CN111062599B - Human resource scheduling model training and scheduling method and device based on personnel relationship - Google Patents

Human resource scheduling model training and scheduling method and device based on personnel relationship Download PDF

Info

Publication number
CN111062599B
CN111062599B CN201911251878.1A CN201911251878A CN111062599B CN 111062599 B CN111062599 B CN 111062599B CN 201911251878 A CN201911251878 A CN 201911251878A CN 111062599 B CN111062599 B CN 111062599B
Authority
CN
China
Prior art keywords
self
encoder
layer
personnel
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201911251878.1A
Other languages
Chinese (zh)
Other versions
CN111062599A (en
Inventor
谢莹
许荣斌
林元模
汪欣梅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Anhui University
Putian University
Original Assignee
Anhui University
Putian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Anhui University, Putian University filed Critical Anhui University
Priority to CN201911251878.1A priority Critical patent/CN111062599B/en
Publication of CN111062599A publication Critical patent/CN111062599A/en
Application granted granted Critical
Publication of CN111062599B publication Critical patent/CN111062599B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06311—Scheduling, planning or task assignment for a person or group
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling

Abstract

The embodiment of the invention provides a human resource scheduling model training and scheduling method and device based on personnel relationship, wherein the training method comprises the following steps: constructing a human resource network by taking each person in work as a node and taking the character attribute and the working position corresponding to the node as edges among the nodes; acquiring a relation matrix of the human resource network according to an adjacent matrix corresponding to the human resource network and the degree of the personnel in the position; respectively training at least two self-encoders by using the relation matrix training until the loss function of each self-encoder is converged; and stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of the self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder. The model trained by the embodiment of the invention can be used for obtaining more accurate results when the personnel to be scheduled are scheduled.

