required value by utilizing a second characteristic extraction sub-network to obtain required time sequence characteristics; step S606, processing at least one supply value by utilizing a third feature extraction sub-network to obtain a supply time sequence feature; step S607, processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by utilizing the prediction sub-network to obtain a demand prediction result and a supply prediction result of the sample company for the sample position; and step S608, obtaining the trained neural network based on the real demand result, the real supply result, the demand prediction result and the supply prediction result. The operations of step S603 to step S607 in fig. 6 are similar to the operations of step S201 to step S204 and their sub-steps in fig. 2, and are not repeated here.

From the previous preliminary data analysis, the demand and supply of different companies follow a long tailed distribution. The end-to-end model naturally yields good performance for these companies with large amounts of training data, but does not accurately predict the "small sample" situation for these long-tailed companies. Therefore, a meta-learner based on loss-driven sampling can be introduced to train the overall prediction framework and specially optimize the long-tailed task. First, the end-to-end prediction model can be optimized using the negative log likelihood loss of the Poisson distribution

Wherein y represents y _D Or y _S Eta represents eta _D And η _S . To simplify the operation, the last term of the equation can be estimated from the Stirling equation as

The optimization objective of the model ensemble is to combine the predicted losses of demand and supply

And optimized by back propagation. In addition, the problem of long tail distribution can be alleviated by a meta-learning problem. The goal is to extract globally shared meta-knowledge from different companies to enable fast adaptation and more accurate predictions when predicting the needs and supplies of companies with limited data. In particular, the talent demand-supply forecast for each company can be formulated as a separate task and the following set of tasks can be constructed.

Defining a task set as: marking

For company c _i E C supply and demand forecast task, the task set is defined as supply and demand forecast task set for each company

Where | C | is the number of companies.

Different from the existing meta-learning method of the equal probability sampling task, a loss-driven sampling strategy can be designed in the meta-learning process so as to strengthen the research of the model on the long-tail task. Intuitively, the more training-lossy task indicates the greater prediction error, requiring additional effort to learn, according to the equation. Meta-training based on loss-driven sampling is described in detail below.

First, model parameters θ are initialized randomly and for each task

Setting equal probability

In the j-th round (epoch), the model parameter adaptation process of several steps (steps) can be run. In each adaptation step, a task can be sampled according to the sampling probability

As a support set, its loss is evaluated

And obtain parameter updates

Where theta' is the updated parameter,

is a task

Learning device f _θ Gradient of losses of j-th round of (a). Correspondingly, a task can be sampled in the same way

As a target set

Wherein

Evaluated in the meta learning step mentioned before.

Before the next round, the sampling probability can be updated as:

wherein

Is the sampling probability of the j +1 epoch,

is the validation set loss for the jth epoch.

According to some embodiments, as shown in fig. 7, the step S601 of determining the model hyper-parameters of the neural network includes: step S701, determining initial hyper-parameters of a neural network; step S702, determining a sampling probability corresponding to each of a plurality of companies; step S703, determining a target sample set from respective samples of a plurality of companies based on respective sampling probabilities of the plurality of companies; step S704, training the initial hyper-parameter by using a target sample set to obtain an updated hyper-parameter and loss values corresponding to a plurality of companies; and step S705, in response to determining that the preset convergence condition is not met, based on the loss value of each of the plurality of target companies, adjusting the respective sampling probabilities of the plurality of companies to update the target sample set and further train the updated hyper-parameters. Therefore, through the method, the model trained based on the learned hyperparameters can give accurate prediction results for different positions of different companies (particularly for the companies and positions in the long tail distribution).

In some embodiments, a person skilled in the art may set a corresponding convergence condition according to a requirement, for example, whether to converge may be determined according to a loss value corresponding to a sample of each company (or a sample set obtained by sampling from a sample of each company), which is not limited herein.

It is understood that one skilled in the art can use various ways to train the neural network based on the learned hyper-parameters to obtain a trained neural network with demand/supply forecasting capability. For example, one skilled in the art can construct a loss function based on the demand truth, the supply truth, the demand forecast, and the supply forecast, and adjust parameters of the neural network based on the constructed loss function and the learned hyperparameters to obtain a trained neural network.

