CN115564490A - Supplier evaluation method based on attention mechanism - Google Patents

Supplier evaluation method based on attention mechanism Download PDF

Info

Publication number
CN115564490A
CN115564490A CN202211295379.4A CN202211295379A CN115564490A CN 115564490 A CN115564490 A CN 115564490A CN 202211295379 A CN202211295379 A CN 202211295379A CN 115564490 A CN115564490 A CN 115564490A
Authority
CN
China
Prior art keywords
evaluation
matrix
decision
iteration
constructing
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.)
Pending
Application number
CN202211295379.4A
Other languages
Chinese (zh)
Inventor
毛恒熙
牟明
于沛
王闯
朱守园
万胜来
郭雨枫
许政�
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.)
Avic Airborne System General Technology Co ltd
Original Assignee
Avic Airborne System General Technology Co ltd
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 Avic Airborne System General Technology Co ltd filed Critical Avic Airborne System General Technology Co ltd
Priority to CN202211295379.4A priority Critical patent/CN115564490A/en
Publication of CN115564490A publication Critical patent/CN115564490A/en
Pending legal-status Critical Current

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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Economics (AREA)
  • Educational Administration (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Tourism & Hospitality (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Operations Research (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention belongs to the technical field of supplier management, and provides a supplier evaluation method based on an attention mechanism, which comprises the following steps: step 1: determining evaluation items and weights to obtain maximum characteristic root lambda max Constructing an evaluation result iteration operator A; step 2: introducing a regular term and an attention mechanism, improving an iteration target L, constructing a neural network, iterating until an iteration threshold constraint is met, and outputting a weight output matrix A of an evaluation result iteration operator A output (ii) a And step 3: based on maximum characteristic root lambda max Constructing a consistency check operator and judging the current weight output matrix A output Whether it is valid. The method constructs an evaluation result iterative operator as a measure selected by a supplierAnd an attention mechanism and a consistency checking operator are introduced into the function, so that the evaluation accuracy is improved, and the evaluation quality of a supplier is guaranteed.