Description

Human resource scheduling model training and scheduling method and device based on personnel relationship
Technical Field
The invention relates to the field of human resource scheduling, in particular to a human resource scheduling model training and scheduling method and device based on personnel relationship.
Background
In the background of the times of economic globalization and knowledge economy, human resources are important factors influencing the survival and development of an enterprise. The configuration of human resources, i.e. the matching degree between personnel and positions, directly affects the reasonable utilization and the overall configuration efficiency of other resources of an enterprise. Reasonable human resource allocation is a key factor for determining whether an enterprise can develop continuously, stably and rapidly. Therefore, how to achieve reasonable human resource allocation is an urgent technical problem to be solved. The commonly used human resource allocation methods include multiple linear regression analysis, analytic hierarchy process analysis, delphire analysis, and the like, which all belong to linear analysis methods. However, the proper allocation of human resources is influenced by many factors, such as scale, economic strength, culture, and the external environment of the enterprise. At the same time, these factors are mostly non-linear. Therefore, the traditional linear analysis method has the defects of strong subjectivity, low accuracy and weak capability of reflecting actual conditions.
At present, the thesis 'Wangbexin, Liuyi and the like', a human resource demand prediction method [ J ] based on a gray BP neural network model, statistics and prediction, 2018 ', the' Zhongwangbexin and the like analyze the matching relationship between human resource management practice and organization conditions based on the BP neural network model, establish a prediction model and give a good prediction effect; the literature studies a large number of influencing factors related to human resources; the literature describes human resources through attributes such as skills, salaries and contribution degree to projects, so that scheduling constraints are established and the human resources are scheduled by using a genetic algorithm.
However, the inventor finds that the influence of the similarity and the specific relation among the persons in the human resource allocation process on the human resource allocation is not considered in the prior art, and therefore, the technical problem that the human resource allocation effect is inaccurate exists in the prior art.
Disclosure of Invention
The technical problem to be solved by the invention is how to provide a human resource scheduling model training and scheduling method and device based on personnel relationship to improve the accuracy of human resource allocation.
The invention solves the technical problems through the following technical means:
the embodiment of the invention provides a human resource scheduling model training method based on personnel relationship, which comprises the following steps:
the method comprises the following steps of taking each person in work as a node, taking character attributes corresponding to the node and a working post as edges among the nodes to construct a human resource network, wherein the character attributes comprise: proficient in one or a combination of office software, strong logical thinking, team cooperation, personality, sex, personnel relationship, graduation colleges and specialties;
acquiring a relation matrix of the human resource network according to the adjacent matrix corresponding to the human resource network and the degree of the personnel in the position;
respectively training at least two self-encoders by using the relation matrix training until the loss function of each self-encoder is converged;
and stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of the self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder.
In actual work, the personnel who are at the job who are suitable for the same job position often have certain similarity, and by applying the embodiment of the invention, the personnel and the job position are used as nodes, and the human resource network is constructed by using the attributes of the personnel at the job position as edges, so that the characteristic reconstruction of the job position is favorably carried out by using a neural network model based on the commonality among the existing personnel, the abstract description result of the job position can be obtained, and further, a more accurate result can be obtained when the model trained by the embodiment of the invention is used for the job positioning of the personnel at the job position.
Optionally, the target model comprises four stacked self-encoders, wherein,
the input layer of the first layer self-encoder is used for receiving the relation matrix, and the output of the hidden layer of the first layer self-encoder is used as the input of the input layer of the second layer self-encoder;
the output of the hidden layer of the second layer self-encoder is used as the output of the input layer of the third layer self-encoder;
the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder;
the fourth layer is output from the hidden layer of the encoder as the output of the stacked self-encoder.
Optionally, the
Training at least two self-encoders respectively by using the relation matrix training until the loss function of each self-encoder converges, comprising:
using formulas
Figure BDA0002309268880000031
The loss from the encoder in the current iteration is calculated, wherein,
COST is the loss of the self-encoder in the current iteration;
Figure BDA0002309268880000033
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m is a unit ofiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; o. oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; w is a group ofOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000032
hiOutputting the corresponding coding layer for the ith personnel,
Figure BDA0002309268880000041
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a trained self-encoder;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
Optionally, the
Training at least two self-encoders respectively by using the relation matrix training until the loss function of each self-encoder converges, comprising:
using the formula J-COST + λ Tr (HLH)T) The loss from the encoder in the current iteration is calculated, wherein,
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation and control reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure BDA0002309268880000042
sijThe fitness values of the personnel i and the personnel j to the posts are obtained; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure BDA0002309268880000043
a matrix of real numbers N x N;
Figure BDA0002309268880000044
Figure BDA0002309268880000045
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; w is a group ofOIs the weight of the decoding layer; dOBias for decoding layers; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000051
hiOutputting the corresponding coding layer for the ith personnel,
Figure BDA0002309268880000052
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a trained self-encoder;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
The embodiment of the invention also provides a human resource scheduling method based on the personnel relationship, which comprises the following steps:
acquiring personnel attributes of personnel to be subjected to post setting, and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
and inputting the attribute matrix into the target model to obtain the post information corresponding to the personnel to be positioned, wherein the target model is a human resource scheduling model based on any training and personnel relationship.
