CN115330142A

CN115330142A - Training method of joint capacity model, capacity requirement matching method and device

Info

Publication number: CN115330142A
Application number: CN202210878392.6A
Authority: CN
Inventors: 李满伟; 秦川; 马海平; 申大忠; 祝恒书; 张敬帅; 姚开春
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-07-25
Filing date: 2022-07-25
Publication date: 2022-11-11
Anticipated expiration: 2042-07-25
Also published as: CN115330142B

Abstract

The disclosure provides a training method of a joint ability model, an ability requirement matching method and a device, and relates to the technical field of artificial intelligence, in particular to the field of deep learning. The specific implementation scheme is as follows: processing the sample data by adopting a plurality of capability prediction models to obtain a plurality of capability information; processing the various kinds of capability information and the demand information by adopting a demand matching model to obtain a demand matching result, wherein the demand information comprises capability information required to be met by a target scene; and training the demand matching model and the multiple capacity prediction models based on the demand matching result to update the parameters of the demand matching model and the multiple capacity prediction models. According to the method, the demand matching model and the multiple capacity prediction models are trained by using multiple sample data, and the combined capacity model obtained after training can meet more comprehensive and more-dimensional capacity prediction and demand matching, and is suitable for richer application scenarios.

Description

Training method of joint capacity model, capacity requirement matching method and device

Technical Field

The present disclosure relates to the field of artificial intelligence technology, and more particularly, to the field of deep learning.

Background

The capability assessment can select the most appropriate candidate by assessing the consistency of the candidate's skills with the job requirements, which is a key task in talent recruitment. Traditional assessment of competency involves multiple processes, taking different forms of assessment methods, often resulting in assessment conclusions that are discrete, noisy and unreliable. Expert decision making or natural language processing techniques and the like can be adopted for decision making at present.

Disclosure of Invention

The disclosure provides a training method of a joint capacity model, a capacity requirement matching method and a capacity requirement matching device.

According to an aspect of the present disclosure, there is provided a method for training a joint capability model, the joint capability model including a multiple capability prediction model and a demand matching model, the method including:

processing the sample data by adopting the multiple capability prediction model to obtain multiple capability information;

processing the various kinds of capability information and the demand information by adopting the demand matching model to obtain a demand matching result, wherein the demand information comprises capability information required to be met by a target scene;

and training the demand matching model and the multiple capacity prediction models based on the demand matching result so as to update the parameters of the demand matching model and the multiple capacity prediction models.

According to another aspect of the present disclosure, there is provided a capability requirement matching method, including:

processing various data to be processed of a candidate by adopting a plurality of capability prediction models to obtain a plurality of capability information of the candidate;

and processing the various kinds of capability information and the demand information by adopting a demand matching model to obtain a demand matching result of the candidate, wherein the demand information comprises capability information which needs to be met by a target scene.

According to another aspect of the present disclosure, there is provided a training apparatus for a joint capacity model including a plurality of capacity prediction models and a demand matching model, the apparatus including:

the capability prediction module is used for processing the sample data by adopting the multiple capability prediction models to obtain multiple capability information;

the demand matching module is used for processing the various kinds of capacity information and demand information by adopting the demand matching model to obtain a demand matching result, and the demand information comprises capacity information required to be met by a target scene;

and the training module is used for training the demand matching model and the multiple capability prediction models based on the demand matching result so as to update the parameters of the demand matching model and the multiple capability prediction models.

According to another aspect of the present disclosure, there is provided a capability requirement matching apparatus including:

the capability prediction module is used for processing various data to be processed of the candidate by adopting various capability prediction models to obtain various capability information of the candidate;

and the demand matching module is used for processing the various kinds of capacity information and demand information by adopting a demand matching model to obtain a demand matching result of the candidate, wherein the demand information comprises capacity information required to be met by the target scene.

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 a method according to any one of the embodiments of the present disclosure.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to perform a method according to any one of the embodiments of the present disclosure.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method according to any one of the embodiments of the present disclosure.

According to the embodiment of the disclosure, the demand matching model and the multiple capability prediction models are trained by using multiple sample data, and the combined capability model obtained after training can meet more comprehensive and more-dimensional capability prediction and demand matching, and is suitable for richer application scenarios.

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 drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a flow diagram of a method of training a joint capability model according to an embodiment of the present disclosure;

FIG. 2 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 3 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 4 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 5 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 6 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 7 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 8 is a flow diagram of a method of training a joint capability model according to another embodiment of the present disclosure;

FIG. 9 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 10 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 11 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 12 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 13 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 14 is a flow diagram of a capability requirement matching method according to an embodiment of the present disclosure;

FIG. 15 is a schematic diagram of a training apparatus for a joint capacity model according to an embodiment of the present disclosure;

FIG. 16 is a schematic diagram of a training apparatus for a joint capacity model according to another embodiment of the present disclosure;

FIG. 17 is a schematic diagram of a capability requirement matching device according to an embodiment of the present disclosure;

FIG. 18 is a schematic structural diagram of a capability requirement matching device according to another embodiment of the present disclosure;

fig. 19 is a block diagram of a joint capacity diagnostic method in talent recruitment according to the present disclosure;

FIG. 20 is a framework diagram of the written modeling of the present disclosure;

FIG. 21 is a framework diagram of the resume modeling of the present disclosure;

FIG. 22 is a block diagram of an electronic device used to implement methods of 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 and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

FIG. 1 is a flow chart diagram of a method for training a joint capability model according to an embodiment of the present disclosure. The joint capability model may include a plurality of capability prediction models and a demand matching model, and the training method may include:

s101, processing sample data by adopting a plurality of capability prediction models to obtain a plurality of capability information;

s102, processing the various kinds of capability information and the demand information by adopting a demand matching model to obtain a demand matching result, wherein the demand information comprises capability information required to be met by a target scene;

s103, training the demand matching model and the multiple capacity prediction models based on the demand matching result so as to update parameters of the demand matching model and the multiple capacity prediction models.

In embodiments of the present disclosure, the joint capability model may include a variety of capability prediction models and demand matching models. According to different application scenarios, different capability prediction models can be constructed. For example, in a job-hunting scenario, the plurality of capability prediction models may include a written-ability prediction model, a resume capability prediction model, an interview capability prediction model, and the like. For another example, in the scenes of learning, consulting and the like, the multiple capability prediction models may include a written-test capability prediction model, an interview capability prediction model and the like.

In the embodiment of the disclosure, the requirement matching model can match various kinds of capability information with specific requirements to obtain a requirement matching result. For example, the demand match results may include an enrollment probability, an admission probability, an enrollment probability, and the like. For another example, the requirement matching result may also include a result of whether the requirement is matched with a specific post or school.

In the embodiment of the disclosure, in the training process, multiple capability prediction models can be adopted to respectively perform corresponding processing on multiple sample data to obtain multiple capability information. The multiple sample data may include a batch of multiple sample data, and may be specifically classified according to the type of the capability prediction model, for example, a batch of 3A sample data includes a sample written test data, a sample resume data, and a sample interview data. A is a positive integer. Of course, the number of sample written data, sample resume data, and sample interview data may also be different.

In the embodiment of the present disclosure, the requirement information includes capability information that the target scene needs to satisfy. The demand information may be different for different target scenarios. For example, the requirement information of the application scenario includes a post-related skill specifically required for a certain post, which may also be referred to as skill requirement information of the post, for example: familiar computer languages, technical expertise, etc. For another example, the requirement information of the entrance scenario includes learning-related skills specifically required by a certain school, and may also be referred to as skill requirement information of the learning, such as english level, discipline score, competition reward, and the like. And then, matching the predicted capacity information of each capacity prediction model with specific demand information, such as skill demand information of a post, by using a demand matching model to obtain a demand matching result. And then training the demand matching model and each capacity prediction model based on a demand matching result to update parameters of the demand matching model and each capacity prediction model until the overall loss function is converged, and stopping training. And performing demand matching prediction on data related to the capabilities of the candidates of various data sources by using the multiple capability prediction models and the demand matching models in the trained joint capability model.

In the embodiment of the disclosure, the demand matching model and the multiple capability prediction models are trained by using multiple sample data, and the trained joint capability model can meet more comprehensive and more-dimensional capability prediction and demand matching, and is suitable for richer application scenarios. Further, the processing efficiency of some application scenes can be improved, for example, the recruitment efficiency in job hunting scenes can be improved. Furthermore, the subjectivity factor of the prediction result can be reduced, and the method has strong expandability.