Description

Supplier evaluation method based on attention mechanism
Technical Field
The invention relates to the technical field of supplier management, in particular to a supplier evaluation method based on an attention mechanism.
Background
With the rapid development of economic and electronic information technologies, the management of suppliers becomes more important for the development and competition of enterprises and becomes a research hotspot. The supplier evaluation is a basic link of supplier management, and refers to a process that an enterprise evaluates potential suppliers in various ways before establishing a cooperative relationship with the suppliers, and continuously tracks and feeds back the potential suppliers in the process after establishing the cooperation. The evaluation and selection of suppliers are the precondition for the successful development of purchasing management and supply management, and are the foundation for enterprises to maintain long-term competitive advantages under the global situation. The accurate and efficient supplier evaluation algorithm can not only reduce the purchasing cost of enterprises and improve the product quality, but also integrate industrial resources and promote the benign and orderly growth of the whole industrial chain, thereby having great application value for the research on the supplier evaluation method.
The existing supplier evaluation algorithm mainly uses the calculation of purchasing cost to carry out comparative analysis, and selects the supplier with lower purchasing cost by calculating the sum of various expenditures including selling price, purchasing cost, transportation cost and the like. The method is selected from the perspective of purchasing cost, has great limitation and is often in conflict with the strategic goals of enterprises. Therefore, some scholars introduce a data envelope analysis DEA, and the dimensionality of evaluation of suppliers is increased by constructing multi-index input and output. Data envelope analysis cannot effectively reflect the preference of decision makers for judgment criteria, and meanwhile, due to the correlation among the unanalyzed criteria, repeated evaluation is caused. Based on the method, scholars propose a hierarchical analysis algorithm AHP, determine evaluation criteria through experts, and determine weights by adopting comprehensive consultation grading. However, AHP needs to satisfy iteration targets and constraint conditions, and when there are many element targets and the problem scale is large, it is difficult to satisfy global consistency requirements.
Disclosure of Invention
Embodiments of the present invention provide an attention-based supplier evaluation method to solve or partially solve the above problems.
The supplier evaluation method based on the attention mechanism provided by the embodiment of the invention comprises the following steps:
step 1: determining evaluation items and weights to obtain maximum characteristic root lambda max Constructing an evaluation result iteration operator A;
and 2, step: introducing a regular term and an attention mechanism, improving an iteration target L, constructing a neural network, iterating until an iteration threshold constraint is met, and outputting a weight output matrix A of an evaluation result iteration operator A output ；
And step 3: based on maximum characteristic root lambda max Constructing a consistency check operator and judging the current weight output matrix A output Whether it is valid.
According to a specific implementation manner of the embodiment of the present invention, the step 1 includes the following steps:
step 1.1: constructing a hierarchical evaluation model, determining a target layer, a criterion layer and a sub-criterion layer of a supplier evaluation decision event, and constructing a judgment matrix A aiming at the evaluation items of the sub-criterion layer 0 ：
Figure BDA0003902832920000021
Step 1.2: calculating each factor a based on the judgment matrix ij The weight W for the target layer is specifically:
for the judgment matrix A0, the product is calculated according to row elements, and then the sphere is put
Figure BDA0003902832920000022
The power of:
Figure BDA0003902832920000023
will be provided with
Figure BDA0003902832920000024
Normalization:
Figure BDA0003902832920000025
the indicator vector is then expressed as:
W＝(W 1 ，W 2 …W n ) T (4)；
step 1.3: based on the weight W, the maximum feature root λ is calculated max ：
AW＝λ max W (5)；
Further obtaining:
Figure BDA0003902832920000031
step 1.4: constructing an evaluation result iterative operator, specifically:
defining a current supplier evaluation decision event as U i Then U is i The matrix U is embedded subject to the decision event,
Figure BDA0003902832920000032
wherein m is the number of historical decision events, d is the number of decision users stored in the database, i.e. the number of decision parallel lines;
defining an evaluation item matrix
Figure BDA0003902832920000033
Wherein n is the number of evaluation items,
constructing an evaluation result iterative operator A:
A＝UV T (7)；
the iterative optimization problem can be defined as:
Figure BDA0003902832920000034
wherein, | | A-UV T || F Denotes A and its approximate value UV T Frobenius norm of (1).