The embodiment of the invention provides a human resource scheduling model training device based on personnel relationship, which comprises:
the construction module is used for constructing a human resource network by taking each person in work as a node and taking the character attribute and the working position corresponding to the node as edges among the nodes, wherein the personnel attribute comprises the following steps: proficient in one or a combination of office software, strong logical thinking, team cooperation, personality, sex, personnel relationship, graduation colleges and specialties;
the first acquisition module is used for acquiring a relation matrix of the human resource network according to the adjacent matrix corresponding to the human resource network and the degree of the personnel in the position;
the training module is used for training at least two self-encoders by using the relation matrix respectively until the loss function of each self-encoder is converged;
and the stacking module is used for stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of the self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder.
Optionally, the target model comprises four stacked self-encoders, wherein,
the input layer of the first layer self-encoder is used for receiving the relation matrix, and the output of the hidden layer of the first layer self-encoder is used as the input of the input layer of the second layer self-encoder;
the output of the hidden layer of the second layer self-encoder is used as the output of the input layer of the third layer self-encoder;
the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder;
the hidden layer output of the fourth layer self-encoder is taken as the output of the stacked self-encoder.
Optionally, the training module is configured to:
using formulas
Figure BDA0002309268880000061
The loss from the encoder in the current iteration is calculated, wherein,
COST is the loss of the self-encoder in the current iteration;
Figure BDA0002309268880000063
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staff memberiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000062
hiOutputting the corresponding coding layer for the ith personnel,
Figure BDA0002309268880000071
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
Optionally, the training module is configured to:
using the formula J-COST + λ Tr (HLH)T) The loss from the encoder in the current iteration is calculated, wherein,
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation and control reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure BDA0002309268880000072
sijThe fitness values of the personnel i and the personnel j to the posts are obtained; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure BDA0002309268880000073
a matrix of real numbers N x N;
Figure BDA0002309268880000074
Figure BDA0002309268880000075
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; o. oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAre multiplied byWHIs the weight of the coding layer; d is a radical ofHA bias for the coding layer; wOWeight of decoding layer; d is a radical ofOBias for decoding layers; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of neurons in the hidden layer, and
Figure BDA0002309268880000076
hioutputting the coding layer corresponding to the ith personnel,
Figure BDA0002309268880000081
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
The embodiment of the invention also provides a human resource scheduling device based on the personnel relationship, which comprises:
the second acquisition module is used for acquiring the personnel attributes of the personnel to be subjected to post setting and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
and the input module is used for inputting the attribute matrix into the target model to obtain the position information corresponding to the personnel to be positioned, wherein the target model is a human resource scheduling model based on the personnel relationship and trained on any one of the training modes.
The invention has the advantages that:
in practice, the personnel who are at the job who are suitable for the same job position often have certain similarity, and by applying the embodiment of the invention, the personnel who are at the job position and the job position are used as nodes, and the human resource network is constructed by using the attributes of the personnel who are at the job position as edges, so that the characteristic reconstruction of the job position is favorably carried out by using a neural network model based on the commonality among the existing personnel, the abstract description result of the job position can be obtained, and further, a more accurate result can be obtained when the model trained by the embodiment of the invention is used for carrying out the job positioning of the personnel who are at the job position.
Drawings
FIG. 1 is a schematic flow chart of a human resource scheduling model training method based on human relationships according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a stack self-encoder in the human resource scheduling model training method based on human relationships according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a comparison of training results of post fitness constraints on hidden layers of a stacked self-encoder according to an embodiment of the present invention;
fig. 4 is a diagram illustrating sensitivity results of loss functions of different λ in different numbers of hidden layers according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a human resource scheduling model training apparatus based on a human relationship according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
Fig. 1 is a schematic flowchart of a human resource scheduling model training method based on human relationships according to an embodiment of the present invention, and as shown in fig. 1, the method includes:
s101: the method comprises the following steps of taking each person in work as a node, taking character attributes corresponding to the node and a working post as edges among the nodes to construct a human resource network, wherein the character attributes comprise: proficient in office software, strong logical thinking, team collaboration, personality, gender, personnel relationship, graduation colleges and specialties, or a combination thereof.
In the human resource allocation process, each person has its specific attributes and outstanding working advantages, for example, if a person has the attribute X { "smart Java", "logical thinking strong" … … "team collaboration" }, then the person is more suitable for the Java development station, and the attributes needed by the persons suitable for the same station should be similar. A particular human resources network can be constructed based on the attributes each person has.
Given N job seekers, a human resources network G ═ (V, E, Y) may be formed, where V ═ V (V, E, Y)1,v2,...,vN) Representing N persons who are on duty for the set of nodes; e ═ EijThe set of edges between two nodes in the tree. Y is the class mark of the person, i.e. the matched position, Y ═ YL,...,YN-LIn which the class label Y of the staff with allocated positionLA＝{Y1,...,YLH, obviously, the personnel class label Y of the unallocated positionUL＝{YL+1,...,YN}。
S102: and acquiring a relation matrix of the human resource network according to the adjacent matrix corresponding to the human resource network and the degree of the personnel in the position.
Illustratively, in the human resources network G, the contiguous matrix of the human resources network is a non-negative symmetric matrix
Figure BDA0002309268880000101
If there is an edge connection between nodes i and j, it indicates that there is a connection between person i and person j: for example, where there is an intersection, similarity, a, in the work of employee i and employee j ij1 is ═ 1; otherwise aij0, i.ltoreq.N for all 1. ltoreq.i, aii0; r is a real number matrix.