FIG. 2 is a flow chart diagram of a method for training a joint capability model according to another embodiment of the present disclosure. The method of this embodiment includes one or more features of the training method embodiments described above. In a possible implementation manner, the method for processing the sample data by using the multiple capability prediction model to obtain multiple capability information includes at least two steps:

s201, processing sample written test data by using a written test capability prediction model to obtain written test capability information;

s202, processing sample interview data by adopting an interview capability prediction model to obtain interview capability information;

and S203, processing the sample resume data by adopting the resume capability prediction model to obtain resume capability information.

In the disclosed embodiment, sample written test data can be selected from written test data of a plurality of candidates and generated. For example, the candidate's written test data may include text generated from the candidate's test paper, such as an enrollment test, an admission test, and the like. The written test data can include written test questions and answer results of candidates for each question. Each test question can have a corresponding number, and the answer result of each question of the candidate can be represented by a specific numerical value. For example, 0 indicates correct and 1 indicates error. For another example, 10 means 100 points, 9 means 90 points, 8 means 80 points, and the like. In the disclosed embodiments, the questions may also be referred to as exercises, practice exercises, or questions, etc.

In the disclosed embodiment, sample interview written data can be selected and generated from interview data of a plurality of candidates. The interview data of the candidate can comprise evaluation texts of the interviewer on various skills of the candidate in the interview process, and can also comprise texts related to the skills of the candidate extracted from audio and video data of the interview and the like. For example, the skills of the candidates may include, but are not limited to, english-language grades, computer-level grades, computer software use skills, graduate college professional related skills, and the like.

In the disclosed embodiment, sample resume data may be selected and generated from resume data of several candidates. The resume data of the candidate may include information on whether the candidate has a certain skill or not.

At least two kinds of capability information of the candidate can be obtained through at least two kinds of the stroke trial capability prediction model, the interview capability prediction model and the resume capability prediction model, and therefore the combined capability model can be updated through a combined training method based on the comprehensive capability of the candidate, and therefore more comprehensive capability prediction and requirement matching are achieved.

In a possible embodiment, as shown in fig. 3, processing the sample written test data by using the written test capability prediction model to obtain written test capability information includes:

s301, extracting candidate characteristics from the sample written test data;

and S302, obtaining the stroke test capability information of the candidate based on the candidate characteristics.

In the embodiment of the disclosure, the candidate feature in the written description data may include a macroscopic characterization of the candidate with respect to the written description, and may be represented in the form of a vector or the like, and the specific capability of the candidate embodied in the written description can be accurately obtained through the candidate feature in the written description data. For example, the candidate feature extracted from the sample written data may be candidate u _i One-hot vector of

The candidate's written ability information may be a macro characterization of the candidate with respect to the written examination

May be a d-dimensional vector.

Wherein W ^U May be an embedded matrix, W ^U May have an initial value. May be applied to W during subsequent training ^U And learning and updating, and further updating the written test ability information. After training is finished, W after training in the stroke test capability prediction model is utilized ^U And candidate characteristics in the real stroke test data of a certain candidate can accurately predict the stroke test capability information of the candidate.

In a possible implementation mode, the loss function of the stroke capability prediction model is constructed based on the probability of correct answer predicted by the stroke capability prediction model and the actual answer result; and the probability of correct answer is obtained by predicting the interaction information of the candidate and the question obtained by the stroke ability prediction model based on the sample stroke data. Parameters of the stroke test capability prediction model can be updated by using the loss function of the stroke test capability prediction model, so that a more accurate stroke test capability prediction model is obtained.

For example, the loss function of the stroke capability prediction model is loss ^E ：

Wherein,

is candidate u _i Answering questions e _j The real situation of (a) is,

predicting candidate u for a model _i Answering questions e _j The case (1). Minimizing the loss function based on written data may learn the ability of each candidate

U denotes a candidate set.

Representing a set of topics.

In one possible implementation, as shown in fig. 4, the step of obtaining the interaction information between the candidate and the topic includes:

s401, extracting candidate characteristics and question characteristics from the sample written test data;

s402, obtaining the stroke test capability information based on the candidate characteristics;

s403, obtaining written examination skill representation information based on the written examination capability information and the overall skill representation information;

s404, obtaining title difficulty information and title distinguishing degree information based on the title characteristics;

s405, obtaining interaction information of the candidate and the question based on the stroke skill characterization information, the question difficulty information, the question distinguishing degree information and the question associated skill information.

In one example, assume that there are N candidates, M exercises, K skills, e.g., number u _i ∈R ¹ ^×N Exercise number e _j ∈R ^1×M . Wherein R is ^1×N Representing a 1 × N vector. R ^1×M Representing a 1 × M vector, other similar expressions have similar meanings.

The candidate features extracted from the sample written data may include candidates u _i One-hot vector of

The title feature can include a title e _j One-hot vector of

The test ability information obtained based on the candidate characteristics can be macro-characterization of the candidate about the test

The overall skill characterization information may be a skill characterization matrix h ^S ∈R ^K×d . Based on written test ability informationThe skill characterization information of the written test obtained from the overall skill characterization information can be a characterization vector of the candidate on the specific skill of the written test

sigmoid represents an activation function. The superscript T denotes transpose.

The topic difficulty information obtained based on the topic characteristics can be a topic difficulty parameter

The topic distinguishing degree information obtained based on the topic characteristics can be a topic distinguishing degree parameter

For example, the topic difficulty parameter can be a vector and the discrimination parameter of the topic can be a scalar. W is a group of ^diff 、W ^disc May be a learnable embedded matrix.

The topic associated skill information can be a topic associated skill vector

Shows the skill situation of topic investigation. For example, if topic e _j And (5) inspecting the skill s, wherein the s-th position is 1, and otherwise, the s-th position is 0. Based on the stroke skill characterization information, the topic difficulty information, the topic discrimination information and the topic associated skill information, the obtained interaction information of the candidate and the topic can be referred to the following formula:

wherein x represents interactive information;

characterization vectors on specific skills for a candidate with respect to a written test, i.e. written test skillsThe information can be represented;

the question difficulty parameter, namely the question difficulty information;

dividing parameters for the titles, namely, title dividing information;

indicates a topic associated skill vector, i.e., topic associated skill information. Symbol

By multiplication in para-position, i.e.

And

multiply by the alignment.

In embodiments of the present disclosure, the stroke capability prediction model may include a neural network Fully Connected Layers (FCLs) of candidate interactions with topics. The input information of the neural network FCL may include interaction information x of the candidate and the topic. The output information of the neural network FCL may include a probability of predicting the answer to be correct

The trial-and-error capability information of the candidate can also be obtained based on the candidate characteristics and the learnable embedded matrix in the neural network. The stroke capability information may be a stroke capability vector.

The interaction information of the candidate and the questions can be obtained based on a project reaction theory formula of educational psychology, so that the interaction situation of the candidate and each question in the written test data can be determined more scientifically and reasonably, and a more appropriate loss function can be obtained.

In one possible implementation, as shown in fig. 5, the processing the sample interview data by using the interview capability prediction model to obtain the interview capability information includes:

s501, extracting initial interview representation information from the sample interview data;

s502, obtaining a first mean value and a first variance based on the interview representation information;

s503, obtaining first representation information conforming to Gaussian distribution based on the first mean value and the first variance, and sampling from the first representation information to obtain a first distribution latent variable;

s504, obtaining first theme distribution information based on the first distribution latent variable;

s505, obtaining a first topic set based on the first topic representation information and the overall skill representation information;

s506, obtaining predicted interview representation information based on the first theme distribution information and the first theme set;

and S507, obtaining interview capability information based on the first theme distribution information and the first theme representation information.

In one example, the initial interview characterization information can be input vectors of an interview capability prediction model

The input vector includes skills extracted from the interview text. For example, if there is a skill s in the interview text, s-th (the s-th position) is 1, otherwise it is 0.

Will vector

Respectively obtaining a first mean value mu through two full-connection layers (or called full-connection neural networks) ^A And the first square difference σ ^A . Then, the signal is adjusted to fit into Gaussian distribution N (mu) ^A ,(σ ^A ) ² ) Is sampled to obtain a first distributed latent variable such as a Gaussian random vector

Then, the softmax layer is passed to obtain the first topic distribution information such as topic distribution(Vector)

Representing first topic representation information, e.g. topic representation matrix t ^A And overall skill characterization information such as skill characterization matrix h ^S Cross multiplication and softmax layer crossing to obtain the first theme set beta ^A (the set may be a matrix). Wherein, beta ^A May include distribution of subject matter over the skill word

Indicating the proficiency of each topic k at the corresponding skill,

finally, the first theme is distributed with information

With the first topic set beta ^A Multiplying to obtain a predicted value, i.e. predicted interview characterization information, e.g.