According to another aspect of the present disclosure, a neural network 800 is provided. As shown in fig. 8, the neural network includes: a first feature extraction subnetwork 810 configured to determine corporate features of a target company and job features of a target job; a second feature extraction sub-network 820 configured to process at least one demand value 802 corresponding to at least one time stamp included in the target time series to obtain a demand timing feature, wherein each demand value of the at least one demand value indicates a talent demand degree of the target position by the target company at the corresponding time stamp; a third feature extraction sub-network 830 configured to process the at least one supply value 804 corresponding to the at least one timestamp to obtain supply timing features, wherein each supply value of the at least one supply value indicates a degree of talent supply of the target position by the target company at the corresponding timestamp; and a forecasting subnetwork 840 configured to process the company characteristics, the job characteristics, the demand timing characteristics, and the supply timing characteristics to obtain the demand forecast 806 and the supply forecast 808 for the target job of the target company. The operation of the neural network 800 is similar to the operations of steps S201 to S204 in fig. 2, and is not repeated herein.

According to some embodiments, as shown in FIG. 9, a first feature extraction subnetwork 900 may comprise: an embedding subnetwork 910 configured to determine company embedding characteristics of the target company corresponding to each of the at least one timestamp and job embedding characteristics of the target job corresponding to each timestamp; a company temporal feature extraction sub-network 920 configured to fuse at least one company-embedded feature of the target company corresponding to the at least one timestamp to obtain a company feature 922 of the target company corresponding to the target time series; and a job time series feature extraction sub-network 930 configured to fuse the at least one job embedding feature of the target job corresponding to the at least one timestamp to obtain a job feature 932 for the target job corresponding to the target time series.

According to some embodiments, the embedded sub-networks are company-job bitmap based graph neural networks. The company-job graph characterization may include relationships between a plurality of companies of the target company and relationships between a plurality of jobs including the target job, and/or the company-job graph characterizes relationships between the plurality of companies and the plurality of jobs. The embedding subnetwork may be further configured to, for each of the at least one timestamp, process, based on the company-job graph, a plurality of company embedding features corresponding to the plurality of companies with respect to a last timestamp and a plurality of job embedding features corresponding to the plurality of jobs with respect to the last timestamp to obtain a company embedding feature for the timestamp and a job embedding feature for the timestamp for each of the plurality of companies.

According to some embodiments, a company-job graph includes a first sub-graph that may characterize relationships between a plurality of companies, a second sub-graph that characterizes relationships between a plurality of jobs, and a third sub-graph that characterizes relationships between a plurality of companies and a plurality of jobs. Embedding the sub-network 910 may include: a first graph neural network 912 configured to, for each of the at least one timestamp, process at least one first related company embedded feature for a last timestamp corresponding to at least one first related company adjacent to the target company in the first sub-graph to obtain a company embedded feature first component for the timestamp corresponding to the target company; a second graph neural network 914 configured to process, for each of the at least one timestamp, at least one first related position embedded feature for a last timestamp in the second sub-graph corresponding to at least one first related position adjacent to the target position to obtain a position embedded feature first component for the timestamp corresponding to the target position; and a third graph neural network 916 configured to, for each of the at least one timestamp, process at least one second relevant position embedding feature for a last timestamp corresponding to at least one second relevant position adjacent to the target company in the third sub-graph to obtain a company embedding feature second component for the timestamp corresponding to the target company; and processing at least one second related company embedded feature for a previous timestamp corresponding to at least one second related company adjacent to the target position in the third sub-graph to obtain a position embedded feature second component for the timestamp corresponding to the target position. The embedded subnetwork may be further configured to: obtaining a company embedded feature corresponding to the target company with respect to the timestamp based on the first component of the company embedded feature and the second component of the company embedded feature; and obtaining a job embedding feature corresponding to the target job with respect to the timestamp based on the job embedding feature first component and the job embedding feature second component.