Using symmetric matrices in embodiments of the invention
Figure BDA0002309268880000102
In network of drawing human resourcesSimilarity of persons when a ∈ RNIn time, there are:
Figure BDA0002309268880000103
wherein wijSimilarity between person i and person j; a isikIs a contiguous matrix vector aiThe kth element of (1); i.e. the ith row and kth column elements in the person similarity relationship matrix, refer to the similarity relationship between the ith person and the kth person. σ is a scale control parameter of each dimension, and the value can be a proper value which is predetermined in an experiment according to an experiment result; exp is an exponential function with a natural constant as the base.
Further, in consideration of the relationship between persons suitable for the same station, the person relationship matrix M ═ M is usedij]∈RN ×NEvaluating the relationship of the persons who are present on the same post, wherein,
Figure BDA0002309268880000111
wherein m isijIs the relationship between person i and person j; k is a radical ofiIs the degree of person i; k is a radical of formulajIn degrees of person j.
From the above, it can be seen that the M matrix is the remachining of the W matrix, and it is hoped to better express the similarity relationship between people. M is calculated based on the element values within k and W. Wherein k isiThe degree of the ith person is represented by the value of the ith person, the value of the degree of the ith person refers to the value accumulation of the ith row in the W matrix, and the physical meaning of the representation is how many edges of the ith point in the network are connected with the ith point. If the degree is large, the number of edges connected with the node is large, and the importance of the node is high.
S103: and respectively training at least two self-encoders by using the relation matrix training until the loss function of each self-encoder is converged.
Illustratively, the first self-encoder training process is described as an example, and the original self-encoder is reconstructed for a deep self-encoder networkObtaining a hidden layer representation of the data personnel relation matrix B
Figure BDA0002309268880000112
H is to be1As input to the next self-encoder, and so on. The 4 th self-encoder is then trained, and a new data representation can be obtained by reconstructing the output of the 3 rd self-encoder
Figure BDA0002309268880000113
Taking the self-encoder of the first layer as an example for explanation, in the embodiment of the present invention, the personnel relation matrix M is used as the input of the self-encoder, and the personnel relation matrix M is mapped to a low-dimensional embedding after encoding
Figure BDA0002309268880000121
In which d is<N，
hi＝f(mi)＝s(WHmi+dH)
Wherein h isiSelf-encoding an encoded output therein for the first layer; m isiThe element in the corresponding relation matrix of the ith staff; n is the number of the staff; and WHIs the weight of the coding layer; dHIs the bias of the coding layer; s (-) is a non-linear mapping function, i.e. Softmax function
Figure BDA0002309268880000122
Figure BDA0002309268880000123
The hidden layer representation H is reflected to the original data space through decoding, so that the effect of reconstructing the original data is achieved:
oi＝g(hi)＝s(WOhi+dO)
wherein o isiCorresponding to m for the self-encoder in the ith staffiThe decoded output of (1); wOIs a weight of a decoding layer, and WO∈RN*d；dOIs a decoding layerAnd d is offset fromO∈RN*1. The self-encoder mainly learns the parameter theta ═ WH，dH，WO，dOUnder the condition, original data are reconstructed to minimize the error between the original data M and the reconstructed data O, and a low-dimensional nonlinear representation H is obtained.
Thus, the loss function from the encoder may be:
Figure BDA0002309268880000124
wherein, the first and the second end of the pipe are connected with each other,
cost is the loss of the self-encoder in the current iteration;
Figure BDA0002309268880000125
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational relationship matrix; n is the number of the staff; sigma is a summation function; o. oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); | | non-woven hair2Is a 2 norm function; θ ═ WH,dH,WO,dO}。
Further, in order to improve the working performance of the neurons, the embodiment of the present invention applies a constraint condition to the self-encoder neural network, that is, a sparsity constraint is added to each hidden layer, and the new sparsity constraint effectively extracts the information of the network structure:
Figure BDA0002309268880000131
wherein, the first and the second end of the pipe are connected with each other,
sigma is a summation function; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000132
j is the jth neuron of the hidden layer; log is a logarithmic function with base 10.
Thus, the loss from the encoder per iteration can be found:
Figure BDA0002309268880000133
and (3) calculating, wherein,
wherein COST is a loss function obtained in the current iteration; wHIs the weight of the coding layer; dHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000134
hiAnd outputting the corresponding coding layer for the ith personnel.
The personnel relation matrix is used as input and is input into the self-encoder,
using formulas
Figure BDA0002309268880000135
Calculating the loss of the self-encoder in the current iteration;
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
The self-encoder is then optimized using a back-propagation algorithm with a random gradient descent, for each point, whose derivative represents the instantaneous rate of change at that point, i.e. the slope at that point, representing the fastest gradient descending from that point to the other points. The back propagation algorithm calculates the error of the last output result of the network and reversely and gradually propagates the error. The directional propagation uses the basic idea of chain derivation (implicit function derivation). Error of calculationAnd then, the error is required to be reversely propagated back, the error of the last layer is firstly calculated, then the parameter of the previous node is updated, and gradient descent is adopted to sequentially update layer by layer until final convergence is known. During each iteration, the parameter theta is set to { W ═ WH，dH，WO，dOAnd updating and continuously training the obtained network, thereby effectively realizing personnel post matching.
Figure BDA0002309268880000141
Figure BDA0002309268880000142
Figure BDA0002309268880000143
Figure BDA0002309268880000144
And then, returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
Further, in order to better implement scheduling of human resources, the embodiment of the present invention introduces the degree that the person i and the person j are suitable for the same work station into the self-encoder as prior information, for example, if both the person i and the person j are suitable for the same work station, then the person i and the person j should have the characteristic of similar attributes, and further, a station fitness matrix can be defined
Figure BDA0002309268880000145
To represent the fitness of the person i and the person j to a certain position, and the threshold range is [0.1, 1%]. Meanwhile, Euclidean distance is adopted to measure the similarity of hidden layers before and after training, namely D (h)i，hj). So the post fitness constraint can be written as:
Figure BDA0002309268880000151
where Tr (,) is the trace of the matrix, the diagonal matrix
Figure BDA0002309268880000152
Is the summation per row of the matrix S,
Figure BDA0002309268880000153