According to a specific implementation manner of the embodiment of the present invention, in step 1.1, a 0 The middle element assignment uses the 1-9 scale method of Santy and satisfies:
Figure BDA0003902832920000035
according to a specific implementation manner of the embodiment of the present invention, the step 2 specifically is:
step 2.1: introducing a L2 regular term and a regular term of a gradient hidden matrix into the target function L:
Figure BDA0003902832920000036
further obtaining:
Figure BDA0003902832920000037
Figure BDA0003902832920000038
Figure BDA0003902832920000039
in the formula, L represents an objective function, i is a decision event matrix index, j is an evaluation item matrix index, and Ω represents a judgment matrix A 0 A subset of (A) r And λ g The method comprises the following steps that two regularization coefficients are adopted, M is the total number of provider evaluation decision events in a set omega, and N is the total number of evaluation items in the set omega;
step 2.2: introducing the attention mechanism, U can be further expressed as:
Figure BDA0003902832920000041
in the formula of U i Evaluation score vector for i-th decision, U p Is the historical decision vector before the ith time;
step 2.3: constructing a neural network according to the step 2.1 and the step 2.2;
step 2.4: connecting decision event vectors in a database, and constructing a decision event embedding matrix U;
step 2.5: constructing a historical decision event vector U according to the sequence number of the current iteration p ；
Step 2.6: initializing a supplier evaluation item matrix V to W of step 1.2, and setting a neural network learning rate gamma and a regular hyper-parameter lambda r And λ g Iteration threshold L 0 Maximum iteration number Maxgen, initial iteration number k =1;
step 2.7: according to step 2.2, U is i And U i And U p The convolution vector is used as input data, input into a full connection layer, and a decision event input vector is obtained through a nonlinear ReLU activation function:
Figure BDA0003902832920000042
step 2.8: obtaining a supplier evaluation matrix V according to the step 2.6, and obtaining a supplier evaluation item input vector for the iteration number k:
V input ＝V k (15)；
step 2.9: inputting the decision event input vector and the supplier evaluation item input vector into a neural network to obtain a weight output matrix of an evaluation result iteration operator A:
A output ＝Softmax(FCs2(FCs1(U input ，V input )) (16)；
step 2.10: calculating an objective function value L according to the formula (10) in the step 2.1;
step 2.11: the iteration number k = k +1, if L is more than or equal to L 0 And if k is less than Maxgen, returning to the step 2.7, otherwise, outputting the current optimal solution, and entering the step 3.1.
According to a specific implementation manner of the embodiment of the present invention, the step 3 specifically includes:
step 3.1: obtaining the maximum characteristic root lambda of the initial evaluation matrix according to the step 1.3 max Constructing a consistency check operator as follows:
Figure BDA0003902832920000051
wherein
Figure BDA0003902832920000052
n represents the order of the evaluation matrix, C 0 Is a test constant;
step 3.2: the one-time checking operator consistency of step 3.1 and a threshold value theta 0 Comparing if constancy is less than or equal to theta 0 If not, the current optimal solution output in the step 2.11 is invalid, and the step 1 is returned.
The embodiment of the invention at least has the following technical effects:
firstly, the attention mechanism introduced by the invention comprehensively considers the relationship between the historical decision and the supplier evaluation item, and can refine the core evaluation item according to the attention point of the current enterprise to improve the evaluation accuracy; and a consistency check operator is introduced to check the evaluation item combination obtained by the neural network, so that the logic problem of the evaluation matrix is avoided, and the evaluation quality of a supplier is guaranteed.
Secondly, the neural network is introduced to replace the traditional explicit solution, the training data such as the historical decision number stored in the database is increased along with the increase of the data quantity such as the decision times and the like, the attention factor is more accurate, and the effect of better use when being used is achieved.
Thirdly, the invention constructs an evaluation result iterative operator as a measurement function selected by a supplier, comprehensively considers the influence of evaluation item weight and decision events, avoids the waste of historical data and improves the evaluation precision.
Drawings
In order to more clearly illustrate the detailed description of the invention or the technical solutions in the prior art, the drawings that are needed in the detailed description of the invention or the prior art will be briefly described below. Throughout the drawings, like elements or portions are generally identified by like reference numerals. In the drawings, elements or portions are not necessarily drawn to scale.
FIG. 1 is a flow chart illustrating the steps of a supplier evaluation method based on attention mechanism according to an embodiment of the present invention;