According to some embodiments, the plurality of company-embedded features for the initial time stamp corresponding to the plurality of companies and the plurality of position-embedded features for the initial time stamp corresponding to the plurality of positions are obtained by random initialization, the initial time stamp being a previous time stamp of the target time series.

According to some embodiments, the second feature extraction sub-network may comprise: a demand timestamp feature extraction sub-network configured to determine, for each of at least one timestamp, a demand timestamp feature corresponding to the timestamp based on the demand value corresponding to the timestamp, the company feature of the target company, and the position feature of the target position; and a demand timing feature extraction sub-network configured to process at least one demand timestamp feature corresponding to the at least one timestamp to obtain a demand timing feature.

According to some embodiments, the third feature extraction subnetwork may comprise: a provisioning timestamp feature extraction sub-network configured to determine, for each of at least one timestamp, a provisioning timestamp feature corresponding to the timestamp based on a provisioning value corresponding to the timestamp, a company feature of the target company, and a position feature of the target position; and a supply timing feature extraction sub-network configured to process the at least one supply timestamp feature corresponding to the at least one timestamp to obtain a supply timing feature.

According to some embodiments, as shown in fig. 10, prediction subnetwork 1000 may comprise: a comprehensive characteristic calculation sub-network 1010 configured to perform fusion processing on the company characteristic 1004, the position characteristic 1006, the demand timing characteristic 1002, and the supply timing characteristic 1008 to obtain a supply and demand comprehensive characteristic; a demand forecasting sub-network 1020 configured to process the feature obtained by further fusing the demand time sequence feature and the supply and demand comprehensive feature to obtain a demand forecasting result 1022; and a supply prediction subnetwork 1030 configured to process the feature obtained by further fusing the supply timing feature and the supply-and-demand comprehensive feature to obtain a supply prediction result 1032.

According to some embodiments, the integrated feature computation subnetwork may be further configured to: fusing the company characteristics and the position characteristics to obtain prior knowledge characteristics; and fusing the prior knowledge characteristic, the demand time sequence characteristic and the supply time sequence characteristic to obtain a supply and demand comprehensive characteristic.

According to some embodiments, the demand forecast may characterize the talent demand level of the target position by the target company at a later time stamp of the target time series, and the supply forecast may characterize the talent supply level of the target position by the target company at the later time stamp.

In the technical scheme of the disclosure, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations and do not violate the good customs of the public order.

According to an embodiment of the present disclosure, there is also provided an electronic device, a readable storage medium, and a computer program product.

Referring to fig. 11, a block diagram of a structure of an electronic device 1100, which may be a server or a client of the present disclosure, which is an example of a hardware device that may be applied to aspects of the present disclosure, will now be described. Electronic device is intended to represent various forms of digital electronic computer devices, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 11, the device 1100 comprises a computing unit 1101, which may perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a Random Access Memory (RAM) 1103. In the RAM 1103, various programs and data necessary for the operation of the device 1100 may also be stored. The calculation unit 1101, the ROM 1102, and the RAM 1103 are connected to each other by a bus 1104. An input/output (I/O) interface 1105 is also connected to bus 1104.

A number of components in device 1100 connect to I/O interface 1105, including: an input unit 1106, an output unit 1107, a storage unit 1108, and a communication unit 1109. The input unit 1106 may be any type of device capable of inputting information to the device 1100, and the input unit 1106 may receive input numeric or character information and generate key signal inputs related to user settings and/or function controls of the electronic device, and may include, but is not limited to, a mouse, a keyboard, a touch screen, a track pad, a track ball, a joystick, a microphone, and/or a remote control. Output unit 1107 may be any type of device capable of presenting information and may include, but is not limited to, a display, speakers, a video/audio output terminal, a vibrator, and/or a printer. Storage unit 1108 may include, but is not limited to, a magnetic or optical disk. The communication unit 1109 allows the device 1100 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and/or chipsets, such as bluetooth (TM) devices, 802.11 devices, wiFi devices, wiMax devices, cellular communication devices, and/or the like.