The specific values in the predicted interview representation information and the initial interview representation information are generally different. In addition, if training is needed to continue, the predicted interview representation information can be used as the initial interview representation information of the next time. Interview capability information

May be a vector. Furthermore, t ^A May comprise

A specific vector representing a topic k.

The first topic distribution information described above may indicate the grasping condition under each topic (skill set) in accordance with the probability of each topic.

The first topic characterization information may be a matrix indicating that there are multiple topics in the system regarding the skill of the interview, wherein each topic represents a group of closer skill words. Each topic may have a token vector, i.e. a topic-specific vector.

Initial interview representation information, predicted interview representation information, theme distribution, theme set and the like related to interviews are obtained through sample interview data, subsequent training is facilitated, and more accurate interview capability information can be obtained.

In one possible embodiment, the loss function of the interview capability prediction model is constructed based on the first topic distribution information, the first distribution latent variable, the predicted interview representation information and the first topic set. Parameters of the interview capability prediction model can be updated by using the loss function of the interview capability prediction model, and a more accurate interview capability prediction model is obtained.

For example, an example of a formula for a loss function of an interview capability prediction model is as follows:

wherein,

is a face test _i The distribution of the subject matter of (a),

is a potential variable of the distribution and is,

is a face test _i Characterization of beta ^A Is the subject set of interviews. Where "|" represents a prior probability problem; "|" indicates juxtaposition.

Representation-based prior distribution

And posterior distribution

The distance of (c).

In one possible implementation, as shown in fig. 6, processing the sample resume data by using the resume capability prediction model to obtain resume capability information includes:

s601, extracting initial resume representation information from the sample resume data;

s602, obtaining a second mean value and a second variance based on the resume representation information;

s603, obtaining second characterization information which accords with Gaussian distribution based on the second mean value and the second variance, and sampling from the second characterization information to obtain a second distribution latent variable;

s604, obtaining second topic distribution information based on the second distribution latent variable;

s605, obtaining a second topic set based on the second topic representation information and the overall skill representation information;

s606, obtaining predicted resume representation information based on the second theme distribution information and the second theme set;

s607, resume capability information is obtained based on the second theme distribution information and the second theme representation information.

In one example, the resume characterization information may be an input vector of a resume capability prediction model

The input vector includes skills extracted from the resume text. For example, if there is a skill s in the resume text, the s-th position is 1, otherwise it is 0.

Will vector

Respectively obtaining a second mean value mu through two full-connection layers (or called full-connection neural networks) ^R And a firstVariance of two σ ^R . Then, the following is satisfied with Gaussian distribution N (mu) ^R ,(σ ^R ) ² ) Is sampled to obtain a second distribution latent variable such as a Gaussian random vector

Then, the softmax layer is passed to obtain second topic distribution information such as topic distribution vector

Characterizing second topic information, e.g., topic characterization matrix t ^R And overall skill characterization information such as skill characterization matrix h ^S Cross multiplication and softmax layer crossing to obtain a second theme set beta ^R (the set may be a matrix). Wherein, beta ^R May include distribution of subject matter in the skill word (vector)

Indicating the proficiency of each topic k at the corresponding skill,

finally, the second theme is distributed with information

With a second topic set β ^R Multiplying to obtain a predicted value, i.e. predicted resume characterization information

In addition, resume capability information

May be a vector. Furthermore, t ^R May comprise

A specific vector representing a topic k.

The second topic distribution information described above may indicate the grasping condition under each topic (skill group) in correspondence with the probability of each topic.

The second topic characterization information may be a matrix indicating that there are multiple topics in the system regarding the skills of the resume, where each topic represents a group of closer skill words. Each topic may have a token vector, i.e. a topic-specific vector.

And obtaining topic distribution, topic set and the like related to the resume through sample resume data, and obtaining predicted resume characterization information based on initial resume characterization information prediction, thereby facilitating subsequent training and obtaining more accurate resume capability information.

In one possible embodiment, the loss function of the resume capability prediction model is constructed based on the second topic distribution information, the second distribution latent variable, the predicted resume characterization information, and the second topic set. Parameters of the resume capability prediction model can be updated by using the loss function of the resume capability prediction model, so that a more accurate resume capability prediction model is obtained.

For example, an example of a formula for a loss function of a resume capability prediction model is as follows:

wherein,

is a resume r _i The distribution of the subject matter of (a),

is a potential variable of the distribution and is,

is a resume r _i Characterization of beta ^R Is a collection of topics of resumes. Where "|" represents a prior probability problem; "|" indicates juxtaposition。

Representation-based prior distribution

And posterior distribution

Of the distance of (c).

In one possible implementation, as shown in fig. 7, processing the multiple kinds of capability information and the requirement information by using a requirement matching model to obtain a requirement matching result includes:

s701, obtaining overall capability information based on the multiple types of capability information and the attention weight;

s702, obtaining specific skill information based on the overall ability information and the skill characterization information;

and S703, obtaining a predicted requirement matching result based on the specific skill information and the post skill requirement information.

For example, the written ability information is

Resume capability information of

The interview ability information is

Obtaining the overall macro capability of the candidate, namely the overall capability information h by adopting an attention mechanism _i ∈R ^1×d Wherein

and h _i May be a vector.

Wherein, a _E 、a _R 、a _A To focus on the weights, these weights may be updated during the training process.

The matching degree between the candidate and the position can be identified through the requirement matching model, and a proper position can be recommended for the candidate or a proper candidate can be recommended for the position based on the overall capability information of the candidate and the specific skill requirement information of the position.

In one possible implementation, the penalty function of the demand matching model is constructed based on the predicted demand matching results and the real demand matching results. Parameters of the demand matching model can be updated by using the loss function of the demand matching model, and a more accurate demand matching model is obtained.

For example, after obtaining the macro-ability of the candidate, the candidate is associated with the skill characterization matrix h ^S Cross-ride learning of a candidate, e.g. job seeker, in a particular skill _i ∈R ^1×K ，α _i May be a vector.

α _i ＝sigmoid(h _i ×(h ^S ) ^T )

Skill requirement gamma of post _p ∈R ^1×K Indicates the position p _p Requirement for specific skills, gamma _p May be a vector, whose formula is exemplified as follows:

wherein,

is the one-hot vector of the position number, W ^J Is a learnable training matrix.

Finally, the vector alpha of the candidate in the mastery condition of the specific skill is obtained _i Skill requirement vector gamma with position _p Bit multiplication, predicted probability of successful application:

wherein alpha is _i,s Grasping condition vector γ representing skill s (s-th skill) _p,s Representing a position p _p The requirement vector of skill s.

An example of a penalty function for a demand matching model according to the above example is as follows:

wherein, y _i Can represent candidate u _i Whether the application for the successful tag is true or not,

model prediction candidates u can be represented _i The probability of success is applied.

In one possible embodiment, as shown in fig. 8, training the demand matching model and the multiple capacity prediction models based on the demand matching result to update the parameters of the demand matching model and the multiple capacity prediction models includes:

s801, obtaining an integral loss function based on the loss function of the demand matching model and the loss functions of the multiple capability prediction models;

s802, under the condition that updating is determined to be needed based on the overall loss function, the demand matching model is updated by using the loss function of the demand matching model, and corresponding capacity prediction models are respectively updated by using the loss functions of the multiple capacity prediction models.

For example, an example of an overall loss function is as follows:

loss＝α*loss ^E +β*loss ^R +γ*loss ^A +η*loss ^D

where α, β, γ, and η are weights corresponding to the loss functions of the capability prediction models and the demand matching models, and may be empirical values. In practical application, the specific values of α, β, γ, and η can be adjusted according to the convergence speed of each model.

In the embodiment of the present disclosure, in the case where the value of the overall loss function is not less than the set threshold value, the parameters of the demand matching model and each capability prediction model may be updated. When the value of the overall loss function is smaller than the set threshold value, the overall loss function converges, and updating of the overall model may be stopped. The multiple models included in the joint capability model can be jointly trained through the integral loss function and the loss functions of the models, and the training efficiency is improved.

In one possible embodiment, updating the demand matching model with a penalty function of the demand matching model includes:

and updating at least one of attention weight, overall skill characterization information and post skill requirement information in the requirement matching model based on the loss function of the requirement matching model. And updating the parameters of the demand matching model, so that the demand matching model can obtain a more appropriate candidate and post matching result.

In the embodiment of the present disclosure, based on the above example, the parameters that need to be updated in the demand matching model may include one or more of the following: attention weights such as a _E 、a _R 、a _A Global skill characterization information such as h ^S And position skill requirement information such as gamma _j Embedded matrix W of ^J . And if the loss function of the requirement matching model is not less than the set threshold value, at least one of the attention weight, the overall skill characterization information and the post skill requirement information in the requirement matching model can be updated. And under the condition that the value of the overall loss function is smaller than the set threshold value, the loss function of the demand matching model can stop updating the demand matching model.

In a possible implementation, updating the corresponding capability prediction models respectively by using the loss functions of the plurality of capability prediction models includes at least two steps:

updating at least one item of written test ability information, subject difficulty information, subject distinguishing degree information and overall skill representation information of the written test ability prediction model based on a loss function of the written test ability prediction model;

updating at least one of the parameters of the full-connection layer, the first theme representation information and the overall skill representation information of the interview capability prediction model based on the loss function of the interview capability prediction model;

and updating at least one of the full-link layer parameter, the second theme representation information and the overall skill representation information of the resume capability prediction model based on the loss function of the resume capability prediction model.

For example, if the combined capability model includes a written ability prediction model and an interview ability prediction model, the parameters of the written ability prediction model are updated according to the loss function of the written ability prediction model, and the parameters of the interview ability prediction model are updated according to the loss function of the interview ability prediction model. For another example, if the combined capability model includes a stroke capability prediction model and a resume capability prediction model, the parameters of the stroke capability prediction model are updated according to the loss function of the stroke capability prediction model, and the parameters of the resume capability prediction model are updated according to the loss function of the resume capability prediction model.

In the embodiment of the present disclosure, in the updating of multiple models, the overall skill characterization information needs to be updated, so the overall skill characterization information may be updated in a certain order. For example, updating the overall skill representation information according to the sequence of the resume capability prediction model, the written test capability prediction model and the interview capability prediction model. For another example, the overall skill characterization information is updated according to the sequence of the stroke test ability prediction model, the resume ability prediction model and the interview ability prediction model. And the overall skill characterization information after the last model is updated can be used as the update basis of the next model.

By updating the parameters of each capability prediction model, more comprehensive and more-dimensional capability prediction can be met.

FIG. 9 is a flow diagram illustrating a capability requirement matching method according to an embodiment of the disclosure. The matching method may include:

s901, processing various data to be processed of a candidate by adopting a various capability prediction model to obtain various capability information of the candidate;

s902, processing the various kinds of capability information and the demand information by adopting a demand matching model to obtain a demand matching result of the candidate, wherein the demand information comprises capability information which needs to be met by a target scene.

The capability prediction model and the requirement matching model in this embodiment may be capability prediction models and requirement matching models in a combined capability model obtained by training using any one of the training methods in the above embodiments. The plurality of data to be processed may include, for example, one or more of real written-test data, real interview data, and real resume data of the candidate. For example, the real stroke test data, the real interview data and the real resume data of a certain candidate are input into the joint capability model, and the real stroke test data, the real interview data and the real resume data of the candidate are respectively processed by the multiple capability prediction models of the joint capability model to obtain the stroke test capability information, the interview capability information and the resume capability information of the candidate.

In the embodiment of the disclosure, the joint capability model obtained after training can meet more comprehensive and more-dimensional capability prediction and requirement matching, and is suitable for richer application scenarios.

In a possible implementation, as shown in fig. 10, the processing the multiple types of data to be processed of the candidate by using the multiple types of capability prediction models to obtain the multiple types of capability information of the candidate includes at least two steps of:

s1001, processing the to-be-processed stroke test data by adopting a stroke test capability prediction model to obtain stroke test capability information;

s1002, processing interview data to be processed by adopting an interview capability prediction model to obtain interview capability information;

s1003, processing resume data to be processed by adopting a resume capability prediction model to obtain resume capability information.

In the embodiment of the present disclosure, the to-be-processed written examination data may be actual written examination data of the candidate, such as written examination question records of school test paper, job entry test paper, and the like. The interview data to be processed can be real interview data of the candidate, such as an assessment text of an interviewer on various skills of a certain candidate in an interview process, a text related to the skills of the certain candidate extracted from audio and video data of interview, and the like. The to-be-processed resume data may be actual resume data of a candidate, such as resume electronic text of a certain candidate.

At least a plurality of kinds of capability information of the candidate can be obtained through at least two of the stroke trial capability prediction model, the interview capability prediction model and the resume capability prediction model, and further more comprehensive and more accurate demand matching can be achieved based on the plurality of kinds of capability information of the candidate.

In a possible embodiment, as shown in fig. 11, processing the to-be-processed stroke test data by using a stroke test capability prediction model to obtain stroke test capability information includes:

s1101, extracting candidate characteristics from the written examination data to be processed;

and S1102, acquiring the stroke test capability information of the candidate based on the candidate characteristics.

In the embodiment of the disclosure, the candidate feature in the written examination data to be processed, that is, the real written examination data, may include a macro characterization of the candidate with respect to the written examination, and the specific ability embodied by the candidate in the written examination may be accurately obtained through the candidate feature in the written examination data. For example, the candidate feature extracted from the to-be-processed written data, i.e., the truth, may be the candidate u ₁ One-hot vector of

The candidate's written-test ability information may be a macro characterization of the candidate with respect to the written-test

May be a d-dimensional vector.

Wherein W ^U May be an embedded matrix, W ^U May have an initial value. May be applied to W during subsequent training ^U To learn and update, and furtherAnd updating the written test capability information. After training is finished, W after training in the stroke test capability prediction model is utilized ^U The one-hot encoding with the real stroke test data of a certain candidate can predict the stroke test capability information of the candidate.

In addition, the trained stroke test capability prediction model can be dynamically updated based on the real stroke test data, and the specific process can be referred to as a process of updating by using the training stroke test data. The manner of constructing the computational loss function of the stroke capability prediction model can also be referred to the related description in the training method of the above embodiment.

In a possible implementation, as shown in fig. 12, processing the interview data to be processed by using the interview capability prediction model to obtain interview capability information includes:

s1201, extracting initial interview representation information from the interview data to be processed;

s1202, obtaining a first mean value and a first variance based on the interview representation information;

s1203, obtaining first representation information according with Gaussian distribution based on the first mean value and the first square difference, and sampling from the first representation information to obtain a first distribution latent variable;

s1204, obtaining first topic distribution information based on the first distribution latent variable;

s1205, obtaining a first topic set based on the first topic representation information and the overall skill representation information;

s1206, obtaining predicted interview representation information based on the first theme distribution information and the first theme set;

s1207, obtaining interview capability information based on the first theme distribution information and the first theme representation information.

Referring to the example of FIG. 5, candidates u are extracted from interview data to be processed, i.e., real interview data ₁ Initial interview characterization information of

As an input vector to the interview ability prediction model. Vector quantity

Respectively obtaining a first mean value mu through two full-connection layers (or called full-connection neural networks) ^A And first square difference sigma ^A . Then, the signal is adjusted to fit into Gaussian distribution N (mu) ^A ,(σ ^A ) ² ) Is sampled to obtain a first distribution latent variable such as a Gaussian random vector

Then, the softmax layer is passed to obtain first topic distribution information such as topic distribution vector

Representing first topic representation information, e.g. topic representation matrix t ^A And overall skill characterization information such as skill characterization matrix h ^S Cross multiplication and softmax layer crossing to obtain the first theme set beta ^A . Finally, the first theme is distributed with information

With the first topic set beta ^A Multiplying to obtain a predicted value, i.e. predicted interview characterization information

In addition, interview capability information

Initial interview representation information, predicted interview representation information, theme distribution, theme set and the like related to the interview can be obtained through real interview data of the candidate, and more accurate interview capability information of the candidate can be obtained.

In addition, the trained stroke ability prediction model can be dynamically updated based on the real interview data, and the specific process can be referred to as the process of updating by using the training interview data. The way of constructing the computational loss function of the interview capability prediction model can also be referred to the related description in the training method of the above embodiment.

In a possible implementation, as shown in fig. 13, processing resume data to be processed by using a resume capability prediction model to obtain resume capability information includes:

s1301, extracting initial resume representation information from the resume data to be processed;

s1302, obtaining a second mean value and a second variance based on the resume representation information;

s1303, obtaining second characterization information conforming to Gaussian distribution based on the second mean value and the second variance, and sampling from the second characterization information to obtain a second distribution latent variable;

s1304, obtaining second topic distribution information based on the second distribution latent variable;

s1305, obtaining a second topic set based on the second topic representation information and the overall skill representation information;

s1306, obtaining predicted resume representation information based on the second theme distribution information and the second theme set;

and S1307, obtaining resume capability information based on the second theme distribution information and the second theme representation information.