FIG. 2 illustrates a supplier evaluation decision event hierarchy diagram in an embodiment of the invention;
FIG. 3 illustrates an attention-based neural network architecture diagram in an embodiment of the present invention.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and therefore are only examples, and the protection scope of the present invention is not limited thereby.
It is to be noted that, unless otherwise specified, technical or scientific terms used herein shall have the ordinary meaning as understood by those skilled in the art to which the invention pertains.
Fig. 1 is a flowchart illustrating steps of a supplier evaluation method based on an attention mechanism according to an embodiment of the present invention, and referring to fig. 1, the method includes the following steps:
step 1: determining evaluation items and weights to obtain maximum characteristic root lambda max And constructing an evaluation result iteration operator A. The method comprises the following specific steps:
step 1.1: combined AHP supplier valuationEstablishing a hierarchical evaluation model according to the preference of a method and a decision maker on the criteria, determining a target layer, a criterion layer and a sub-criterion layer of a supplier evaluation decision event as shown in figure 2, and constructing a judgment matrix A according to the evaluation items of the sub-criterion layer 0 ：
Figure BDA0003902832920000061
A 0 The middle element assignment uses the 1-9 scale method of Santy and satisfies:
Figure BDA0003902832920000062
step 1.2: calculating each factor a based on the judgment matrix ij The weight W for the target layer is specifically:
for decision matrix A 0 The product is calculated according to the row element and then the product is calculated
Figure BDA0003902832920000063
The power of:
Figure BDA0003902832920000064
will be provided with
Figure BDA0003902832920000065
Normalization:
Figure BDA0003902832920000066
the indicator vector can be expressed as:
W＝(W 1 ，W 2 …W n ) T (4)。
step 1.3: based on the weight W, the maximum feature root λ is calculated max ：
AW＝λ max W (5)；
Further, it can be obtained:
Figure BDA0003902832920000071
step 1.4: constructing an evaluation result iterative operator, specifically:
defining a current supplier evaluation decision event as U i Then U is i Embedding matrices from dependent decision events
Figure BDA0003902832920000072
Wherein m is the number of historical decision events, d is the number of decision users stored in the database, i.e. the number of decision parallel lines;
defining an evaluation item matrix
Figure BDA0003902832920000073
Wherein n is the number of evaluation items, and an evaluation result iterative operator A is constructed:
A＝UV T (7)；
the iterative optimization problem can be defined as:
Figure BDA0003902832920000074
wherein, | | A-UV T || F Denotes A and its approximate value UV T Frobenius norm of (1).
Step 2: introducing a regular term and an attention mechanism, improving an iteration target L, constructing a neural network, iterating until an iteration threshold constraint is met, and outputting a weight output matrix A of an evaluation result iteration operator A output . The method specifically comprises the following steps:
step 2.1: introducing a L2 regular term and a regular term of a gradient hidden matrix into the target function L:
Figure BDA0003902832920000075
further, the method can be obtained as follows:
Figure BDA0003902832920000076
Figure BDA0003902832920000077
Figure BDA0003902832920000078
in the formula, L represents an objective function, namely an iterative loss function, i is a decision event matrix index, j is an evaluation item matrix index, and omega represents a judgment matrix A 0 A subset of (A) r And λ g The method comprises the following steps that two regularization coefficients are adopted, M is the total number of provider evaluation decision events in a set omega, and N is the total number of evaluation items in the set omega.
The traditional neural network is easy to fall into overfitting, generally, the reason is that the network model considers each point in a sample, the finally formed function has large fluctuation, the function value is changed violently due to slight change of an independent variable, two regular terms of r (U, V) and g (U, V) are introduced, the weight is reduced, the fluctuation of the function value is reduced, and overfitting is avoided.
Step 2.2: when the supplier evaluates, different enterprises have own evaluation criteria, so the evaluation criteria of the current enterprise and the preference of the current decision maker for the criteria need to be fully considered. The attention mechanism is that the attention of the model to different behaviors is different when predicting, the history of the behaviors which are 'relevant' is regarded a little, and the history of the behaviors which are 'irrelevant' can be even ignored. And (3) for the decision process of supplier evaluation, introducing an attention mechanism, taking the relation between the decision event in the current round and the historical decision event as a weight, and if the correlation degree is higher, proving that the evaluation item is more concerned by the current enterprise.
In a conventional supplier evaluation algorithm, the decision events of formula (10) are embedded into a matrix U, typically a single-evaluation decision U i Simply added or weighted sum of fixed weights, i.e.
Figure BDA0003902832920000081
Directing attention to the mechanism, U can be further expressed as:
Figure BDA0003902832920000082
in the formula of U i Evaluation score vector for i-th decision, U p Is the historical decision vector before the ith time. Due to the addition of attention mechanism, U i Weight w of i Is just by U p And U i Is determined by the relationship of (A), i.e., g (U) in the formula (12) p ，U i )；
Step 2.3: constructing a neural network according to step 2.1 and step 2.2, as shown in fig. 3;
step 2.4: connecting decision event vectors in a database to construct a decision event matrix U;
step 2.5: constructing a historical decision event vector U according to the sequence number of the current iteration p ；
Step 2.6: initializing a supplier evaluation item matrix V to W of step 1.2, and setting a neural network learning rate gamma and a regular hyper-parameter lambda r And λ g Iteration threshold L 0 Maximum iteration number Maxgen, initial iteration number k =1;
step 2.7: according to step 2.2, U is i And U i And U p The convolution vector is used as input data, input into a full connection layer, and a decision event input vector is obtained through a nonlinear ReLU activation function:
Figure BDA0003902832920000091
step 2.8: obtaining a supplier evaluation matrix V according to the step 2.6, and obtaining a supplier evaluation item input vector for the iteration number k:
V input ＝V k (15)；
step 2.9: inputting the decision event input vector and the supplier evaluation item input vector into a neural network to obtain a weight output matrix of a construction evaluation result iteration operator A:
A ouptut ＝Softmax(FCs2(FCs1(U input ，V input )) (16)；
step 2.10: calculating an objective function value L according to the formula (10) in the step 2.1;
step 2.11: increasing the iteration number k = k +1, if L is more than or equal to L 0 If k is less than Maxgen, returning to the step 2.7, otherwise, outputting the current optimal solution, and entering the step 3.1;
and 3, step 3: based on maximum characteristic root lambda max Constructing a consistency check operator and judging the current weight output matrix A output Whether it is valid.
Step 3.1: constructing a consistency checking operator;
obtaining the maximum characteristic root lambda of the initial evaluation matrix according to the step 1.3 max Constructing a consistency check operator as follows:
Figure BDA0003902832920000092
wherein
Figure BDA0003902832920000093
n represents the order of the evaluation matrix, C 0 The constant can be checked by a random consistency index value-taking table.
Step 3.2: according to step 3.1, the consistency check result is calculated, and the consistency check threshold is set as theta 0 . If constancy is less than or equal to theta 0 If not, the current optimal solution output in the step 2.11 is invalid, and the step 1.1 is returned to.
And if the current optimal solution output in the step 2.11 is effective, subsequently inputting the scores of all the index items according to the evaluation matrix, and obtaining the objective scores of all the suppliers after cross calculation of corresponding weights and levels. And the current optimal solution output in the step 2.11 is invalid, and the step 1.1 is returned, namely, the reasonable initial evaluation is input againPrice matrix A 0 The method is re-executed.
It should be noted that, the modules are arranged according to a streaming layout, which is only one embodiment of the present invention, and may also be arranged in other manners, and the present invention is not limited to this.
The embodiment of the invention has the following technical effects:
firstly, the method obtains the initial index weight based on the AHP supplier evaluation algorithm, retains the advantages of the AHP algorithm, focuses on the correlation among analysis criteria, and refines the preference of a decision maker to the criteria;
secondly, an evaluation result iterative operator is constructed and used as a measurement function selected by a supplier, the influence of evaluation item weight and decision events is comprehensively considered, the waste of historical data is avoided, and the evaluation precision is improved;
thirdly, the attention mechanism introduced by the invention comprehensively considers the relationship between the historical decision and the supplier evaluation item, and can refine the core evaluation item according to the attention point of the current enterprise, thereby improving the evaluation accuracy;
fourthly, the neural network is introduced to replace the traditional explicit solution, along with the increase of data quantity such as decision times and the like, training data such as historical decision numbers and the like stored in the database are increased, the attention factor is more accurate, and the effect of better use when being used is achieved;
and fifthly, the consistency check operator is introduced to check the evaluation item combination obtained by the neural network, so that the logic problem of the evaluation matrix is avoided, and the evaluation quality of a supplier is guaranteed.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (5)

1. A supplier evaluation method based on attention mechanism, characterized by comprising the following steps:
step 1: determining evaluation items and weights to obtain maximum characteristic root lambda max Constructing an evaluation result iteration operator A;
and 2, step: introducing a regular term and an attention mechanism, improving an iteration target L, constructing a neural network, iterating until an iteration threshold constraint is met, and outputting a weight output matrix A of the evaluation result iteration operator A output ；
And step 3: based on the maximum feature root λ max Constructing a consistency check operator and judging the current weight output matrix A output Whether it is valid.
2. The supplier evaluation method according to claim 1, wherein the step 1 includes the steps of:
step 1.1: constructing a hierarchical evaluation model, determining a target layer, a criterion layer and a sub-criterion layer of a supplier evaluation decision event, and constructing a judgment matrix A aiming at the evaluation items of the sub-criterion layer 0 ：
Figure FDA0003902832910000011
Step 1.2: calculating each factor a based on the judgment matrix ij The weight W for the target layer is specifically:
for decision matrix A 0 By row element, the product is calculated and then the product is calculated
Figure FDA0003902832910000012
The power of:
Figure FDA0003902832910000013
will be provided with
Figure FDA0003902832910000014
Normalization:
Figure FDA0003902832910000015
the indicator vector is then expressed as:
W＝(W 1 ,W 2 …W n ) T (4)；
step 1.3: based on the weight W, the maximum feature root λ is calculated max ：
AW＝λ max W (5)；
Further obtaining:
Figure FDA0003902832910000021
step 1.4: constructing an evaluation result iterative operator, specifically:
defining a current supplier evaluation decision event as U i Then U is i The matrix U is embedded subject to the decision event,
Figure FDA0003902832910000022
wherein m is the number of historical decision events, d is the number of decision users stored in the database, i.e. the number of decision parallel lines;
defining an evaluation item matrix
Figure FDA0003902832910000023
Wherein n is the number of evaluation items,
constructing an evaluation result iterative operator A:
A＝UV T (7)；
the iterative optimization problem is defined as:
Figure FDA0003902832910000024
wherein, | | A-UV T || F Denotes A and its approximate value UV T Frobenius norm of (1).
3. The supplier evaluation method according to claim 2, characterized in that: in said step 1.1, A 0 The middle element assignment uses the 1-9 scale method of Santy and satisfies:
Figure FDA0003902832910000025
4. the supplier evaluation method according to claim 2, wherein the step 2 is specifically:
step 2.1: introduction into the objective function L
Figure FDA00039028329100000210
Regularization term and regularization term of the gradient latent matrix:
Figure FDA0003902832910000026
further obtaining:
Figure FDA0003902832910000027
Figure FDA0003902832910000028
Figure FDA0003902832910000029
in the formula, L represents an objective function, i is a decision event matrix index, j is an evaluation item matrix index, and Ω represents a judgment matrix A 0 A subset of (A) r And λ g The method comprises the following steps that two regularization coefficients are adopted, M is the total number of provider evaluation decision events in a set omega, and N is the total number of evaluation items in the set omega;
step 2.2: with attention paid to the mechanism, U is further expressed as:
Figure FDA0003902832910000031
in the formula of U i Evaluation score vector for i-th decision, U p Is the historical decision vector before the ith time;
step 2.3: constructing a neural network according to the step 2.1 and the step 2.2;
step 2.4: connecting decision event vectors in a database to construct a decision event matrix U;
step 2.5: constructing a historical decision event vector U according to the sequence number of the current iteration p ；
Step 2.6: initializing a supplier evaluation item matrix V to W of step 1.2, and setting a neural network learning rate gamma and a regular hyper-parameter lambda r And λ g Iteration threshold L 0 Maximum iteration number Maxgen, initial iteration number k =1;
step 2.7: according to step 2.2, U is i And U i And U p The convolution vector is used as input data, input into a full connection layer, and a decision event input vector is obtained through a nonlinear ReLU activation function:
Figure FDA0003902832910000032
step 2.8: obtaining a supplier evaluation matrix V according to the step 2.6, and obtaining a supplier evaluation item input vector for the iteration number k:
V input ＝V k (15)；
step 2.9: inputting the decision event input vector and the supplier evaluation item input vector into a neural network to obtain a weight output matrix of an evaluation result iteration operator A:
A output ＝Softmax(FCs2(FCs1(U input ,V input )) (16)；
step 2.10: calculating an objective function value L according to the formula (10) in the step 2.1;
step 2.11: the iteration number k = k +1, if L is more than or equal to L 0 And k is<And (5) the Maxgen returns to the step 2.7, otherwise, the current optimal solution is output.
5. The supplier evaluation method according to claim 4, wherein the step 3 specifically comprises:
step 3.1: obtaining the maximum characteristic root lambda of the initial evaluation matrix according to the step 1.3 max Constructing a consistency check operator as follows:
Figure FDA0003902832910000041
wherein
Figure FDA0003902832910000042
n represents the order of the evaluation matrix, C 0 Is a test constant;
step 3.2: the one-time checking operator constenstxy of the step 3.1 and a threshold value theta 0 Comparing if constancy is less than or equal to theta 0 If not, the current optimal solution output in the step 2.11 is invalid, and the step 1 is returned.
CN202211295379.4A 2022-10-21 2022-10-21 Supplier evaluation method based on attention mechanism Pending CN115564490A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211295379.4A CN115564490A (en) 2022-10-21 2022-10-21 Supplier evaluation method based on attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211295379.4A CN115564490A (en) 2022-10-21 2022-10-21 Supplier evaluation method based on attention mechanism