hjand outputting the corresponding coding layer for the jth personnel.
Thus, a new loss function will be obtained as:
J＝COST+λTr(HLHT) Wherein, in the step (A),
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure BDA0002309268880000154
sijThe fitness values of the personnel i and the personnel j to the posts are obtained; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure BDA0002309268880000155
a matrix of real numbers N x N;
Figure BDA0002309268880000156
Figure BDA0002309268880000157
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d);
θ＝{WH,dH,WO,dOand W isHIs the weight of the coding layer; d is a radical ofHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000158
hiOutputting the coded layer corresponding to the ith personnel;
Figure BDA0002309268880000159
furthermore, in model training, the formula J ═ COST + λ Tr (HLH) can be usedT) Calculating the loss of the self-encoder in the current iteration;
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dO}。
S104: and stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of the self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder.
In order to better implement human resource scheduling and improve the matching degree between personnel and posts, fig. 2 is a schematic structural diagram of a stack self-encoder in a human resource scheduling model training method based on personnel relationship provided in an embodiment of the present invention, as shown in fig. 2, an embodiment of the present invention discloses a new deep network structure, a stack self-encoder is formed by stacking four self-encoders, as shown in fig. 2, an input layer of a first layer self-encoder is used for receiving a relationship matrix, and an output of a hidden layer of the first layer self-encoder is used as an input of an input layer of a second layer self-encoder; the output of the hidden layer of the second layer self-encoder is used as the output of the input layer of the third layer self-encoder; the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder; the hidden layer output of the fourth layer self-encoder is taken as the output of the stacked self-encoder.
In order to explain the effect of the embodiment of the present invention, a target model after training of the present invention is taken as an example for explanation.
The structure of the trained target model is as follows:
the stacked autoencoder network includes four autoencoders. The layer configuration of the experimental data is set to [ 2000-; the second self-encoder is 512-256-512; the third self-encoder is 256-128-256; the fourth self-encoder is 128-64-128.
Then, randomly selecting different numbers of personnel data in the training set, and respectively carrying out cross testing on the human resource demand prediction method [ J ] of Wangben, Liuyi and the like based on a gray BP neural network model, statistics and prediction, 2018, the disclosed BP neural network algorithm, Yang L, Cao X, He D, et al, modulated based community detection with deep detection [ C ]// International Joint Conference on scientific understanding apparatus associated with AAAI Press,2016: 2252-.
The Accuracy Accuracy is adopted as the Accuracy of the matching degree of the human resources, namely:
Figure BDA0002309268880000171
wherein the content of the first and second substances,
TP represents the number of correctly divided positive cases, FP represents the number of incorrectly divided positive cases, TN represents the number of correctly divided negative cases, and FN represents the number of incorrectly divided negative cases.
Table 1 shows the comparison result of the human resource scheduling accuracy obtained in the embodiment of the present invention with respect to the prior art,
TABLE 1
Figure BDA0002309268880000172
Figure BDA0002309268880000181
As shown in table 1, under the condition of different numbers of people, the new human resource depth nonlinear reconstruction algorithm proposed by the embodiment of the present invention is superior to the DNR (DNR _ L2) that applies the L2 paradigm to the error loss function and the DNR (DNR _ CE) that applies sigmoid-cross entropy to the error function, and achieves a better prediction effect. The new algorithm not only fully considers the related attributes among the personnel, but also improves the working efficiency and avoids the complexity of manually matching the personnel positions by adopting the position fitness constraint prior information through personnel similarity processing. Therefore, in practical application, the working efficiency of a human resource department is greatly improved, and remarkable results are obtained.
In order to evaluate the improvement effect of the post fitness matrix on the algorithm of the embodiment of the invention, 2000 experimental data are respectively trained in a stack self-encoder to evaluate four hidden layers (one hidden layer, two hidden layers, three hidden layers and four hidden layers).
Fig. 3 is a diagram illustrating a comparison of training results of the post fitness constraint on the hidden layer of the stacked self-encoder according to an embodiment of the present invention, as shown in fig. 3. Compared with the evaluation of the four hidden layers, the method can show that the four hidden layers trained by the stack self-encoder can better embody the remarkable characteristics of people, thereby achieving higher human resource allocation prediction capability. Therefore, the deep neural network-based human resource deep nonlinear reconstruction model provided by the embodiment of the invention can obtain a better personnel position matching effect, effectively and reasonably allocate human resources of automobile manufacturing enterprises, and realize optimization of the human resources.
The enterprise human resource allocation should satisfy a reasonable match of people to a particular post. In the allocation process, people suitable for the same post often have certain similarity, in order to better predict the most suitable post of the staff not allocated with the post, different attributes of the people are utilized to construct a human resource network based on a personnel relation module, and a novel human resource deep nonlinear reconstruction model is provided for human resource scheduling, so that a human resource allocation task can be completed more reasonably.
In addition, the embodiment of the invention uses lambda as a parameter between the regulation and control reconstruction error and the constraint, and different cost functions, namely final values of the loss functions, can be obtained under different lambda values. Fig. 4 is a schematic diagram of the sensitivity result of the loss function of different λ in different numbers of hidden layers provided in the embodiment of the present invention, as shown in fig. 4, when the new method provided in the embodiment of the present invention adopts four hidden layers, the cost function can be stabilized at a lower value, which also means the best effect in the present work.
The embodiment of the invention provides the method and the device which fully consider the characteristics of the personnel and explore the effective factors such as the similarity and the personnel relationship among the personnel. The relation module based on the personnel carries out personnel post matching work of semi-supervised learning, and the prior information of post fitness constraint of related personnel is combined, so that the overall prediction capability is greatly improved. The algorithm was tested on an automobile manufacturing enterprise dataset, different numbers of samples were randomly taken from the dataset for further testing, and the impact of the post fitness constraint on the algorithm was tested. Experimental results show that the new model provided by the embodiment of the invention achieves remarkable results in personnel post matching, so that the working efficiency of human resource allocation is effectively improved.