Referring to the example of FIG. 6, candidate u is extracted from resume data to be processed, i.e., real resume data ₁ Interview characterization information of

As an input vector to the resume capability prediction model. Vector quantity

Respectively obtaining a second mean value mu through two full-connection layers (or called full-connection neural networks) ^R And a second variance σ ^R . Then, the signal is adjusted to fit into Gaussian distribution N (mu) ^R ,(σ ^R ) ² ) Is sampled to obtain a second distribution latent variable such as a Gaussian random vector

Then, the softmax layer is passed to obtain second topic distribution information such as topic distribution directionMeasurement of

Characterizing second topic information, e.g., topic characterization matrix t ^R And overall skill characterization information such as skill characterization matrix h ^S Cross multiplication and softmax layer crossing to obtain a second theme set beta ^R . Finally, the second theme is distributed with information

In addition, interview capability information

Initial resume characterization information, predicted resume characterization information, theme distribution, theme set and the like related to the resumes are obtained through the real resume data of the candidate, and more accurate resume capability information of the candidate can be obtained.

In addition, the trained resume capability prediction model can be dynamically updated based on the real resume data, and the specific process can be referred to as a process of updating by using the training resume data. The manner of constructing the computational loss function of the resume ability prediction model can also be referred to the related description in the training method of the above embodiment.

In one possible implementation, as shown in fig. 14, the processing the multiple kinds of capability information and requirement information by using the requirement matching model to obtain the requirement matching result of the candidate includes:

s1401, obtaining overall capability information based on the multiple types of capability information and attention weight;

s1402, obtaining specific skill information based on the overall ability information and the skill characterization information;

and S1403, obtaining a predicted requirement matching result based on the specific skill information and the post skill requirement information.

E.g. based on candidate u ₁ The stroke capability information of

Resume capability information of

The interview ability information is

Obtaining the overall macro capability of the candidate, namely the overall capability information h by adopting an attention mechanism ₁ ∈R ^1×d 。

Wherein, a _E 、a _R 、a _A To focus the weights, the weights may be trained weights.

In obtaining the macroscopic power h of the candidate ₁ Then, h is mixed ₁ And skill characterization matrix h ^S Cross-ride obtains the mastery condition alpha of a candidate, such as a job seeker, in a particular skill ₁ ∈R ^1×K 。

α ₁ ＝sigmoid(h ₁ ×(h ^S ) ^T )

Position p ₁ Skill requirement gamma of ₁ ∈R ^1×K An example of the formula is as follows:

wherein,

is post p ₁ One-hot vector of (W) ^J Is the trained matrix.

Finally, the candidate u ₁ Vector alpha of mastery in a particular skill ₁ And post p ₁ Skill requirement vectorγ ₁ Bit multiplication, predicted probability of success application:

wherein alpha is _1,s Represents candidate u ₁ The vector of the mastery of the skill s, γ _1,s Representing a position p ₁ The requirement vector gamma of skill s ₁ 。

FIG. 15 is a schematic diagram of a training apparatus for a joint capacity model including a plurality of capacity prediction models and a demand matching model according to an embodiment of the disclosure, and the apparatus may include:

the capability prediction module 1501 is used for processing the sample data by adopting a plurality of capability prediction models to obtain a plurality of capability information;

a requirement matching module 1502, configured to process the multiple types of capability information and requirement information by using a requirement matching model to obtain a requirement matching result, where the requirement information includes capability information that needs to be met by a target scene;

the training module 1503 is configured to train the demand matching model and the multiple capability prediction models based on the demand matching result to update parameters of the demand matching model and the multiple capability prediction models.

In one possible implementation, as shown in fig. 16, the capability prediction module 1501 includes at least two of the following:

the stroke test capability prediction sub-module 15011 is used for processing the sample stroke test data by adopting a stroke test capability prediction model to obtain stroke test capability information;

the interview capability prediction sub-module 15012 is used for processing sample interview data by adopting the interview capability prediction model to obtain interview capability information;

and the resume capability prediction sub-module 15013 is used for processing the sample resume data by adopting the resume capability prediction model to obtain resume capability information.

In one possible implementation, the trial capability prediction sub-module 15011 is further configured to extract candidate features from the sample trial data; and obtaining the stroke test capability information of the candidate based on the candidate characteristic.

In a possible implementation mode, the loss function of the stroke capability prediction model is constructed based on the probability of correct answer predicted by the stroke capability prediction model and the actual answer result; and the probability of correct answer is obtained by predicting the interaction information of the candidate and the question obtained by the stroke ability prediction model based on the sample stroke data.

In a possible implementation, the trial capability prediction sub-module 15011 is further configured to obtain interaction information of the candidate and the topic, including:

extracting candidate features and topic features from the sample written data;

obtaining stroke trial capability information based on the candidate features;

obtaining written test skill representation information based on the written test capability information and the overall skill representation information;

obtaining question difficulty information and question distinguishing degree information based on the question characteristics;

and obtaining the interaction information of the candidate and the question based on the stroke skill characterization information, the question difficulty information, the question distinguishing degree information and the question associated skill information.

In one possible implementation, the interview capability prediction sub-module 15012 is configured to:

extracting initial interview representation information from the sample interview data;

obtaining a first mean value and a first variance based on the interview representation information;

obtaining first representation information conforming to Gaussian distribution based on the first mean value and the first variance, and sampling the first representation information to obtain a first distribution latent variable;

obtaining first topic distribution information based on the first distribution latent variable;

obtaining a first topic set based on the first topic representation information and the overall skill representation information;

obtaining predicted interview representation information based on the first theme distribution information and the first theme set;

and obtaining interview capability information based on the first theme distribution information and the first theme representation information.

In one possible embodiment, the loss function of the interview capability prediction model is constructed based on the first topic distribution information, the first distribution latent variable, the predicted interview representation information and the first topic set.

In one possible implementation, the resume capability prediction sub-module 15013 is further configured to:

extracting initial resume characterization information from the sample resume data;

obtaining a second mean value and a second variance based on the resume characterization information;

obtaining second representation information which accords with Gaussian distribution based on the second mean value and the second variance, and sampling from the second representation information to obtain a second distribution latent variable;

obtaining second topic distribution information based on the second distribution latent variable;

obtaining a second topic set based on the second topic representation information and the overall skill representation information;

obtaining predicted resume representation information based on the second theme distribution information and the second theme set;

and obtaining resume capability information based on the second theme distribution information and the second theme representation information.

In one possible embodiment, the loss function of the resume capability prediction model is constructed based on the second topic distribution information, the second distribution latent variable, the predicted resume characterization information, and the second topic set.

In one possible implementation, as shown in FIG. 16, the requirement matching module 1502 includes:

an overall capability sub-module 15021 for obtaining overall capability information based on the plurality of capability information and the attention weight;

a specific skill submodule 15022 for deriving specific skill information based on the overall capability information and the skill characterization information;

and a requirement matching sub-module 15023 for obtaining a predicted requirement matching result based on the specific skill information and the post skill requirement information.

In one possible implementation, the penalty function of the demand matching model is constructed based on the predicted demand matching results and the real demand matching results.

In one possible implementation, the training module 1503 includes:

an overall loss sub-module 15031 configured to obtain an overall loss function based on the loss function of the demand matching model and the loss functions of the multiple capability prediction models;

an updating sub-module 15032 is configured to, in a case where it is determined that updating is required based on the overall loss function, update the demand matching model using the loss functions of the demand matching model, and update the corresponding capacity prediction models using the loss functions of the plurality of capacity prediction models, respectively.

In one possible implementation, the update sub-module 15032 is configured to update the demand matching model with a penalty function of the demand matching model, including: updating at least one of attention weight, global skill information, and post skill requirement information in the requirement matching model based on the loss function of the requirement matching model.

In one possible embodiment, the updating sub-module 15032 is further configured to update the corresponding capability prediction models with the loss functions of the plurality of capability prediction models, respectively, including at least two of:

Fig. 17 is a schematic structural diagram of a capability requirement matching apparatus according to an embodiment of the present disclosure, which may include:

the capability prediction module 1701 is used for processing various to-be-processed data of a candidate by adopting various capability prediction models to obtain various capability information of the candidate;

a requirement matching module 1702, configured to process the multiple pieces of capability information and requirement information by using a requirement matching model to obtain a requirement matching result of the candidate, where the requirement information includes capability information that needs to be met by the target scene.

In one possible implementation, as shown in fig. 18, the capability prediction module 1701 includes at least two of:

the stroke test capability prediction submodule 17011 is configured to process the to-be-processed stroke test data by using a stroke test capability prediction model to obtain stroke test capability information;

an interview capability prediction submodule 17012 for processing interview data to be processed by using an interview capability prediction model to obtain interview capability information;

and the resume capability prediction submodule 17013 is configured to process the resume data to be processed by using the resume capability prediction model, so as to obtain resume capability information.

In a possible implementation, the tentative capability prediction module 17011 is further configured to extract candidate features from the tentative data to be processed; and obtaining the stroke test capability information of the candidate based on the candidate characteristic.

In one possible implementation, the interview capability prediction module 17012 is further configured to:

extracting initial interview representation information from the interview data to be processed;

obtaining first characterization information conforming to Gaussian distribution based on the first mean value and the first variance, and sampling from the first characterization information to obtain a first distribution latent variable;

In one possible implementation, the resume capability prediction submodule 17013 is configured to:

extracting initial resume representation information from the resume data to be processed;

In one possible implementation, the requirement matching module 1702 includes:

an overall capability submodule 17021 configured to obtain overall capability information based on the plurality of types of capability information and the attention weight;

a skill-specific submodule 17022 configured to obtain skill-specific information based on the overall capability information and the skill characterization information;

and a requirement matching submodule 17023 configured to obtain a predicted requirement matching result based on the specific skill information and the post skill requirement information.

For a description of specific functions and examples of each module and sub-module of the apparatus in the embodiment of the present disclosure, reference may be made to the description of corresponding steps in the foregoing method embodiments, and details are not repeated here.

In the related art, expert decision making includes: the expert (personnel department, etc.) makes a comprehensive decision on factors such as the stroke test results, the mastery skills in the resume, the project experience, the interview evaluation report, and the like of the candidate based on past experience. However, the expert decision has strong dependence on the manual experience, and has the problems of low decision efficiency, lack of interpretability, strong subjectivity of comprehensive decision and the like. Moreover, different experts may have different decision preferences and insufficient scalability due to different past experiences.

In the related art, the matching-based algorithm simply matches the resume information of the candidate and the skill requirement of the post using a natural language processing-based technique, and makes a decision based on the matching situation. The scheme can assist experts in making decisions, but only matches resumes and post information of candidates, multi-source evaluation data (written test and interview evaluation) of the candidates are not considered comprehensively, and certain interpretability is lacked.

The training method of the joint capacity model of the embodiment of the disclosure can be used in the joint capacity diagnosis process in talent recruitment, and can improve the capacity evaluation performance in talent recruitment by jointly modeling the multi-source heterogeneous evaluation result.

Application scenarios include, but are not limited to: and integrating multi-source evaluation results of the candidates to obtain comprehensive evaluation of the candidates, and selecting the most suitable candidate according to the skill requirements of different posts, so that the best matching between the posts and talents is realized, the efficiency of talent recruitment is improved, and talent loss caused by competition among the posts is reduced.

In the joint competency diagnosis process in talent recruitment, the competency of a candidate may be modeled based on the resume data, the written test data, and the interview data of the candidate, respectively. Such as resume-based capabilities, written-trial-based capabilities, interview-based capabilities, and the like. And intelligently integrating the multi-source heterogeneous evaluation result of the candidate by adopting an attention mechanism to obtain the comprehensive capability evaluation of the candidate. And restoring the expert decision process in the historical recruitment records by predicting the matching degree of the comprehensive abilities of the candidates and the skill requirements of the post (or called the capability requirements of the post). And the training and learning of the relevant parameters of each model can be realized in iteration by combining the prediction result and the real result, such as the actual job-entering result. Referring to fig. 19, the joint capacity diagnosis process in talent recruitment can specifically include the following steps.

The specific steps of each link are described in detail below with reference to the main mathematical symbols and descriptions in the embodiment of the present disclosure in table 1.

Table 1: mathematical symbols and explanations

Multi-source candidate capability modeling:

1. candidate capability modeling based on written test data

Suppose there are N candidates in the system, such as job seeker, M exercises (also called questions, exercises, etc.), K skills.

Wherein the candidate number u _i ∈R ^1×N Exercise number e _j ∈R ^1×M Skill characterization matrix h ^S ∈R ^K×d ,W ^U 、W ^diff 、W ^disc Respectively, are learnable embedded matrices. Candidate u _i Exercise e _j One-hot direction ofRespectively in the amount of

And

macroscopic characterization (vector) of candidates with respect to written examination

R ^1×d The superscript d of (d) represents the macroscopic capability dimension.

Characterization of candidates in a particular skill (vector) with respect to a written test

R ^1×K The superscript K of (a) represents the dimension of the skill.

Difficulty parameter (vector) of exercises

Discriminability parameter of exercises

Exercise correlation skill vector

The skill situation of the problem investigation is shown. For example, if problem e _j Investigating the skills s, then

Is 1, otherwise is 0.

Project reaction theory formula based on educational psychology:

wherein the symbols

Indicating the multiplication by the para position.

Referring to FIG. 20, a neural network FCL for interactions between candidates (e.g., students, job seekers, etc.) and problems in a stroke ability prediction model takes x as an input to the FCL and outputs x as

I.e. the probability that the prediction is correct.

Based on the stroke test answering data of the candidate, the capability parameters of the candidate and the parameter representation of the exercises are modeled, then the interactive formula of the candidate and the exercises is constructed according to the psychological project reaction theory idea, the neural network with strong fitting capability can be adopted to automatically learn the interaction between the candidate and the exercises, and the probability that the candidate correctly answers the exercises is output. And comparing the probability of correct answer exercises with the real answer results to calculate a loss function, and updating the parameters of the stroke ability prediction model. Loss function loss based on candidate ability modeling of written test ^E ：

Wherein,

is candidate u _i Answering exercise e _j The real situation of (a) is,

predicting candidate u for a model _i Answering problem e _j The case (1). Ability to learn each candidate based on written-to-try recording minimizing the loss function

2. Candidate capability modeling based on resume/interview data:

referring to FIG. 21, the input of the resume capability prediction model is resume r _i Characterization (vector)

Including skills extracted from the resume text. For example, if there is a skill s in the resume text, then

Is 1, otherwise is 0.

Will be provided with

Respectively passing through two fully-connected neural networks to obtain a mean value mu ^R Sum variance σ ^R Then from fitting into a Gaussian distribution N (μ) ^R ,(σ ^R ) ² ) In which the latent variables of the distribution, e.g. Gaussian random vectors, are sampled

Then, the distribution (vector) of the theme is obtained after softmax

Characterizing (matrix) t topic ^R And skill characterization (matrix) h ^S Cross multiplication and softmax passing to obtain a theme set (matrix) beta ^R Finally will

And beta ^R Multiplying to obtain a predicted value.

Distribution of subject matter (vector)

The probability corresponding to each topic represents the grasping condition under each topic (skill set).

Subject characterization (matrix)

Assume that there is K a skill on resume in the system ^R Each topic representing a set of more closely related skills words, such as convolution, machine learning, and the like. Each topic has a token vector, i.e., a topic-specific vector.

A specific vector representing a topic k.

Distribution of topics among skill words

Indicating the proficiency of each topic k at the corresponding skill,

h ^S a skill characterization matrix is represented.

Finally, the subject of the resume is distributed (vector)

With topic representation (matrix) t ^R Multiplication, the macroscopic ability (vector) of the candidate, e.g. job seeker, on the resume can be obtained:

interview modeling, topic distribution (vector) for interview evaluation based on principles similar to resume modeling

With topic representation (matrix) t ^A Multiplying to obtain the macroscopic ability (vector) of the candidate in interview evaluation:

after modeling the candidate capabilities based on the resume/interview data, the candidates' capabilities can be learned from their resumes/interviews based on their resume/interview data. Since topic models are known for interpreting hidden decision logic in efficient representations with high interpretability, it is here possible to mine the latent capability characterization embodied in the resume based on topic models. The loss function modeled based on the resume's candidate capabilities may be:

wherein,

is a resume r _i The distribution of the subject matter of (a),

is a potential variable of the distribution and is,

is a resume r _i Characterization of beta ^R Is a collection of topics of resumes.

The loss function modeled based on interview candidate capabilities may be:

wherein,

is a face test _i The distribution of the subject matter of (a),

is a potential variable of the distribution and is,

Representation-based prior distribution

And posterior distribution

The distance of (c).

The following is an example of a specific formula reasoning process of a loss function of the resume, and interviews are similar and are not described again:

(1) Loss function:

(2) The generation process comprises the following steps: one generative model may be defined for each resume using the above model,

the subscript s in the above formula represents the count value of the skill; k represents the number of knowledge points, and corresponds to the skill dimension K described above.

Wherein it is assumed that

Is from a gaussian random vector

The obtained product is obtained by the method in the step (1),

(3) And (3) reasoning process: inferring topic distribution probabilities for resumes using variational inference algorithms

The basic idea is to optimize the parameters

Therefore the KL distribution is used to approach the true posterior distribution. Here, the variation distribution may be defined as a gaussian distribution,

wherein,

f ₁ ,f ₂ is a parameter of the fully-connected neural network layer.

Resume R _i Is embedded in

f _r Are parameters of the neural network, see fig. 21.

Finally, can be paired with

And deducing each candidate u by minimizing the lower bound evidence _i Is/are as follows

The formula of the loss function of (a) is as follows,

(II) multi-source candidate capability integration: based on the multi-source ability of the candidate obtained in the step (I), the emphasis on different abilities of the candidate is learned respectively by adopting an attention mechanism, and the multi-source candidate ability is integrated by using a linear superposition function to obtain the comprehensive ability of the candidate.

(III) modeling the position skill requirement: and extracting the requirement condition of the post on the specific skill based on the information data of the post.

(IV) candidate-position matching:

an example of a modeling process for matching a candidate, such as a job seeker, to a post is as follows:

ability to model candidates in written, resume, interview data, respectively

Then, an attention mechanism is adopted to obtain the overall macroscopic ability (vector) h of the candidate _i ∈R ^1×d I.e. by

Wherein, a _E 、a _R 、a _A Is the attention weight.

After obtaining the macro ability of the candidate, the macro ability is compared with a skill characterization matrix h ^S Cross multiplication obtains a vector alpha of the mastery condition of the candidate in a specific skill _i ∈R ^1×K ，

α _i ＝sigmoid(h _i ×(h ^S ) ^T )

Skill requirement vector gamma for a post _p ∈R ^1×K : representing a position p _p The requirement situation (vector) for a specific skill, the formula is as follows,

wherein,

is at post p _p One-hot vector of, W ^J Is a learnable training matrix.

Finally, the mastery condition alpha of the job seeker in the specific skill is determined _i Skill requirement vector gamma with position _p Carrying out bit-alignment multiplication to predict the probability of job entry,

wherein alpha is _i,s Grasping condition vector γ representing skill s (s-th skill) _p,s A requirement vector representing a skill s for a position.

And (4) matching the comprehensive capability of the candidate obtained in the step (II) with the post skill requirement obtained in the step (III), judging whether the capability of the candidate is suitable for the post, and predicting whether the candidate successfully applies for the post based on the matching result. Candidate-post matching modeled loss function:

wherein, y _i Is candidate u _i Whether the application for the successful tag is true or not,

is a model prediction candidate u _i The probability of success is applied.

(V) calculating a loss function: and (4) comparing the prediction result obtained in the step (IV) with the actual recruitment condition, finding out all the situations of prediction errors, and calculating a loss function. For example, an example of an overall loss function:

loss＝α*loss ^E +β*loss ^R +γ*loss ^A +η*loss ^D

and (VI) updating parameters: and (5) updating the parameters of the whole model by using the loss function in the step (five). Performing combined training: and if the calculation result of the loss function in the step (V) shows that the parameters need to be updated, respectively adjusting the parameters of each model according to the loss function corresponding to each model. If the result of the computation of the loss function of (five) indicates that convergence is required, all models stop tuning.

The solution of the disclosed embodiments, after diagnosing the comprehensive ability of the candidate, may have the following specific applications.

(1) Candidate competency delineation:

the mastery condition of the candidate on a specific skill can be obtained through the mapping operation, and the weak/skilled skill point of the candidate is obtained.

After the new skill characterization is obtained, the potential characterization of the comprehensive ability of the candidate can be mapped to the skill, and the mastery condition of the candidate on the new skill can be obtained.

And obtaining the overall mastery condition of the candidate on the relevant skill set through a linear superposition function.

(2) Talent retrieval service:

the framework can sort the candidates according to the mastery condition of a certain skill combination, and then recommend K optimal candidates for a specific post according to the requirement of the post on the skills, so that talents lost due to competition in the same post can be effectively avoided, and the post is ensured to recruit the most appropriate candidates.

The scheme of the embodiment of the disclosure is applied to the talent recruitment scene and has at least one of the following characteristics:

the comprehensiveness of the product. The framework based on the scheme of the embodiment of the disclosure can provide a joint capacity diagnosis method in talent recruitment, compared with the traditional matching-based method, the data of candidates is considered more comprehensively, and the capacity evaluation performance in talent recruitment is improved by modeling the candidate capacity from different dimensions.

High efficiency. The framework based on the scheme of the embodiment of the disclosure can directly make a decision and also can assist experts in making a decision, and compared with the traditional manual decision method, the recruitment efficiency can be greatly improved.

Interpretability. The framework based on the scheme of the embodiment of the disclosure can depict the mastering condition of the candidate on a specific skill and the requirement condition of the post on the specific skill, has no black box and has good interpretability.

Objectivity. Based on the framework of the scheme of the embodiment of the disclosure, the comprehensive capacity of the candidate is modeled based on the multi-source data of the candidate, and compared with the traditional manual integration multi-source evaluation result of human experts, the subjective factors are reduced.

And (4) expandability. For different scenes and different candidate data, the framework based on the scheme of the embodiment of the disclosure can be applicable, and has strong expandability.

And (4) balance. The framework based on the scheme of the embodiment of the disclosure can sort the candidates according to the specific skill combination so as to ensure that talents recruited on posts are the most suitable, and can effectively avoid talent loss caused by competition on the same post.

In the technical scheme of the disclosure, the acquisition, storage, application and the like of the personal information of the related user all accord with the regulations of related laws and regulations, and do not violate the customs of public sequences.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

Fig. 22 illustrates a schematic block diagram of an example electronic device 2200 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate 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. 22, the apparatus 2200 includes a computing unit 2201, which can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 2202 or a computer program loaded from a storage unit 2208 into a Random Access Memory (RAM) 2203. In the RAM2203, various programs and data required for the operation of the device 2200 may also be stored. The computing unit 2201, ROM 2202, and RAM2203 are connected to each other via a bus 2204. An input/output (I/O) interface 2205 is also connected to bus 2204.

A number of components in the device 2200 are connected to the I/O interface 2205, including: an input unit 2206 such as a keyboard, a mouse, or the like; an output unit 2207 such as various types of displays, speakers, and the like; a storage unit 2208 such as a magnetic disk, an optical disk, or the like; and a communication unit 2209 such as a network card, modem, wireless communication transceiver, etc. The communication unit 2209 allows the device 2200 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

The computing unit 2201 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 2201 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 model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The computing unit 2201 performs the various methods and processes described above, such as a training method of a joint capability model or a capability requirement matching method. For example, in some embodiments, the training method or the capacity requirement matching method of the joint capacity model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 2208. In some embodiments, some or all of the computer programs may be loaded and/or installed onto device 2200 via ROM 2202 and/or communications unit 2209. When loaded into RAM2203 and executed by computing unit 2201, a computer program may perform one or more steps of the above-described method of training a joint capability model or method of capability requirement matching. Alternatively, in other embodiments, the computing unit 2201 may be configured in any other suitable manner (e.g., by means of firmware) to perform a training method or a capability requirement matching method of the joint capability model.

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 codes 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 codes, when executed by the processor or controller, cause the functions/operations 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 can 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 may be a cloud server, a server of a distributed system, or a server combining 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 executed in parallel or 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.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A method of training a joint competency model, the joint competency model comprising a multi-competency prediction model and a demand matching model, the method comprising:

processing the sample data by adopting the multiple capability prediction models to obtain multiple capability information;

training the demand matching model and the multiple capability prediction models based on the demand matching result to update parameters of the demand matching model and the multiple capability prediction models.

2. The method of claim 1, wherein processing the sample data using the plurality of capability prediction models to obtain a plurality of capability information comprises at least two of:

processing the sample stroke test data by adopting a stroke test capability prediction model to obtain stroke test capability information;

processing the sample interview data by adopting an interview capability prediction model to obtain interview capability information;

and processing the sample resume data by adopting a resume capability prediction model to obtain resume capability information.

3. The method of claim 2, wherein processing the sample written test data using the written test capability prediction model to obtain written test capability information comprises:

extracting candidate features from the sample written data;

and obtaining the stroke test capability information of the candidate based on the candidate characteristics.

4. The method according to claim 3, wherein the loss function of the stroke capability prediction model is constructed based on the probability of correct answer predicted by the stroke capability prediction model and the actual answer result; and the probability of correct answer is obtained by predicting the interaction information of the candidate and the question, which is obtained by the stroke ability prediction model based on the sample stroke data.

5. The method of claim 4, wherein the step of obtaining the interaction information of the candidate and the topic comprises:

extracting candidate features and topic features from the sample written data;

obtaining stroke trial capability information based on the candidate features;

obtaining question difficulty information and question distinguishing degree information based on the question features;

and obtaining the interaction information of the candidate and the question based on the stroke skill characterization information, the question difficulty information, the question distinguishability information and the question associated skill information.