The computing unit 1101 can be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning network algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 1101 performs the various methods and processes described above, such as a neural network-based information processing method and/or a training method of a neural network. For example, in some embodiments, the neural network-based information processing method and/or the neural network training method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 1108. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 1100 via ROM 1102 and/or communication unit 1109. When the computer program is loaded into RAM 1103 and executed by the computing unit 1101, one or more steps of the neural network based information processing method and/or the training method of the neural network described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to perform the neural network-based information processing method and/or the training method of the neural network by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, causes the functions/acts specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may 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 a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

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 a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user may provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; 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 may be received in any form, including acoustic, speech, or tactile input.

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: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The Server can be a cloud Server, also called a cloud computing Server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server may also be a server of a distributed system, or a server incorporating a blockchain.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be performed in parallel, sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

While embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it is to be understood that the above-described methods, systems and apparatus are merely illustrative embodiments or examples and that the scope of the invention is not to be limited by these embodiments or examples, but only by the claims as issued and their equivalents. Various elements in the embodiments or examples may be omitted or may be replaced with equivalents thereof. Further, the steps may be performed in an order different from that described in the present disclosure. Further, various elements in the embodiments or examples may be combined in various ways. It is important that as technology evolves, many of the elements described herein may be replaced with equivalent elements that appear after the present disclosure.

Claims

1. A neural network-based information processing method, the neural network including a first feature extraction sub-network, a second feature extraction sub-network, a third feature extraction sub-network, and a prediction sub-network, the method comprising:

determining company features of the target company and position features of the target position by utilizing the first feature extraction sub-network;

processing at least one demand value corresponding to at least one time stamp included in a target time sequence by using a second feature extraction sub-network to obtain demand time sequence features, wherein each demand value in the at least one demand value indicates the talent demand degree of the target company on the target position at the corresponding time stamp;

processing at least one supply value corresponding to the at least one timestamp using a third feature extraction sub-network to obtain supply time-series features, wherein each supply value of the at least one supply value indicates a talent supply degree of the target company for the target position at the corresponding timestamp; and

and processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by utilizing a prediction sub-network to obtain a demand prediction result and a supply prediction result of the target company for the target position.

2. The method of claim 1, wherein the first feature extraction sub-network comprises an embedding sub-network, a corporate time series feature extraction sub-network, and a job time series feature extraction sub-network, wherein determining corporate features of the target corporate and job features of the target job using the first feature extraction sub-network comprises:

determining, with the embedding subnetwork, a company embedding feature of the target company corresponding to each of the at least one timestamp and a job embedding feature of the target job corresponding to each timestamp;

fusing at least one company embedded feature of the target company corresponding to the at least one timestamp by using the company time series feature extraction sub-network to obtain a company feature of the target company corresponding to the target time series; and

and utilizing the position time sequence feature extraction sub-network to fuse at least one position embedding feature of the target position corresponding to the at least one time stamp so as to obtain a position feature corresponding to the target time sequence and aiming at the target position.

3. The method of claim 2, wherein the embedded sub-network is a corporate-job graph-based graph neural network, wherein the corporate-job graph characterization includes relationships between a plurality of companies of the target company and relationships between a plurality of jobs including the target job, and/or the corporate-job graph characterizes relationships between the plurality of companies and the plurality of jobs,

wherein determining, with the embedding subnetwork, a company embedding characteristic of the target company corresponding to each of the at least one timestamp and a position embedding characteristic of the target position corresponding to each timestamp comprises:

for each of the at least one timestamp, processing, with the embedding sub-network, a plurality of company embedding features corresponding to the plurality of companies with respect to a last timestamp and a plurality of position embedding features corresponding to the plurality of positions with respect to a last timestamp based on the company-position map to obtain company embedding features for the timestamp for each of the plurality of companies and position embedding features for the timestamp for each of the plurality of positions.