Furthermore, the output of the last hidden layer in the stack self-encoder can be clustered by adopting a k-means clustering algorithm, so that the output result of the target model has better representation capability, and a predicted post class mark with better representation effect can be obtained.
In addition, the embodiment of the invention can also assist a decision maker to reasonably arrange working personnel and improve the judging and decision-making capability.
In practice, the personnel who are at the job who are suitable for the same job position often have certain similarity, and by applying the embodiment of the invention, the personnel who are at the job position and the job position are used as nodes, and the human resource network is constructed by using the attributes of the personnel who are at the job position as edges, so that the characteristic reconstruction of the job position is favorably carried out by using a neural network model based on the commonality among the existing personnel, the abstract description result of the job position can be obtained, and further, a more accurate result can be obtained when the model trained by the embodiment of the invention is used for carrying out the job positioning of the personnel who are at the job position.
Example 2
Based on the method of the embodiment 1, the embodiment 2 of the present invention further provides a human resource scheduling method based on the personnel relationship, where the method includes:
acquiring personnel attributes of personnel to be subjected to post setting, and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
and inputting the attribute matrix into the target model to obtain the position information corresponding to the personnel to be positioned, wherein the target model is a human resource scheduling model based on personnel relationship and trained according to any one of claims 1 to 4.
It should be noted that the specific process of the embodiment of the present invention has been disclosed in embodiment 1, and details of the embodiment of the present invention are not described herein again.
By applying the embodiment of the invention, more accurate results can be obtained when the personnel waiting to be post fix the post.
Example 3
Corresponding to the embodiment 1 of the invention, the embodiment of the invention also provides a human resource scheduling model training device based on the personnel relationship.
Fig. 5 is a schematic structural diagram of a human resource scheduling model training apparatus based on human relationships according to an embodiment of the present invention, and as shown in fig. 5, the apparatus includes:
a building module 501, configured to build a human resource network by using each person who is on duty as a node and using a person attribute and a work position corresponding to the node as edges between the nodes, where the person attribute includes: proficient in one or a combination of office software, strong logical thinking, team cooperation, personality, sex, personnel relationship, graduation colleges and specialties;
a first obtaining module 502, configured to obtain a relationship matrix of the human resource network according to an adjacency matrix corresponding to the human resource network and a degree of an incumbent;
a training module 503, configured to train at least two autoencoders using the relationship matrix until a loss function of each autoencoder converges;
a stacking module 504, configured to perform stacking processing on the trained self-encoders to obtain a target model, where an output of a hidden layer of each of the self-encoders in the target model except for the first self-encoder is used as an input of a next self-encoder.
In practice, the personnel who are at the job who are suitable for the same job position often have certain similarity, and by applying the embodiment of the invention, the personnel who are at the job position and the job position are used as nodes, and the human resource network is constructed by using the attributes of the personnel who are at the job position as edges, so that the characteristic reconstruction of the job position is favorably carried out by using a neural network model based on the commonality among the existing personnel, the abstract description result of the job position can be obtained, and further, a more accurate result can be obtained when the model trained by the embodiment of the invention is used for carrying out the job positioning of the personnel who are at the job position.
In one embodiment of the present invention, the target model includes four stacked self-encoders, wherein,
the input layer of the first layer self-encoder is used for receiving the relation matrix, and the output of the hidden layer of the first layer self-encoder is used as the input of the input layer of the second layer self-encoder;
the output of the hidden layer of the second layer self-encoder is used as the output of the input layer of the third layer self-encoder;
the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder;
the hidden layer output of the fourth layer self-encoder is taken as the output of the stacked self-encoder.
In a specific implementation manner of the embodiment of the present invention, the training module 503 is configured to:
the training module is configured to:
using formulas
Figure BDA0002309268880000221
The loss from the encoder in the current iteration is calculated, wherein,
COST is the loss of the self-encoder in the current iteration;
Figure BDA0002309268880000222
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; d is a radical ofHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000223
hiOutputting the corresponding coding layer for the ith personnel,
Figure BDA0002309268880000224
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, useInverse propagation algorithm for random gradient descent updates theta ═ WH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
In a specific implementation manner of the embodiment of the present invention, the training module 503 is configured to:
using the formula J-COST + λ Tr (HLH)T) The loss from the encoder in the current iteration is calculated, wherein,
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation and control reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure BDA0002309268880000231
sijThe fitness values of the personnel i and the personnel j to the posts are obtained; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure BDA0002309268880000232
a matrix of real numbers N x N;
Figure BDA0002309268880000233
Figure BDA0002309268880000234
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOH, and WHIs the weight of the coding layer; d is a radical ofHIs the bias of the coding layer; wOIs the weight of the decoding layer; d is a radical ofOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure BDA0002309268880000235
hiOutputting the corresponding coding layer for the ith personnel,
Figure BDA0002309268880000236
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
Example 4
Corresponding to embodiment 2 of the present invention, an embodiment of the present invention further provides a human resource scheduling apparatus based on a human relationship, where the apparatus includes:
the second acquisition module is used for acquiring the personnel attributes of the personnel to be subjected to post setting and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
and the input module is used for inputting the attribute matrix into the target model to obtain the position information corresponding to the personnel to be positioned, wherein the target model is the human resource scheduling model based on the personnel relationship and trained in the embodiment 1.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (8)