6. The method of any one of claims 2 to 5, wherein processing the sample interview data using an interview ability prediction model to obtain interview ability information comprises:

interview representation information extracted from the sample interview data;

obtaining first representation information which accords with Gaussian distribution based on the first mean value and the first square difference, and sampling the first representation information to obtain a first distribution latent variable;

obtaining predicted interview representation information based on the first topic distribution information and the first topic set;

7. The method of claim 6, wherein a loss function of the interview capability prediction model is constructed based on the first topic distribution information, the first distribution latent variable, the predicted interview characterization information, and the first topic set.

8. The method of any one of claims 2 to 7, wherein processing the sample resume data using the resume capability prediction model to obtain resume capability information comprises:

resume characterization information extracted from the sample resume data;

obtaining second characterization information which accords with Gaussian distribution based on the second mean value and the second variance, and sampling from the second characterization information to obtain a second distribution latent variable;

9. The method of claim 8, wherein a loss function of the resume capability prediction model is constructed based on the second topic distribution information, the second distribution latent variable, the predicted resume characterization information, and the second topic set.

10. The method according to any one of claims 1 to 9, wherein processing the plurality of types of capability information and requirement information using the requirement matching model to obtain a requirement matching result comprises:

obtaining overall capability information based on the plurality of types of capability information and the attention weight;

obtaining specific skill information based on the overall ability information and the skill characterization information;

and obtaining a predicted requirement matching result based on the specific skill information and the post skill requirement information.

11. The method of claim 10, wherein the penalty function of the demand matching model is constructed based on predicted demand matching results and real demand matching results.

12. The method of any of claims 1-11, wherein training the demand matching model and the plurality of capacity prediction models based on the demand matching results to update parameters of the demand matching model and the plurality of capacity prediction models comprises:

obtaining an overall loss function based on the loss function of the demand matching model and the loss functions of the multiple capability prediction models;

and under the condition that updating is determined to be needed based on the overall loss function, updating the demand matching model by using the loss function of the demand matching model, and respectively updating the corresponding capacity prediction models by using the loss functions of the plurality of capacity prediction models.

13. The method of claim 12, wherein updating the demand matching model with a penalty function of the demand matching model comprises:

updating at least one of attention weight, overall skill characterization information, and post skill requirement information in the demand matching model based on a loss function of the demand matching model.

14. The method according to claim 12 or 13, wherein updating the corresponding capability prediction models with the loss functions of the plurality of capability prediction models, respectively, comprises at least two steps of:

updating at least one of the parameters of the full-link layer, the first theme representation information and the overall skill representation information of the interview capability prediction model based on the loss function of the interview capability prediction model;

15. A capacity requirement matching method, comprising:

processing various to-be-processed data of a candidate by adopting a plurality of capability prediction models to obtain a plurality of capability information of the candidate;

16. The method of claim 15, wherein processing the plurality of data to be processed of the candidate using a plurality of capability prediction models to obtain a plurality of capability information of the candidate comprises at least two steps of:

processing the to-be-processed stroke test data by adopting a stroke test capability prediction model to obtain stroke test capability information;

processing the interview data to be processed by adopting an interview capability prediction model to obtain interview capability information;

and processing the resume data to be processed by adopting a resume capability prediction model to obtain resume capability information.

17. The method of claim 16, wherein processing the written test data to be processed by using the written test capability prediction model to obtain written test capability information comprises:

extracting candidate features from the written data to be processed;

18. The method of claim 16 or 17, wherein processing the interview data to be processed by using the interview capability prediction model to obtain interview capability information comprises:

interview representation information extracted from the interview data to be processed;

and obtaining interviewing capability information based on the first theme distribution information and the first theme representation information.

19. The method according to any one of claims 16 to 18, wherein processing resume data to be processed by using a resume capability prediction model to obtain resume capability information comprises:

resume representation information extracted from the resume data to be processed;

20. The method according to any one of claims 15 to 19, wherein processing the plurality of types of capability information and requirement information using a requirement matching model to obtain a requirement matching result of the candidate comprises:

obtaining specific skill information based on the overall capability information and the skill characterization information;

21. An apparatus for training a joint capacity model, the joint capacity model including a multiple capacity prediction model and a demand matching model, the apparatus comprising:

22. The apparatus of claim 21, wherein the capability prediction module comprises at least two of:

the stroke test capability prediction sub-module is used for processing the sample stroke test data by adopting a stroke test capability prediction model to obtain stroke test capability information;

the interview capability prediction submodule is used for processing sample interview data by adopting an interview capability prediction model to obtain interview capability information;

and the resume capability prediction submodule is used for processing the sample resume data by adopting the resume capability prediction model to obtain resume capability information.

23. The apparatus of claim 22, wherein the triage capability prediction sub-module is configured to extract candidate features from the sample triage data; and obtaining the stroke test capability information of the candidate based on the candidate characteristics.

24. The apparatus of claim 23, wherein the loss function of the stroke capability prediction model is constructed based on the probability of correct answer predicted by the stroke capability prediction model and the actual answer result; and the probability of correct answer is obtained by predicting the interaction information of the candidate and the question, which is obtained by the stroke ability prediction model based on the sample stroke data.

25. The apparatus of claim 24, wherein the stroke capability prediction sub-module is configured to obtain interaction information of the candidate and the topic, and comprises:

extracting candidate features and topic features from the sample written data;

obtaining stroke trial capability information based on the candidate features;

obtaining topic difficulty information and topic distinguishing degree information based on the topic characteristics;

26. The apparatus of any one of claims 22 to 25, the interview capability prediction sub-module to:

interview representation information extracted from the sample interview data;

27. The apparatus of claim 26, wherein a loss function of the interview capability prediction model is constructed based on the first topic distribution information, the first distribution latent variable, the predicted interview characterization information, and the first topic set.

28. The apparatus of any of claims 22 to 27, the resume capability prediction sub-module to:

resume characterization information extracted from the sample resume data;

29. The apparatus of claim 28, wherein the loss function of the resume capability prediction model is constructed based on the second topic distribution information, the second distribution latent variable, the predicted resume characterization information, and the second topic set.

30. The apparatus of any of claims 21 to 29, wherein the demand matching module comprises:

the overall capacity submodule is used for obtaining overall capacity information based on the various kinds of capacity information and the attention weight;

the specific skill submodule is used for obtaining specific skill information based on the overall ability information and the skill characterization information;

and the requirement matching submodule is used for obtaining a predicted requirement matching result based on the specific skill information and the post skill requirement information.

31. The apparatus of claim 30, wherein the penalty function of the demand matching model is constructed based on predicted demand matching results and real demand matching results.

32. The apparatus of any of claims 21 to 31, wherein the training module comprises:

the overall loss submodule is used for obtaining an overall loss function based on the loss function of the demand matching model and the loss functions of the multiple capability prediction models;

and the updating submodule is used for updating the demand matching model by using the loss function of the demand matching model and respectively updating the corresponding capacity prediction models by using the loss functions of the multiple capacity prediction models under the condition that the updating is determined to be needed based on the overall loss function.

33. The apparatus of claim 32, wherein the update sub-module is configured to update the demand matching model with a penalty function of the demand matching model, comprising: updating at least one of attention weight, overall skill information, and post skill requirement information in the requirement matching model based on a loss function of the requirement matching model.

34. The apparatus according to claim 32 or 33, wherein the updating sub-module is configured to update the respective capacity prediction models with the loss functions of the plurality of capacity prediction models, respectively, and comprises at least two steps of:

35. A capability requirement matching apparatus comprising:

and the requirement matching module is used for processing the various kinds of capability information and the requirement information by adopting a requirement matching model to obtain a requirement matching result of the candidate, wherein the requirement information comprises capability information required to be met by a target scene.

36. The apparatus of claim 35, wherein the capability prediction module comprises at least two of:

the stroke test capability prediction submodule is used for processing the stroke test data to be processed by adopting a stroke test capability prediction model to obtain stroke test capability information;

the interview capability prediction submodule is used for processing interview data to be processed by adopting an interview capability prediction model to obtain interview capability information;

and the resume capability prediction submodule is used for processing the resume data to be processed by adopting the resume capability prediction model to obtain resume capability information.

37. The apparatus of claim 36, wherein the written ability prediction sub-module is configured to extract candidate features from the written data to be processed; and obtaining the stroke test capability information of the candidate based on the candidate characteristics.

38. The apparatus of claim 36 or 37, the interview capability prediction sub-module to:

39. The apparatus of any of claims 36 to 38, the resume capability prediction sub-module to:

40. The apparatus of any of claims 35 to 39, wherein the demand matching module comprises:

41. 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 1-20.

42. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any of claims 1-20.

43. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-20.