4. The method of claim 3, wherein the company-position graph includes a first subgraph characterizing relationships between the plurality of companies, a second subgraph characterizing relationships between the plurality of positions, and a third subgraph characterizing relationships between the plurality of companies and the plurality of positions, the embedded sub-network includes a first graph neural network based on the first subgraph, a second graph neural network based on the second subgraph, and a third graph neural network based on the third subgraph,

wherein for each of the at least one timestamp, processing, with the embedding sub-network, a plurality of company-embedded features of a last timestamp corresponding to the plurality of companies and a plurality of job-embedded features corresponding to the plurality of jobs comprises:

for each of the at least one timestamp,

processing at least one first related company embedded feature with respect to a previous timestamp corresponding to at least one first related company adjacent to the target company in the first sub-graph by using the first graph neural network to obtain a company embedded feature first component with respect to the timestamp corresponding to the target company;

processing, with the second graph neural network, at least one first relevant position embedding feature for a last timestamp corresponding to at least one first relevant position adjacent to the target position in the second sub-graph to obtain a position embedding feature first component for the timestamp corresponding to the target position;

processing at least one second relevant position embedded feature corresponding to at least one second relevant position adjacent to the target company in the third sub-graph and related to a previous time stamp by using the third graph neural network to obtain a company embedded feature second component corresponding to the target company and related to the time stamp;

processing at least one second related company embedded feature corresponding to at least one second related company adjacent to the target position in the third sub-graph with respect to a previous timestamp by using the third graph neural network to obtain a position embedded feature second component corresponding to the target position with respect to the timestamp;

obtaining a company embedded feature corresponding to the target company with respect to the time stamp based on the company embedded feature first component and the company embedded feature second component; and

and obtaining the position embedding feature corresponding to the target position and related to the time stamp based on the position embedding feature first component and the position embedding feature second component.

5. The method of claim 3 or 4, wherein a plurality of company-embedded features corresponding to the plurality of companies with respect to an initial timestamp and a plurality of job-embedded features corresponding to the plurality of jobs with respect to the initial timestamp are derived by random initialization, the initial timestamp being a previous timestamp of the target time series.

6. The method of claim 1, wherein the second feature extraction sub-network comprises a demand timestamp feature extraction sub-network and a demand timing feature extraction sub-network, wherein processing at least one demand value corresponding to at least one timestamp included in the target time series with the second feature extraction sub-network comprises:

for each of the at least one timestamp, determining, with the demand timestamp feature extraction subnetwork, a demand timestamp feature corresponding to the timestamp based on the demand value corresponding to the timestamp, the company feature of the target company, and the job feature of the target job; and

and processing at least one requirement timestamp feature corresponding to the at least one timestamp by using the requirement time sequence feature extraction sub-network to obtain the requirement time sequence feature.

7. The method of claim 6, wherein the third feature extraction sub-network comprises a provisioning timestamp feature extraction sub-network and a provisioning timing feature extraction sub-network, wherein processing the at least one provisioning value corresponding to the at least one timestamp with the third feature extraction sub-network comprises:

for each timestamp of the at least one timestamp, determining, with the supply timestamp feature extraction sub-network, a supply timestamp feature corresponding to the timestamp based on the supply value corresponding to the timestamp, the company feature of the target company, and the position feature of the target position; and

processing at least one provisioning timestamp feature corresponding to the at least one timestamp using the provisioning timing feature extraction subnetwork to obtain the provisioning timing feature.

8. The method of claim 1, wherein the prediction sub-network comprises a synthetic feature computation sub-network, a demand prediction sub-network, and a supply prediction sub-network, wherein processing the company features, the job features, the demand timing features, and the supply timing features with a prediction sub-network comprises:

utilizing the comprehensive characteristic calculation sub-network to perform fusion processing on the company characteristic, the position characteristic, the demand time sequence characteristic and the supply time sequence characteristic to obtain a supply and demand comprehensive characteristic;

processing the feature after the demand time sequence feature and the supply and demand comprehensive feature are further fused by using the demand forecasting sub-network to obtain a demand forecasting result; and

and processing the feature after further fusing the supply time sequence feature and the supply and demand comprehensive feature by using the supply prediction sub-network to obtain the supply prediction result.

9. The method of claim 8, wherein fusing the corporate characteristics, the job characteristics, the demand timing characteristics, and the supply timing characteristics with the integrated characteristic computation sub-network to obtain a supply and demand integrated characteristic comprises:

fusing the company features and the position features to obtain prior knowledge features; and

and fusing the prior knowledge characteristic, the demand time sequence characteristic and the supply time sequence characteristic to obtain the supply and demand comprehensive characteristic.

10. The method of claim 1, wherein the demand forecast characterizes a degree of talent demand of the target company for the target position at a later time stamp of the target time series, and the supply forecast characterizes a degree of talent supply of the target company for the target position at the later time stamp.

11. A method of training a neural network, the neural network comprising a first feature extraction subnetwork, a second feature extraction subnetwork, a third feature extraction subnetwork, and a prediction subnetwork, the method comprising:

determining a model hyperparameter of the neural network;

obtaining at least one demand value and at least one supply value corresponding to at least one timestamp included in a sample time series, wherein each demand value of the at least one demand value indicates a talent demand level of a sample company for a sample position at the corresponding timestamp, and each supply value of the at least one supply value indicates a talent supply level of the sample company for the sample position at the corresponding timestamp;

acquiring a demand real result and a supply real result of the sample company for both the sample positions;

determining company features of the sample company and position features of the sample position using the first feature extraction sub-network;

processing the at least one demand value by utilizing the second feature extraction sub-network to obtain a demand time sequence feature;

processing the at least one provisioning value using the third feature extraction sub-network to obtain a provisioning timing feature;

processing the company characteristics, the position characteristics, the demand time sequence characteristics and the supply time sequence characteristics by using the forecast sub-network to obtain a demand forecast result and a supply forecast result of the sample company for the sample position; and

and obtaining a trained neural network based on the real demand result, the real supply result, the demand prediction result and the supply prediction result.

12. The method of claim 11, wherein determining model hyper-parameters of the neural network comprises:

determining an initial hyper-parameter of the neural network;

determining a sampling probability corresponding to each of a plurality of companies;

determining a target sample set from respective samples of the plurality of companies based on respective sampling probabilities of the plurality of companies;

training the initial hyper-parameter by using the target sample set to obtain an updated hyper-parameter and loss values corresponding to the plurality of companies respectively; and

in response to determining that the preset convergence condition is not satisfied, adjusting respective sampling probabilities of the plurality of companies based on the loss value of each of the plurality of target companies to update the target sample set and further train the updated hyper-parameters.

13. A neural network, comprising:

a first feature extraction sub-network configured to determine company features of a target company and job features of a target job;

a second feature extraction sub-network configured to process at least one demand value corresponding to at least one timestamp included in a target time series to obtain a demand timing feature, wherein each demand value of the at least one demand value indicates a talent demand degree of the target position by the target company at the corresponding timestamp;

a third feature extraction sub-network configured to process at least one supply value corresponding to the at least one timestamp to obtain a supply timing feature, wherein each supply value of the at least one supply value indicates a talent supply degree of the target company for the target position at the corresponding timestamp; and

a forecasting subnetwork configured to process the company characteristics, the job characteristics, the demand timing characteristics, and the supply timing characteristics to obtain a demand forecasting result and a supply forecasting result of the target company for the target job.

14. The neural network of claim 13, wherein the first feature extraction sub-network comprises:

an embedding subnetwork configured to determine company embedding characteristics of the target company corresponding to each of the at least one timestamp and job embedding characteristics of the target job corresponding to each timestamp;

a company time series feature extraction sub-network configured to fuse at least one company embedded feature of the target company corresponding to the at least one timestamp to obtain a company feature of the target company corresponding to the target time series; and

a position time sequence feature extraction sub-network configured to fuse at least one position embedding feature of the target position corresponding to the at least one timestamp to obtain a position feature for the target position corresponding to the target time sequence.

15. The neural network of claim 14, wherein the embedded sub-network is a corporate-role graph-based graph neural network, wherein the corporate-role graph characterization includes relationships between a plurality of companies of the target company and relationships between a plurality of roles including the target role, and/or the corporate-role graph characterizes relationships between the plurality of companies and the plurality of roles,

wherein the embedding sub-network is further configured to, for each of the at least one timestamp, process a plurality of company-embedded features corresponding to the plurality of companies with respect to a last timestamp and a plurality of job-embedded features corresponding to the plurality of jobs with respect to a last timestamp based on the company-job map to obtain a company-embedded feature for the timestamp and a job-embedded feature for the timestamp for each of the plurality of jobs for each of the plurality of companies.

16. The neural network of claim 15, wherein the company-position graph includes a first subgraph characterizing relationships between the plurality of companies, a second subgraph characterizing relationships between the plurality of positions, and a third subgraph characterizing relationships between the plurality of companies and the plurality of positions,

wherein the embedded subnetwork comprises:

a first graph neural network configured to, for each of the at least one timestamp, process at least one first related company embedded feature for a last timestamp corresponding to at least one first related company adjacent to the target company in the first subgraph to obtain a company embedded feature first component for the timestamp corresponding to the target company;

a second graph neural network configured to, for each of the at least one timestamp, process at least one first relevant position embedding feature for a last timestamp corresponding to at least one first relevant position adjacent to the target position in the second sub-graph to obtain a position embedding feature first component for the timestamp corresponding to the target position; and

a third graph neural network configured to, for each of the at least one timestamp,

processing at least one second relevant position embedding feature corresponding to at least one second relevant position adjacent to the target company in the third subgraph and related to a previous timestamp to obtain a company embedding feature second component corresponding to the target company and related to the timestamp; and

processing at least one second related company embedded feature for a previous timestamp corresponding to at least one second related company adjacent to the target position in the third sub-graph to obtain a position embedded feature second component for the timestamp corresponding to the target position;

wherein the embedded subnetwork is further configured to:

obtaining a company embedded feature corresponding to the target company with respect to the timestamp based on the company embedded feature first component and the company embedded feature second component; and

17. The neural network of claim 15 or 16, wherein a plurality of company-embedded features corresponding to the plurality of companies with respect to an initial timestamp and a plurality of position-embedded features corresponding to the plurality of positions with respect to the initial timestamp are derived by random initialization, the initial timestamp being a previous timestamp of the target time series.

18. The neural network of claim 13, wherein the second feature extraction sub-network comprises:

a demand timestamp feature extraction sub-network configured to determine, for each of the at least one timestamp, a demand timestamp feature corresponding to the timestamp based on the demand value corresponding to the timestamp, the company feature of the target company, and the position feature of the target position; and

a demand timing feature extraction sub-network configured to process at least one demand timestamp feature corresponding to the at least one timestamp to obtain the demand timing feature.

19. The neural network of claim 18, wherein the third feature extraction sub-network comprises:

a supply timestamp feature extraction sub-network configured to determine, for each of the at least one timestamp, a supply timestamp feature corresponding to the timestamp based on the supply value corresponding to the timestamp, the company feature of the target company, and the job feature of the target job; and

a supply timing feature extraction sub-network configured to process at least one supply timestamp feature corresponding to the at least one timestamp to obtain the supply timing feature.

20. The neural network of claim 13, wherein the prediction subnetwork comprises:

a comprehensive characteristic calculation sub-network configured to perform fusion processing on the company characteristic, the position characteristic, the demand time sequence characteristic and the supply time sequence characteristic to obtain a supply and demand comprehensive characteristic;

a demand forecasting sub-network configured to process the feature obtained by further fusing the demand time sequence feature and the supply and demand comprehensive feature to obtain the demand forecasting result; and

and the supply prediction sub-network is configured to process the feature after the supply time sequence feature and the supply and demand comprehensive feature are further fused to obtain the supply prediction result.

21. The neural network of claim 20, wherein the synthetic feature computation subnetwork is further configured to:

22. The neural network of claim 13, wherein the demand forecast characterizes a degree of talent demand of the target company for the target position at a later time stamp of the target time series, and the supply forecast characterizes a degree of talent supply of the target company for the target position at the later time stamp.

23. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein

The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-10.

24. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-10.

25. A computer program product comprising a computer program, wherein the computer program realizes the method of any one of claims 1-10 when executed by a processor.