1. The human resource scheduling model training method based on the personnel relationship is characterized by comprising the following steps:
the method comprises the following steps of taking each person in work as a node, and taking the person attribute and the working post corresponding to the node as edges among the nodes to construct a human resource network, wherein the person attribute comprises the following steps: proficient in one or a combination of office software, strong logical thinking, team cooperation, personality, sex, personnel relationship, graduation colleges and specialties;
acquiring a relation matrix of the human resource network according to the adjacent matrix corresponding to the human resource network and the degree of the personnel in the position;
respectively training at least two self-encoders by using the relation matrix training until the loss function of each self-encoder is converged;
stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of other self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder;
wherein the training of at least two autoencoders using the relationship matrix until the loss function of each autoencoder converges comprises:
using formulas
Figure FDA0003612371380000011
Calculating the loss of the self-encoder in the current iteration, wherein COST is the loss of the self-encoder in the current iteration;
Figure FDA0003612371380000012
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; ρ isA sparsity parameter; KL is KL divergence; rhojIs the average activity of hidden layer neurons, an
Figure FDA0003612371380000021
hiOutputting the corresponding coding layer for the ith personnel,
Figure FDA0003612371380000022
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a trained self-encoder;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
2. The human resources scheduling model training method based on human relationships of claim 1, wherein the target model comprises four stacked self-encoders, wherein,
the input layer of the first layer self-encoder is used for receiving the relation matrix, and the output of the hidden layer of the first layer self-encoder is used as the input of the input layer of the second layer self-encoder;
the output of the hidden layer of the second layer self-encoder is used as the input of the input layer of the third layer self-encoder;
the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder;
the hidden layer output of the fourth layer self-encoder is taken as the output of the stacked self-encoder.
3. The human resource scheduling model training method based on human relationship as claimed in claim 1, wherein the training using the relationship matrix respectively trains at least two auto-encoders until the loss function of each auto-encoder converges comprises:
using the formula J-COST + λ Tr (HLH)T) The loss from the encoder in the current iteration is calculated, wherein,
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation and control reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure FDA0003612371380000023
sijThe fitness values of the personnel i and the personnel j to the posts are obtained; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure FDA0003612371380000031
a matrix of real numbers N x N;
Figure FDA0003612371380000032
Figure FDA0003612371380000033
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure FDA0003612371380000034
hiOutputting the corresponding coding layer for the ith personnel,
Figure FDA0003612371380000035
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a trained self-encoder;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
4. The human resource scheduling method based on the personnel relationship is characterized by comprising the following steps:
acquiring personnel attributes of personnel to be subjected to post setting, and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
and inputting the attribute matrix into the target model to obtain the position information corresponding to the personnel to be positioned, wherein the target model is obtained based on the human resource scheduling model training method based on the personnel relationship in any one of claims 1-3.
5. Human resource scheduling model training device based on personnel relation, its characterized in that, the device includes:
the construction module is used for constructing the human resource network by taking each person in work as a node and taking the person attribute and the working post corresponding to the node as edges among the nodes, wherein the person attribute comprises the following steps: one or a combination of official software, strong logical thinking, team cooperation, character, sex, personnel relationship, graduation colleges and major;
the first acquisition module is used for acquiring a relation matrix of the human resource network according to the adjacent matrix corresponding to the human resource network and the degree of the personnel in the position;
the training module is used for training at least two self-encoders by using the relation matrix respectively until the loss function of each self-encoder is converged;
the stacking module is used for stacking the trained self-encoders to obtain a target model, wherein the output of the hidden layers of other self-encoders except the first self-encoder in the target model is used as the input of the next self-encoder;
wherein the training module is configured to:
using formulas
Figure FDA0003612371380000041
The loss from the encoder in the current iteration is calculated, wherein,
COST is the loss of the self-encoder in the current iteration;
Figure FDA0003612371380000042
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; o. oiCorresponding to m for the self-encoder in the ith staff memberiAn output of (d); θ ═ WH,dH,WO,dOAnd W isHIs the weight of the coding layer; dHIs the bias of the coding layer; wOIs the weight of the decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; ρ is a sparsity parameter; KL is KL divergence; ρ is a unit of a gradientjIs the average activity of hidden layer neurons, an
Figure FDA0003612371380000051
hiOutputting the corresponding coding layer for the ith personnel,
Figure FDA0003612371380000052
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
6. The human resources scheduling model training apparatus based on human relationships according to claim 5, wherein the target model comprises four stacked self-encoders,
the input layer of the first layer self-encoder is used for receiving the relation matrix, and the output of the hidden layer of the first layer self-encoder is used as the input of the input layer of the second layer self-encoder;
the output of the hidden layer of the second layer self-encoder is used as the input of the input layer of the third layer self-encoder;
the output of the hidden layer of the third-layer self-encoder is used as the input of the input layer of the fourth-layer self-encoder;
the hidden layer output of the fourth layer self-encoder is taken as the output of the stacked self-encoder.
7. The human resource scheduling model training apparatus based on human relationship as claimed in claim 5, wherein the training module is configured to:
using the formula J-COST + λ Tr (HLH)T) The loss from the encoder in the current iteration is calculated, wherein,
j is the loss of the self-encoder in the current iteration; lambda is a parameter between the regulation and control reconstruction error and the constraint; tr is the trace of the post fitness matrix, and the post fitness matrix is
Figure FDA0003612371380000061
sijThe fitness values of the personnel i and the personnel j to the posts; h is an output matrix of the hidden layer; l is a Laplace matrix of the position fitness matrix; hTTranspose of output matrix for hidden layer;
Figure FDA0003612371380000062
a matrix of real numbers N x N;
Figure FDA0003612371380000063
Figure FDA0003612371380000064
an argument evaluation function when the minimum value is taken for the function; sigma is a summation function; m isiIs an element in the ith occupational personnel relationship matrix; n is the number of the staff; oiCorresponding to m for the self-encoder in the ith staffiAn output of (d); θ ═ WH,dH,WO,dOH, and WHIs the weight of the coding layer; dHIs the bias of the coding layer; wOWeight of decoding layer; dOA bias for decoding the layer; p is the number of hidden layer neurons; rho is a sparsity parameter; rhojIs the average activity of hidden layer neurons, an
Figure FDA0003612371380000065
hiOutputting the corresponding coding layer for the ith personnel,
Figure FDA0003612371380000066
judging whether the loss of the self-encoder in the current iteration is converged;
if so, taking the self-encoder after the current iterative training as a target model;
if not, updating theta to be { W by using a back propagation algorithm with descending random gradientH,dH,WO,dOAnd returning to the step of calculating the loss of the self-encoder in the current iteration until the loss of the self-encoder is converged.
8. Human resources scheduling device based on personnel relationship, characterized in that the device includes:
the second acquisition module is used for acquiring the personnel attributes of the personnel to be subjected to post setting and constructing an attribute matrix aiming at the personnel to be subjected to post setting based on the personnel attributes;
an input module, configured to input the attribute matrix into the target model, so as to obtain the position information corresponding to the person to be scheduled, where the target model is obtained based on the human resource scheduling model training apparatus based on human relationship as claimed in any one of claims 5 to 7.
CN201911251878.1A 2019-12-09 2019-12-09 Human resource scheduling model training and scheduling method and device based on personnel relationship Active CN111062599B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911251878.1A CN111062599B (en) 2019-12-09 2019-12-09 Human resource scheduling model training and scheduling method and device based on personnel relationship