Publications (1)

Publication Number Publication Date
CN115564490A true CN115564490A (en) 2023-01-03

Family

ID=84746648

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211295379.4A Pending CN115564490A (en) 2022-10-21 2022-10-21 Supplier evaluation method based on attention mechanism

Country Status (1)

Country Link
CN (1) CN115564490A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314233A (en) * 2023-09-21 2023-12-29 国网河北省电力有限公司物资分公司 Vendor evaluation method, device, terminal equipment and medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314233A (en) * 2023-09-21 2023-12-29 国网河北省电力有限公司物资分公司 Vendor evaluation method, device, terminal equipment and medium

Similar Documents

Publication Publication Date Title
Więckowski et al. RANCOM: A novel approach to identifying criteria relevance based on inaccuracy expert judgments
Aguirregabiria et al. Dynamic discrete choice structural models: A survey
Martinez-Jaramillo et al. An heterogeneous, endogenous and coevolutionary GP-based financial market
De et al. Green logistics under imperfect production system: A rough age based multi-objective genetic algorithm approach
CN115564490A (en) Supplier evaluation method based on attention mechanism
CN116452052A (en) Patent value evaluation method, system, device and storage medium
CN111861751A (en) Stock quantitative trading method based on deep reinforcement learning, storage medium and equipment
CN110223180A (en) Portfolio Selection Based method based on depth attention network and intensified learning
CN115496290A (en) Medium-and-long-term runoff time-varying probability prediction method based on &#39;input-structure-parameter&#39; full-factor hierarchical combination optimization
Kochliaridis et al. Combining deep reinforcement learning with technical analysis and trend monitoring on cryptocurrency markets
Huang et al. A novel stock trading utilizing long short term memory prediction and evolutionary operating-weights strategy
CN108520380A (en) A kind of inventory forecast method based on improved BP neural network
Corral et al. Poverty mapping in the age of machine learning
CN115471009A (en) Predictive optimized power system planning method
CN115841345A (en) Cross-border big data intelligent analysis method, system and storage medium
CN115798632A (en) Medicine molecular structure aided design method based on neural network
Amin Naseri et al. Modeling portfolio optimization based on fundamental analysis using an expert system in the real estate industry
Cárdenas-Cárdenas et al. Colombian inflation forecast using long short-term memory approach
Alves et al. A new exact method for linear bilevel problems with multiple objective functions at the lower level
CN113435938A (en) Distributed characteristic data selection method in electric power spot market
CN114612231A (en) Stock quantitative trading method and device, terminal device and readable storage medium
Zha et al. A hierarchical reinforcement learning framework for stock selection and portfolio
Li et al. DeepDLP: deep reinforcement learning based framework for dynamic liner trade pricing
Wang et al. Research on Credit Decision Issues of the Small and Medium-Sized Enterprises Based on TOPSIS and Hierarchical Cluster Analysis
Shen et al. PortPortfolio Management based on Deep Reinforcement Learning Method with Data Augment

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