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911251878.1A CN111062599B (en) 2019-12-09 2019-12-09 Human resource scheduling model training and scheduling method and device based on personnel relationship

Publications (2)

Publication Number Publication Date
CN111062599A CN111062599A (en) 2020-04-24
CN111062599B true CN111062599B (en) 2022-06-14

Family

ID=70300159

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911251878.1A Active CN111062599B (en) 2019-12-09 2019-12-09 Human resource scheduling model training and scheduling method and device based on personnel relationship

Country Status (1)

Country Link
CN (1) CN111062599B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111882054B (en) * 2020-05-27 2024-04-12 杭州中奥科技有限公司 Method for cross training of encryption relationship network data of two parties and related equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105608477A (en) * 2016-03-01 2016-05-25 吕云 Method and system for matching portraits with positions
JP2019053541A (en) * 2017-09-15 2019-04-04 Kddi株式会社 Information processing apparatus, information processing method and program
CN109947912A (en) * 2019-01-25 2019-06-28 四川大学 A kind of model method based on paragraph internal reasoning and combined problem answer matches
CN110287325A (en) * 2019-06-28 2019-09-27 南方电网科学研究院有限责任公司 A kind of power grid customer service recommended method and device based on intelligent sound analysis

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11205103B2 (en) * 2016-12-09 2021-12-21 The Research Foundation for the State University Semisupervised autoencoder for sentiment analysis

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105608477A (en) * 2016-03-01 2016-05-25 吕云 Method and system for matching portraits with positions
JP2019053541A (en) * 2017-09-15 2019-04-04 Kddi株式会社 Information processing apparatus, information processing method and program
CN109947912A (en) * 2019-01-25 2019-06-28 四川大学 A kind of model method based on paragraph internal reasoning and combined problem answer matches
CN110287325A (en) * 2019-06-28 2019-09-27 南方电网科学研究院有限责任公司 A kind of power grid customer service recommended method and device based on intelligent sound analysis

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Lin, Yuanmo 等.Semi-supervised human resource scheduling based on deep presentation in the cloud.《EURASIP Journal on Wireless Communications & Networking》.2020,(第1期), *
兰洋.改进谱聚类算法在高等院校人才选拔中的应用.《信阳师范学院学报》.2010,第23卷(第4期), *

Also Published As

Publication number Publication date
CN111062599A (en) 2020-04-24

Similar Documents

Publication Publication Date Title
Wang et al. Application of C4. 5 decision tree for scholarship evaluations
CN108920544A (en) A kind of personalized position recommended method of knowledge based map
CN108257052B (en) Online student knowledge assessment method and system
CN105931116A (en) Automated credit scoring system and method based on depth learning mechanism
SHIRINKINA et al. Methods of estimation of digital competences of industrial enterprises employees by means of neural network modelling.
Flitman Towards analysing student failures: neural networks compared with regression analysis and multiple discriminant analysis
CN111080241A (en) Internet platform-based data-based talent management analysis system
CN108256022A (en) Talent evaluation model building method and personnel evaluation methods and system
CN111062599B (en) Human resource scheduling model training and scheduling method and device based on personnel relationship
Lobry et al. Visual question answering from remote sensing images
CN113705242B (en) Intelligent semantic matching method and device for education consultation service
CN106650769A (en) Linear representation multi-view discrimination dictionary learning-based classification method
Ekong et al. A Machine Learning Approach for Prediction of Students’ Admissibility for Post-Secondary Education using Artificial Neural Network
CN112200208A (en) Cloud workflow task execution time prediction method based on multi-dimensional feature fusion
Shankar et al. Analyzing attrition and performance of an employee using machine learning techniques
Okoye-Ubaka et al. Neural network model for performance evaluation of Academic staff of tertiary institutions
CN113408964A (en) Method for assisting government carbon management based on comprehensive evaluation system
CN116882711B (en) Government affair hall business handling optimization method and system based on big data analysis
Nandar et al. Jobs Prediction Model based on Myanmar Labor Force Survey by using Artificial Neural Network
Thio-ac et al. Philippine Stock Price Forecasting Using Artificial Neural Network
CN114818659B (en) Text emotion source analysis method and system and storage medium
Elena et al. Methods of estimation of digital competences of industrial enterprises employees by means of neural network modelling
PANa et al. A Teaching Evaluation Model Based on SVM Multi Classification
Yang et al. Research on Freshman Registration Prediction Based on Machine Learning
Vanclay¹ A stand growth model for yield regulation in north Queensland rainforests

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant