Methods and systems for constructing contrast matrices, and methods and systems for hierarchical analysis.

By automatically constructing the comparison matrix through a neural network-trained matrix generator, the problem of time-consuming and labor-intensive manual construction in the analytic hierarchy process is solved, improving efficiency and accuracy and expanding the scope of application.

CN117764192BActive Publication Date: 2026-11-10CHINA UNIONPAY
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310953347.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-07-31
Publication Date
2026-11-10
Estimated Expiration
2043-07-31

AI Technical Summary

Technical Problem

The existing analytic hierarchy process requires a lot of manual intervention when constructing the comparison matrix, which is costly in terms of manpower and resources and has limitations, making it difficult to meet the needs of complex decision-making tasks.

Method used

A neural network model is used to train a matrix generator, which automatically constructs a comparison matrix by combining factor features and obtaining labels. The comparison matrix is ​​then generated using a factor discrimination model and a hierarchical scoring model.

Benefits of technology

It reduces the manpower required for constructing the comparison matrix, improves the processing efficiency and accuracy of the analytic hierarchy process, and broadens its application areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117764192B_ABST
    Figure CN117764192B_ABST
Patent Text Reader

Abstract

This application relates to methods and systems for training matrix generators in machine learning, methods and systems for constructing contrast matrices, hierarchical analysis methods and systems, and computer-readable storage media for implementing the above methods. According to one aspect of this application, a method for constructing a contrast matrix is ​​provided, the method comprising the steps of: obtaining a set of factor features; and inputting the set of factor features into a matrix generator to obtain a contrast matrix, wherein the matrix generator is trained using a method for training a matrix generator provided according to another aspect of this application. By automatically constructing a contrast matrix based on a set of factor features using a trained matrix generator, the manual effort in the contrast matrix construction process can be saved, the efficiency and accuracy of using hierarchical analysis to solve problems can be improved, and the application field of hierarchical analysis can be broadened.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to methods and systems for training matrix generators in machine learning, methods and systems for constructing contrast matrices, hierarchical analysis methods and systems, and computer-readable storage media for implementing the above methods. Background Technology

[0002] The Analytic Hierarchy Process (AHP) is a comprehensive evaluation method for system analysis and decision-making. It is a multi-factor decision-making method proposed based on a thorough study of human thought processes, and it can reasonably realize the quantitative processing of qualitative problems.

[0003] The Analytic Hierarchy Process (AHP) transforms complex problem-solving processes into comparisons of the importance of pairwise factors by constructing a hierarchical structure, thus converting difficult-to-quantify qualitative judgments into actionable comparisons of importance. Specifically, AHP transforms multi-factor decision-making tasks into a comparison matrix, and then obtains the priority weight of each factor at each level relative to a factor at the previous level by solving the eigenvectors of the comparison matrix. It is a method that combines qualitative and quantitative approaches. Since its inception, AHP has been widely applied in areas such as corporate capability assessment, merchant value assessment, customer demand rating, supplier selection, manufacturing systems, and strategic decision-making in the financial sector.

[0004] Typically, the Analytic Hierarchy Process (AHP) includes four steps: problem modeling, constructing the comparison matrix, priority derivation, and consistency verification. Currently, the construction of the comparison matrix requires manual comparison and scoring by experts. However, as the complexity of multi-factor decision-making tasks increases, the number of factors involved multiplies, leading to a significant expenditure of human and material resources for constructing the comparison matrix. Furthermore, comparison matrices constructed manually based on limited cognitive biases have limitations, and sometimes it is even impossible to construct them manually, affecting the processing efficiency of the AHP and limiting its application areas. Summary of the Invention

[0005] To address or at least alleviate one or more of the above problems, the following technical solutions are provided.

[0006] According to a first aspect of this application, a method for training a matrix generator is provided, the matrix generator comprising a factor discrimination model and a hierarchical scoring model, the method comprising the following steps: obtaining a set of factor feature combinations and a factor discrimination label and a hierarchical scoring label for each factor feature combination; training a first neural network model using the set of factor feature combinations and the factor discrimination label for each factor feature combination as a first training set and updating the trained first neural network model using a first loss function to obtain the factor discrimination model; and training a second neural network model using the set of factor feature combinations and the hierarchical scoring label for each factor feature combination as a second training set and updating the trained second neural network model using a second loss function to obtain the hierarchical scoring model, wherein the second loss function is constructed at least based on the reciprocal of the factor discrimination result output by the factor discrimination model and the hierarchical scoring result output by the trained second neural network model.

[0007] A method for training a matrix generator according to an embodiment of this application, wherein the matrix generator is used to generate a contrast matrix based on multiple input factors in the analytic hierarchy process.

[0008] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein obtaining a set of factor feature combinations includes: obtaining a plurality of factors and a semantic vector of each of the plurality of factors; combining the semantic vectors of each of the plurality of factors into a set of factor features; combining the factor features in the set of factor features in pairs to generate a plurality of factor feature pairs; and using the plurality of factor feature pairs as the set of factor feature combinations, wherein each factor feature combination includes a semantic vector of a first factor and a semantic vector of a second factor.

[0009] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein obtaining the factor discriminant label for each factor feature combination includes: obtaining the factor discriminant label for each factor feature combination based on the importance of the first factor relative to the second factor.

[0010] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein obtaining a factor discriminant label for each factor feature combination further includes: setting the factor discriminant label to a first value in response to determining that the first factor is more important than the second factor; and setting the factor discriminant label to a second value in response to determining that the second factor is more important than the first factor.

[0011] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein obtaining the hierarchical rating label for each factor feature combination includes: obtaining the hierarchical rating label for each factor feature combination based on the importance of the first factor relative to the second factor.

[0012] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein obtaining a hierarchical rating label for each factor feature combination further includes: determining a rating to represent the importance of the first factor relative to the second factor; setting the hierarchical rating label to the rating in response to the first factor being more important than the second factor; and setting the hierarchical rating label to the reciprocal of the rating in response to the second factor being more important than the first factor.

[0013] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein the first loss function is constructed based on the factor discrimination label of each factor feature combination and the factor discrimination result output by the trained first neural network model.

[0014] According to one embodiment or any of the above embodiments of this application, the method for training a matrix generator, wherein updating the trained first neural network model using a first loss function includes: adjusting the parameters of the trained first neural network model using the first loss function until the first loss function is in a convergent state, so as to obtain the factor discrimination model.

[0015] According to one embodiment or any of the above embodiments of this application, the method for training a matrix generator, wherein the second loss function is constructed based on the hierarchical rating label of each factor feature combination, the factor discrimination result output by the factor discrimination model, the hierarchical rating result output by the trained second neural network model, and the reciprocal of the hierarchical rating result output by the trained second neural network model.

[0016] The method for training a matrix generator according to one embodiment or any of the above embodiments of this application, wherein updating the trained second neural network model using a second loss function includes: adjusting the parameters of the trained second neural network model using the second loss function until the second loss function is in a convergent state, so as to obtain the hierarchical scoring model.

[0017] According to a second aspect of this application, a system for training a matrix generator is provided, the matrix generator including a factor discrimination model and a hierarchical scoring model, the system comprising: an acquisition unit configured to acquire a set of factor feature combinations and a factor discrimination label and a hierarchical scoring label for each factor feature combination; a first training unit configured to train a first neural network model using the set of factor feature combinations and the factor discrimination label for each factor feature combination as a first training set and update the trained first neural network model using a first loss function to obtain the factor discrimination model; and a second training unit configured to train a second neural network model using the set of factor feature combinations and the hierarchical scoring label for each factor feature combination as a second training set and update the trained second neural network model using a second loss function to obtain the hierarchical scoring model, wherein the second loss function is constructed at least based on the reciprocal of the factor discrimination result output by the factor discrimination model and the hierarchical scoring result output by the trained second neural network model.

[0018] According to a third aspect of this application, a computer-readable storage medium is provided, comprising instructions that, when executed, perform the steps of the method for training a matrix generator according to a first aspect of this application.

[0019] According to a fourth aspect of this application, a method for constructing a contrast matrix is ​​provided, the method comprising the steps of: obtaining a set of factor features; and inputting the set of factor features into a matrix generator to obtain a contrast matrix, wherein the matrix generator is trained using the method for training a matrix generator according to a first aspect of this application.

[0020] According to a fifth aspect of this application, a system for constructing a contrast matrix is ​​provided, the system comprising: an acquisition unit configured to acquire a set of factor features; and a matrix generator configured to process the set of factor features to obtain a contrast matrix, wherein the matrix generator is trained using a method for training a matrix generator according to a first aspect of this application.

[0021] According to a sixth aspect of this application, a computer-readable storage medium is provided, comprising instructions that, when executed, perform the steps of the method for constructing a contrast matrix according to a fourth aspect of this application.

[0022] According to a seventh aspect of this application, a hierarchical analysis method is provided, the hierarchical analysis method comprising the following steps: obtaining a set of factor features of multiple factors associated with a target task; inputting the set of factor features into a matrix generator to obtain a contrast matrix, wherein the matrix generator is trained using a method for training a matrix generator according to a first aspect of this application; processing the contrast matrix to determine the priority of the multiple factors; and performing a consistency check on the contrast matrix using the priority of the multiple factors, and determining the weights of the multiple factors using the contrast matrix that passes the consistency check.

[0023] According to an eighth aspect of this application, a hierarchical analysis system is provided, characterized in that the system comprises: a memory; a processor coupled to the memory; and a computer program stored on the memory and running on the processor, the execution of the computer program causing the following operations: acquiring a set of factor features of a plurality of factors associated with a target task; inputting the set of factor features to a matrix generator to obtain a contrast matrix, wherein the matrix generator is trained using a method for training a matrix generator according to a first aspect of this application; processing the contrast matrix to determine the priority of the plurality of factors; and performing a consistency check on the contrast matrix using the priority of the plurality of factors, and determining the weights of the plurality of factors using the contrast matrix that passes the consistency check.

[0024] According to a ninth aspect of this application, a computer-readable storage medium is provided, comprising instructions that, when executed, perform the steps of the hierarchical analysis method according to a seventh aspect of this application.

[0025] According to one or more embodiments of this application, by automatically constructing a contrast matrix based on a set of factor features using a trained matrix generator, the manual labor cost in the contrast matrix construction process can be saved, the efficiency and accuracy of using the analytic hierarchy process can be improved, and the application field of the analytic hierarchy process can be broadened. Attached Figure Description

[0026] The above and / or other aspects and advantages of this application will become clearer and more readily understood from the following description taken in conjunction with the accompanying drawings, in which the same or similar elements are denoted by the same reference numerals. The drawings include:

[0027] Figure 1 A schematic diagram of a typical hierarchical analysis method model is shown.

[0028] Figure 2 A flowchart of a method for training a matrix generator according to one or more embodiments of this application is shown.

[0029] Figure 3A block diagram of a system for training a matrix generator according to one or more embodiments of this application is shown.

[0030] Figure 4 A flowchart of a method for constructing a contrast matrix according to one or more embodiments of this application is shown.

[0031] Figure 5 A block diagram of a system for constructing a contrast matrix according to one or more embodiments of this application is shown.

[0032] Figure 6 A flowchart of a hierarchical analysis method according to one or more embodiments of this application is shown.

[0033] Figure 7 A block diagram of a hierarchical analysis system according to one or more embodiments of this application is shown. Detailed Implementation

[0034] The present application will now be described more fully with reference to the accompanying drawings, which illustrate exemplary embodiments thereof. However, the present application may be implemented in various forms and should not be construed as being limited to the embodiments given herein. The foregoing embodiments are intended to make the disclosure herein complete and thorough, so as to more fully convey the scope of protection of the present application to those skilled in the art.

[0035] In this specification, terms such as “comprising” and “including” indicate that, in addition to having the units and steps that are directly and explicitly stated in the specification and claims, the technical solution of this application does not exclude the presence of other units and steps that are not directly or explicitly stated.

[0036] Unless otherwise specified, terms such as “first” and “second” do not indicate the order of units in terms of time, space, size, etc., but are merely used to distinguish between units.

[0037] In the following, various exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings.

[0038] Figure 1 A schematic diagram of a typical hierarchical analysis method model is shown.

[0039] like Figure 1 As shown, the hierarchical analysis method model 100 divides the multi-factor decision problem into three layers: the first layer is the problem overview, the second layer is the factor system, and the third layer is the candidate decision.

[0040] exist Figure 1In this framework, the first layer, the Problem Simple Description (PSD), describes the decision problem to be solved. The second layer, the factor system, includes multiple factors involved in addressing the decision problem, which can be represented as a set M = {m1, m2, m3, ..., m...}. k ,…,m n}, where m1, m2, m3, ..., m k ,…,m n Let M represent multiple factors, and n represent the size of the set M, i.e., the number of factors. The third layer of candidate decisions includes multiple candidate decisions related to the problem to be solved, which can be represented as a set TC = {tc1, tc2, tc3, ..., tc...}. k ,…,tc o}, where tc1, tc2, tc3, ..., tc k ,…,tc o This represents multiple candidate decisions, where 'o' represents the size of the set TC, i.e., the number of candidate decisions. It's important to note that the processing method for the candidate decision layer varies depending on the decision problem. For example, in a single-choice decision problem, only one candidate decision in the candidate decision set TC is output; in a multi-objective evaluation context, each candidate decision in the candidate decision set TC and its corresponding score can be output.

[0041] In traditional hierarchical analysis (AHP) methods, the process of constructing a comparison matrix based on the factor system of the second level requires expert participation in comparison and scoring. However, when the number of factors to be compared in a decision-making problem increases exponentially, constructing a comparison matrix through manual comparison and scoring is not only time-consuming and labor-intensive, but also difficult to meet practical needs, and may even fail to construct a comparison matrix at all.

[0042] Figure 2 A flowchart of a method for training a matrix generator according to one or more embodiments of this application is shown, wherein the matrix generator includes a factor discrimination model and a hierarchical scoring model. Optionally, the matrix generator can be used to generate a contrast matrix based on multiple input factors in the analytic hierarchy process (AHP).

[0043] like Figure 2 As shown, in step S201, a set of factor feature combinations and factor discrimination labels and hierarchical scoring labels for each factor feature combination are obtained.

[0044] Optionally, in step S201, multiple factors and the semantic vector of each factor can be obtained, the semantic vectors of each factor can be combined into a set of factor features, the factor features in the set of factor features can be combined in pairs to generate multiple factor feature pairs, and the multiple factor feature pairs can be used as a set of factor feature combinations, wherein each factor feature combination may include the semantic vector of the first factor and the semantic vector of the second factor.

[0045] In one embodiment, the multiple factors involved in processing the decision problem to be solved can be represented as a set M = {m1, m2, m3, ..., m}. k ,…,m n}, where m1, m2, m3, ..., m k ,…,m n Let M represent multiple factors, where n represents the size of the set M, i.e., the number of factors. Then, these multiple factors can be represented using natural language. For example, a pre-trained natural language model (e.g., BERT, GPT, etc.) can be used to obtain the semantic vector of each factor, forming a set of factor features ME = {em1, em2, em3, ..., em...}. k ,…,em n},in Indicator m k The semantic vectors are defined as follows: t represents the dimension of the semantic vector, and n represents the size of the set ME, i.e., the number of semantic vectors of the factor. The set ME = {em1, em2, em3, ..., em...} is used to obtain the factor features. k ,…,em n After that, the factor features em1, em2, em3, ..., em in the set of factor features ME can be... k ,…,em n Pairwise combinations are performed to generate multiple factor feature pairs. For example, factor features em1 and em2 can be combined to generate a factor feature pair (em1, em2). For instance, N factors from multiple domains can be collected and represented as features, and then the multiple factor feature pairs can be used as a set CME of factor feature combinations, represented as (em1, em2). i ,em j )∈CME,i≠j,i,j={1,2,3,…,N}, where N represents the number of factors.

[0046] In one embodiment, the factor discrimination label of a factor feature combination can be obtained based on the importance of a first factor relative to a second factor within that combination. When it is determined that the first factor is more important than the second factor, the factor discrimination label of the factor feature combination composed of the first and second factors can be set to a first value; and when it is determined that the second factor is more important than the first factor, the factor discrimination label of the factor feature combination composed of the first and second factors can be set to a second value. For example, for a factor feature combination (em... i ,em j The factor feature combination (em) can be set using the following formula (1). i ,em j Factor discrimination label flag ij :

[0047]

[0048] Among them, em>em j Indicator m i Factor m j Importantly, at this point, factor characteristics can be combined (em) i ,em j Factor discrimination label flag ij Set to 1; and em i j Indicator m j Factor m i Importantly, at this point, factor characteristics can be combined (em) i ,em j Factor discrimination label flag ij Set to 0.

[0049] In one embodiment, a hierarchical rating label for a factor feature combination can be obtained based on the importance of a first factor relative to a second factor within that combination. Optionally, a rating can be determined to represent the importance of the first factor relative to the second factor. The hierarchical rating label is set to the rating when the first factor is more important than the second factor, and the hierarchical rating label is set to the reciprocal of the rating when the second factor is more important than the first factor. For example, the importance rating can include nine levels: {1, 2, 3, 4, 5, 6, 7, 8, 9}. A higher rating indicates a higher importance of the first factor relative to the second factor. For instance, a rating of 1 indicates that the first and second factors are equally important, a rating of 3 indicates that the first factor is slightly more important than the second factor, a rating of 5 indicates that the first factor is significantly more important than the second factor, and a rating of 9 indicates that the first factor is extremely more important than the second factor. For example, for a factor feature combination (em... i ,em​j Factor m can be utilized using the following formula (2). i Relative to factor m j The importance score is used to set the factor feature combination (em). i ,em j The score level label (score_level) ij :

[0050]

[0051] Among them, em i >em j Indicator m i Factor m j Importantly, at this point, factor characteristics can be combined (em) i ,em j The score level label (score_level) ij Set as score; and em i j Indicator m j Factor m i Importantly, at this point, factor characteristics can be combined (em) i ,em j The score level label (score_level) ij Set to 1 / score. For example, score∈{1,2,3,4,5,6,7,8,9} represents factor m. i Compared to factor m j The importance level is rated.

[0052] In step S203, the set of factor feature combinations and the factor discrimination label of each factor feature combination are used as the first training set to train the first neural network model, and the first loss function is used to update the trained first neural network model to obtain the factor discrimination model.

[0053] Referring to the examples above, in obtaining factor feature combinations (em) i ,em j ) and the corresponding factor discrimination label flag ij Then, the first training set SC can be set as follows:

[0054] SC = {sc1, sc2, sc3, ..., sc k ,…,sc N(N-1) / 2} Formula (3)

[0055] Among them, sc N(i-1)-(i(i+1)) / 2+j =((em) i ,em j ​), flag ij N represents the number of factors.

[0056] Optionally, the first neural network model may include a first layer, a second layer, and an output layer, which may be constructed as follows:

[0057] Z1 = W1(em) i ,em j )+b1

[0058] Z2=W2Z1+b2 Formula (4)

[0059]

[0060] Where Z1 is the output of the first layer, (em i ,em j Z1 is the input of the first layer, W1 represents the weight of the first layer, and b1 represents the bias of the first layer; Z2 is the output of the second layer, Z1 is the input of the second layer, W2 represents the weight of the second layer, and b2 represents the bias of the second layer. Z1 is the output of the output layer, and Z2 is the input of the output layer. For example, Where e represents the natural base and K represents the total number of elements in the set containing x.

[0061] Optionally, the first loss function can be constructed based on the factor discrimination label of each factor feature combination and the factor discrimination result output by the trained first neural network model. For example, the first loss function... It can be constructed as follows:

[0062]

[0063] Where ln represents the logarithm to the base e, and y represents the factor label. This represents the factor discrimination result output by the first trained neural network model.

[0064] Optionally, in step S203, the parameters of the trained first neural network model can be adjusted using the first loss function until the first loss function is in a convergent state to obtain the factor discrimination model.

[0065] In step S205, the set of factor feature combinations and the hierarchical rating labels for each factor feature combination are used as the second training set to train the second neural network model, and the trained second neural network model is updated using the second loss function to obtain the hierarchical rating model. The second loss function is constructed based at least on the reciprocal of the factor discrimination result output by the factor discrimination model and the hierarchical rating result output by the trained second neural network model.

[0066] Referring to the examples above, in obtaining factor feature combinations (em) i ,em j ) and the corresponding level score label after score_level ij The second training set SL can be set as follows:

[0067] SL = {sl1, sl2, sl3, ..., sl} k ,…,sl N(N-1) / 2} Formula (6)

[0068] Among them, sc N(i-1)-(i(i+1)) / 2+j =((em) i ,em j ), score_level ij ), where N represents the number of factors.

[0069] Optionally, the second neural network model may include a first layer, a second layer, and an output layer, which may be constructed as follows:

[0070]

[0071] Where Z1 is the output of the first layer, (em i ,em j Z1 is the input of the first layer, W1 represents the weight of the first layer, and b1 represents the bias of the first layer; Z2 is the output of the second layer, Z1 is the input of the second layer, W2 represents the weight of the second layer, and b2 represents the bias of the second layer. Z1 is the output of the output layer, and Z2 is the input of the output layer. For example, Where e represents the natural base and K represents the total number of elements in the set containing x.

[0072] Optionally, the second loss function can be constructed based on the hierarchical rating label for each factor feature combination, the factor discrimination result output by the factor discrimination model, the hierarchical rating result output by the trained second neural network model, and the reciprocal of the hierarchical rating result output by the trained second neural network model. For example, the second loss function... It can be constructed as follows:

[0073]

[0074] Where ln represents the logarithm to the base e, and y represents the hierarchical rating label. This represents the hierarchical scoring result output by the trained second neural network model, while flag represents the factor discrimination result output by the factor discrimination model.

[0075] Alternatively, the second loss function It can be constructed as follows:

[0076]

[0077] Where ln represents the logarithm to the base e, and y represents the hierarchical rating label. This represents the hierarchical scoring result output by the trained second neural network model, while flag represents the factor discrimination result output by the factor discrimination model.

[0078] By incorporating the factor discrimination result flag output by the factor discrimination model and the reciprocal of the hierarchical score output by the trained second neural network model into the second loss function... It can adapt to the hierarchical scoring rules of reciprocals in the process of constructing the comparison matrix in the analytic hierarchy process, so that the hierarchical scoring model can be better combined with the factor discrimination model, thereby improving the accuracy and rationality of matrix generation.

[0079] Optionally, in step S205, the parameters of the trained second neural network model can be adjusted using the second loss function until the second loss function is in a convergent state to obtain a hierarchical scoring model.

[0080] Optionally, after obtaining the factor discrimination model and the hierarchical scoring model, the two models can be concatenated to obtain a matrix generator, where factor features can be combined (em). i ,em j The input is fed into the matrix generator, and the output is the feature combination of that factor (em). i ,em j The score level label (score_level) ij The set of factor characteristics ME = {em1, em2, em3, ..., em} k ,…,em n Multiple factor feature combinations in pairwise combinations (em) i ,em j The input is fed into the matrix generator to obtain the feature combination (em) for each factor. i ,em j The score level label (score_level) ij This is used to construct a contrast matrix. The method for constructing the contrast matrix will be described further below.

[0081] Figure 3 A block diagram of a system for training a matrix generator according to one or more embodiments of this application is shown.

[0082] like Figure 3 As shown, the system 300 for training a matrix generator may include an acquisition unit 310, a first training unit 320, and a second training unit 330.

[0083] The acquisition unit 310 is configured to acquire a set of factor feature combinations and a factor discrimination label and a hierarchical scoring label for each factor feature combination.

[0084] Optionally, the acquisition unit 310 can be configured to acquire multiple factors and the semantic vector of each factor among the multiple factors, combine the semantic vectors of each factor among the multiple factors into a set of factor features, combine the factor features in the set of factor features in pairs to generate multiple factor feature pairs, and use the multiple factor feature pairs as a set of factor feature combinations, wherein each factor feature combination may include the semantic vector of the first factor and the semantic vector of the second factor.

[0085] Optionally, the acquisition unit 310 can be configured to acquire the factor discrimination label of a factor feature combination based on the importance of the first factor relative to the second factor in the factor feature combination. When it is determined that the first factor is more important than the second factor, the factor discrimination label of the factor feature combination composed of the first factor and the second factor can be set to a first value; and when it is determined that the second factor is more important than the first factor, the factor discrimination label of the factor feature combination composed of the first factor and the second factor can be set to a second value.

[0086] Optionally, the acquisition unit 310 can be configured to acquire the hierarchical rating label of a factor feature combination based on the importance of the first factor relative to the second factor in that combination. Optionally, a rating representing the importance of the first factor relative to the second factor can be determined, and the hierarchical rating label is set to the rating when the first factor is more important than the second factor, and the hierarchical rating label is set to the reciprocal of the rating when the second factor is more important than the first factor.

[0087] The first training unit 320 is configured to train a first neural network model using a set of factor feature combinations and a factor discrimination label for each factor feature combination as a first training set, and to update the trained first neural network model using a first loss function to obtain a factor discrimination model. Optionally, the first loss function can be constructed based on the factor discrimination label for each factor feature combination and the factor discrimination result output by the trained first neural network model.

[0088] Optionally, the first training unit 320 can be configured to adjust the parameters of the trained first neural network model using a first loss function until the first loss function is in a convergent state, so as to obtain a factor discrimination model.

[0089] The second training unit 330 is configured to train a second neural network model using the set of factor feature combinations and the hierarchical rating labels for each factor feature combination as a second training set, and to update the trained second neural network model using a second loss function to obtain a hierarchical rating model. The second loss function is constructed based at least on the factor discrimination result output by the factor discrimination model and the reciprocal of the hierarchical rating result output by the trained second neural network model. Optionally, the second loss function can be constructed based on the hierarchical rating labels for each factor feature combination, the factor discrimination result output by the factor discrimination model, the hierarchical rating result output by the trained second neural network model, and the reciprocal of the hierarchical rating result output by the trained second neural network model.

[0090] By incorporating the factor discrimination results output by the factor discrimination model and the reciprocals of the hierarchical scoring results output by the trained second neural network model into the second loss function, it is possible to adapt to the hierarchical scoring rules of reciprocals in the process of constructing the comparison matrix in the analytic hierarchy process. This allows the hierarchical scoring model to be better integrated with the factor discrimination model, thereby improving the accuracy and rationality of matrix generation.

[0091] Optionally, the second training unit 330 can be configured to adjust the parameters of the trained second neural network model using a second loss function until the second loss function is in a convergent state to obtain a hierarchical scoring model.

[0092] Figure 4 A flowchart of a method for constructing a contrast matrix according to one or more embodiments of this application is shown.

[0093] like Figure 4 As shown, in step S401, a set of factor features is obtained.

[0094] Optionally, in step S401, the semantic vector of each of the multiple factors associated with the target task can be obtained, and the semantic vectors of each of the multiple factors can be combined to obtain a set of factor features. For example, the multiple factors associated with the target task can be represented as a set M = {m1, m2, m3, ..., m...} k ,…,m n}, where m1, m2, m3, ..., m k ,…,m n Let M represent multiple factors, where n represents the size of the set M, i.e., the number of factors. Then, these multiple factors can be represented using natural language. For example, a pre-trained natural language model (e.g., BERT, GPT, etc.) can be used to obtain the semantic vector of each factor, forming a set of factor features ME = {em1, em2, em3, ..., em...}. k ,…,em n},in Indicator m k The semantic vector, t represents the dimension of the semantic vector, and n represents the size of the set ME, that is, the number of semantic vectors of the factors.

[0095] In step S403, the set of factor features is input to a matrix generator to obtain a comparison matrix, wherein the matrix generator is trained using a method for training a matrix generator according to one aspect of this application.

[0096] Optionally, in step S403, the set of factor features can be input into a matrix generator to generate hierarchical scoring results for multiple factor feature pairs that are combinations of the factor features in the set of factor features, and a comparison matrix is ​​obtained based on the hierarchical scoring results of the multiple factor feature pairs. For example, the set of factor features ME = {em1, em2, em3, ..., em...} is... k ,…,em n Inputting this into the matrix generator will produce the following comparison matrix A:

[0097]

[0098] Where i represents the row number of the element in the matrix, j represents the column number of the element in the matrix, and a ii =1,a ij =1 / a ij ,n represents the number of factors (also the dimension of the matrix), a ij Features for each factor derived from the matrix generator output (em) i ,em j The score_level of the hierarchical evaluation result. ij The value of .

[0099] According to one or more embodiments of this application, by automatically constructing a contrast matrix based on a set of factor features using a trained matrix generator, the manual labor cost in the contrast matrix construction process can be saved, the efficiency and accuracy of using the analytic hierarchy process can be improved, and the application field of the analytic hierarchy process can be broadened.

[0100] Figure 5 A block diagram of a system for constructing a contrast matrix according to one or more embodiments of this application is shown.

[0101] like Figure 5 As shown, the system 500 for constructing the comparison matrix includes an acquisition unit 510 and a matrix generator 520.

[0102] The acquisition unit 510 is configured to acquire a set of factor features.

[0103] Optionally, the acquisition unit 510 can be configured to acquire the semantic vector of each of a plurality of factors associated with the target task, and to combine the semantic vectors of each of the plurality of factors to obtain a set of factor features. For example, the plurality of factors associated with the target task can be represented as a set M = {m1, m2, m3, ..., m...} k ,…,m n}, where m1, m2, m3, ..., m k ,…,m n Let M represent multiple factors, where n represents the size of the set M, i.e., the number of factors. Then, these multiple factors can be represented using natural language. For example, a pre-trained natural language model (e.g., BERT, GPT, etc.) can be used to obtain the semantic vector of each factor, forming a set of factor features ME = {em1, em2, em3, ..., em...}. k ,…,em n},in Indicator m k The semantic vector, t represents the dimension of the semantic vector, and n represents the size of the set ME, that is, the number of semantic vectors of the factors.

[0104] The matrix generator 520 is configured to process a set of factor features to obtain a contrast matrix, wherein the matrix generator 520 can be trained using a method for training a matrix generator according to one aspect of this application.

[0105] Optionally, the matrix generator 520 can be configured to process a set of factor features to generate hierarchical scoring results for multiple factor feature pairs that are pairwise combinations of factor features in the set of factor features, and to generate a comparison matrix based on the hierarchical scoring results of the multiple factor feature pairs. For example, the matrix generator 520 can process a set of factor features ME = {em1, em2, em3, ..., em...} k ,…,em n}, so as to obtain the comparison matrix A as shown in the above formula (10).

[0106] Figure 6 A flowchart of a hierarchical analysis method according to one or more embodiments of this application is shown.

[0107] like Figure 6 As shown, in step S601, a set of factor features of multiple factors associated with the target task is obtained.

[0108] Optionally, in step S601, the semantic vector of each of the multiple factors associated with the target task can be obtained, and the semantic vectors of each of the multiple factors can be combined to obtain a set of factor features. For example, the multiple factors associated with the target task can be represented as a set M = {m1, m2, m3, ..., m...} k ,…,m n}, where m1, m2, m3, ..., m k ,…,m n Let M represent multiple factors, where n represents the size of the set M, i.e., the number of factors. Then, these multiple factors can be represented using natural language. For example, a pre-trained natural language model (e.g., BERT, GPT, etc.) can be used to obtain the semantic vector of each factor, forming a set of factor features ME = {em1, em2, em3, ..., em...}. k ,…,em n},in Indicator m k The semantic vector, t represents the dimension of the semantic vector, and n represents the size of the set ME, that is, the number of semantic vectors of the factors.

[0109] In step S603, the set of factor features is input to a matrix generator to obtain a comparison matrix, wherein the matrix generator is trained using a method for training a matrix generator according to one aspect of this application.

[0110] Optionally, in step S603, the set of factor features can be input into a matrix generator to generate hierarchical scoring results for multiple factor feature pairs that are pairwise combinations of factor features in the set of factor features, and a comparison matrix is ​​obtained based on the hierarchical scoring results of the multiple factor feature pairs. For example, the set of factor features ME = {em1, em2, em3, ..., em...} is... k ,…,em n Inputting the data into the matrix generator will produce the comparison matrix A shown in formula (10) above.

[0111] In step S605, the comparison matrix is ​​processed to determine the priority of multiple factors.

[0112] Optionally, in step S605, the elements of each column in the comparison matrix can be summed to obtain the sum of the elements in each column, each element in the comparison matrix can be divided by the sum of the elements in the corresponding column to generate a column-averaged comparison matrix, and the average value of each row of the column-averaged comparison matrix can be determined as the priority of multiple factors.

[0113] As an example, after obtaining the comparison matrix A as shown in formula (10) above, the sum of the elements in each column j of the comparison matrix A can be obtained by summing the elements in each column j using the following formula (11):

[0114]

[0115] Where i represents the row number of the element in the matrix, j represents the column number of the element in the matrix, and a ij This represents the elements in the comparison matrix A, where n represents the number of factors (also the dimension of the matrix).

[0116] As an example, after obtaining the sum of the elements in each column, the column-averaged comparison matrix A' can be generated by dividing each element in the comparison matrix by the sum of the elements in the corresponding column using the following formula (12):

[0117]

[0118] Where i represents the row number of the element in the matrix, j represents the column number of the element in the matrix, and a ij ' represents an element in the comparison matrix A after column averaging, and n represents the number of factors (which is also the dimension of the matrix).

[0119] As an example, after obtaining the comparison matrix A' after column averaging, the average value p of each row element in the comparison matrix can be determined by the following formula (13). i Prioritize as multiple factors:

[0120]

[0121] Where i represents the row number of the element in the matrix, j represents the column number of the element in the matrix, and a ij ' represents an element in the comparison matrix A after column averaging, and n represents the number of factors (which is also the dimension of the matrix).

[0122] As an example, when obtaining the average value p of each row of elements... i Then, the priority of the n factors can be represented as a column vector P as follows:

[0123] P = [p1,

[0124] p2,

[0125] …,

[0126] p n ].

[0127] In step S607, the consistency of the comparison matrix is ​​checked using the priority of multiple factors, and the weights of the multiple factors are determined using the comparison matrix that passes the consistency check.

[0128] Optionally, in step S607, the maximum eigenvalue of the comparison matrix can be determined based on the priority of multiple factors; a consistency index of the comparison matrix can be determined based on the maximum eigenvalue and the number of factors; a random index of the comparison matrix can be determined based on its dimension; and a consistency check can be performed on the comparison matrix based on the consistency index and the random index. When the ratio of the consistency index to the random index of the comparison matrix is ​​less than a predetermined threshold, the consistency check can be considered passed.

[0129] As an example, after obtaining the column vector P representing the priorities of multiple factors, the largest eigenvalue λ of the comparison matrix A can be determined by the following formula (14). max :

[0130]

[0131] Where i represents the row number of the element in the matrix, n represents the number of factors (also the dimension of the matrix), and p i Let [AP] represent the i-th element in column vector P. i This represents the i-th element in [AP].

[0132] As an example, in determining the largest eigenvalue λ of the contrast matrix A max Then, the consistency index CI of the comparison matrix A can be determined using the following formula (15):

[0133]

[0134] Where n represents the number of factors (which is also the dimension of the matrix).

[0135] As an example, the random index RI of the comparison matrix A can also be determined based on the dimension n of the comparison matrix A, for example, by using a mathematical lookup table to determine the random index RI of the comparison matrix based on the dimension n of the comparison matrix A.

[0136] As an example, after determining the consistency index CI and the random index RI of the comparison matrix A, the consistency ratio CR of the comparison matrix A can be determined by the following formula (16):

[0137] CR = CI / RI Formula (16)

[0138] For example, when CR is less than 0.1, it can be determined that the comparison matrix passes the consistency test.

[0139] Optionally, in step S607, one or more of the following methods—arithmetic mean, geometric mean, and eigenvector method—can be used to determine the weights of multiple factors through a consistency check comparison matrix, thereby determining the hierarchical analysis results of the target task.

[0140] The following will further describe in detail the implementation of the hierarchical analysis method according to one or more embodiments of this application, taking the evaluation of supplier value as an example.

[0141] refer to Figure 1 The hierarchical analysis method model 100 shown has the following structure: the first layer, the problem overview PSD, can be "Which supplier is the most valuable?"; the second layer, the factor system, can include influence, reputation, execution ability, geographical location, etc.; and the third layer, the candidate decision, can include supplier 1, supplier 2, supplier 3, supplier 4, etc. For example, the 235 factors included in the second layer factor system, such as influence, reputation, execution ability, and geographical location, can be represented as a set M = {m1, m2, m3, ..., m...} k ,…,m 235}, where m1, m2, m3, ..., m k ,…,m 235 These represent 235 factors. Next, multiple factors can be represented using natural language. For example, a factor set M = {m1, m2, m3, ..., m...} can be obtained using a pre-trained natural language model (e.g., BERT, GPT, etc.). k ,…,m 235 The semantic vectors of the 235 factors in} constitute the set of factor features ME = {em1, em2, em3, ..., em k ,…,em 235},in Indicator m k The semantic vector, where 160 represents the dimension of the semantic vector.

[0142] For example, the set of factor features of the obtained 235 factors ME = {em1, em2, em3, ..., em3} can be used as a reference. k ,…,em 235 The input is fed into a matrix generator trained using the method for training a matrix generator according to one aspect of this application to generate a contrast matrix. Then, reference can be made to... Figure 6The steps of the illustrated analytic hierarchy process (AHP) involve processing a comparison matrix to determine the priority of 235 factors, performing a consistency check on the comparison matrix using the priority of these factors, and determining the weights of the 235 factors using the comparison matrix that passes the consistency check. Optionally, one or more of the arithmetic mean, geometric mean, and eigenvector method can be used to process the comparison matrix that passes the consistency check to determine the weights of the aforementioned 235 factors, thereby determining that the first-level problem summary PSD can be the AHP result for "Which supplier is the most valuable?", for example, the AHP result is that supplier 3 is the most valuable.

[0143] Figure 7 A block diagram of a hierarchical analysis system according to one or more embodiments of this application is shown.

[0144] like Figure 7 As shown, the hierarchical analysis system 70 includes a communication unit 710, a memory 720 (e.g., a non-volatile memory such as flash memory, ROM, hard disk drive, magnetic disk, optical disk, etc.), a processor 730, and a computer program 740 stored on the memory 720 and executable on the processor 730.

[0145] The communication unit 710 serves as a communication interface and is configured to establish a communication connection between the hierarchical analysis system 70 and external devices or networks (e.g., cloud servers, mobile terminals, etc.).

[0146] The memory 720 stores a computer program 740 that can be executed by the processor 730. Furthermore, the memory 720 may also store data generated by the processor 730 when executing the computer program (e.g., a set of factor features, a contrast matrix, etc.) and data or commands received from the outside via the communication unit 710.

[0147] Processor 730 is configured to execute computer program 740 to implement a hierarchical analysis method according to one or more embodiments of the present application.

[0148] This application can also be implemented as a computer-readable storage medium including instructions that, when executed, perform a method for training a matrix generator according to one aspect of this application. Alternatively, this application can also be implemented as a computer-readable storage medium including instructions that, when executed, perform a method for constructing a contrast matrix according to one aspect of this application. Furthermore, this application can also be implemented as a computer-readable storage medium including instructions that, when executed, perform a hierarchical analysis method according to one aspect of this application.

[0149] Where applicable, the various embodiments provided in this application may be implemented using hardware, software, or a combination of hardware and software. Furthermore, where applicable, without departing from the scope of this application, the various hardware and / or software components described herein may be combined into composite components comprising software, hardware, and / or both. Where applicable, without departing from the scope of this application, the various hardware and / or software components described herein may be divided into sub-components comprising software, hardware, or both. Additionally, where applicable, it is contemplated that software components may be implemented as hardware components, and vice versa.

[0150] The software (such as program code and / or data) according to this application can be stored on one or more computer storage media. It is also contemplated that the software identified herein can be implemented using one or more networked and / or otherwise general-purpose or special-purpose computers and / or systems. Where applicable, the order of the various steps described herein can be changed, combined into compound steps, and / or divided into sub-steps to provide the features described herein.

[0151] The embodiments and examples presented herein are provided to best illustrate embodiments of this application and its particular applications, thereby enabling those skilled in the art to implement and use this application. However, those skilled in the art will understand that the above description and examples are provided for ease of illustration and example only. The descriptions presented are not intended to cover all aspects of this application or to limit this application to the precise forms disclosed.

Claims

1. A method for training a matrix generator, characterized in that, The matrix generator includes a factor discrimination model and a hierarchical scoring model, and the method includes the following steps: Obtain the set of factor feature combinations, as well as the factor discrimination label and hierarchical scoring label for each factor feature combination; The first neural network model is trained using the set of factor feature combinations and the factor discrimination label of each factor feature combination as the first training set, and the trained first neural network model is updated using the first loss function to obtain the factor discrimination model. as well as The set of factor feature combinations and the hierarchical scoring label for each factor feature combination are used as the second training set to train the second neural network model, and the trained second neural network model is updated using the second loss function to obtain the hierarchical scoring model. The second loss function is constructed based at least on the reciprocal of the factor discrimination result output by the factor discrimination model and the hierarchical scoring result output by the trained second neural network model. Furthermore, the set of combinations of factor features includes: Acquire multiple factors that are relevant to the target task and represented in natural language; The semantic vector of each of the multiple factors is obtained through a pre-trained natural language model; The semantic vectors of each of the multiple factors are combined into a set of factor features; The factor features in the set of factor features are combined pairwise to generate multiple factor feature pairs; and The plurality of factor feature pairs are used as a set of factor feature combinations, wherein each factor feature combination includes the semantic vector of the first factor and the semantic vector of the second factor.

2. The method according to claim 1, wherein the matrix generator is used to generate a contrast matrix based on multiple input factors in the analytic hierarchy process.

3. The method according to claim 1, wherein obtaining the factor discrimination label for each factor feature combination includes: The factor discrimination label for each factor feature combination is obtained based on the importance of the first factor relative to the second factor.

4. The method according to claim 3, wherein obtaining the factor discrimination label for each factor feature combination further comprises: In response to determining that the first factor is more important than the second factor, the factor discrimination label is set to a first value; as well as In response to determining that the second factor is more important than the first factor, the factor discrimination label is set to a second value.

5. The method according to claim 4, wherein obtaining the hierarchical scoring label for each factor feature combination includes: The hierarchical scoring labels for each factor feature combination are obtained based on the relative importance of the first factor compared to the second factor.

6. The method of claim 5, wherein obtaining the hierarchical rating label for each factor feature combination further comprises: Determine a score to represent the importance of the first factor relative to the second factor; The hierarchical rating label is set to the rating in response to the first factor being more important than the second factor; as well as In response to the second factor being more important than the first factor, the hierarchical rating label is set to the reciprocal of the rating.

7. The method according to claim 1, wherein the first loss function is constructed based on the factor discrimination label of each factor feature combination and the factor discrimination result output by the trained first neural network model.

8. The method of claim 1, wherein updating the trained first neural network model using a first loss function comprises: The parameters of the trained first neural network model are adjusted using the first loss function until the first loss function is in a convergent state, so as to obtain the factor discrimination model.

9. The method according to claim 1, wherein the second loss function is constructed based on the hierarchical rating label of each factor feature combination, the factor discrimination result output by the factor discrimination model, the hierarchical rating result output by the trained second neural network model, and the reciprocal of the hierarchical rating result output by the trained second neural network model.

10. The method of claim 1, wherein updating the trained second neural network model using a second loss function comprises: The parameters of the trained second neural network model are adjusted using the second loss function until the second loss function converges to obtain the hierarchical scoring model.

11. A system for training a matrix generator, characterized in that, The matrix generator includes a factor discrimination model and a hierarchical scoring model; the system includes: The acquisition unit is configured to acquire a set of factor feature combinations and a factor discrimination label and a hierarchical scoring label for each factor feature combination; A first training unit is configured to train a first neural network model using the set of factor feature combinations and the factor discrimination label of each factor feature combination as a first training set, and to update the trained first neural network model using a first loss function to obtain the factor discrimination model; and The second training unit is configured to train the second neural network model using the set of factor feature combinations and the hierarchical rating labels for each factor feature combination as the second training set, and to update the trained second neural network model using the second loss function to obtain the hierarchical rating model. The second loss function is constructed based at least on the reciprocal of the factor discrimination result output by the factor discrimination model and the hierarchical scoring result output by the trained second neural network model. Furthermore, the acquisition unit is further configured to: Acquire multiple factors that are relevant to the target task and represented in natural language; The semantic vector of each of the multiple factors is obtained through a pre-trained natural language model; The semantic vectors of each of the multiple factors are combined into a set of factor features; The factor features in the set of factor features are combined pairwise to generate multiple factor feature pairs; and The plurality of factor feature pairs are used as a set of factor feature combinations, wherein each factor feature combination includes the semantic vector of the first factor and the semantic vector of the second factor.

12. The system of claim 11, wherein the matrix generator is used to generate a contrast matrix based on multiple input factors in the analytic hierarchy process.

13. The system of claim 11, wherein the acquisition unit is further configured to: The factor discrimination label for each factor feature combination is obtained based on the importance of the first factor relative to the second factor.

14. The system of claim 13, wherein the acquisition unit is further configured to: In response to determining that the first factor is more important than the second factor, the factor discrimination label is set to a first value; and In response to determining that the second factor is more important than the first factor, the factor discrimination label is set to a second value.

15. The system according to claim 11, wherein the acquisition unit is further configured to: The hierarchical scoring labels for each factor feature combination are obtained based on the relative importance of the first factor compared to the second factor.

16. The system of claim 15, wherein the acquisition unit is further configured to: Determine a score to represent the importance of the first factor relative to the second factor; In response to the first factor being more important than the second factor, the hierarchical rating label is set to the rating; and In response to the second factor being more important than the first factor, the hierarchical rating label is set to the reciprocal of the rating.

17. The system of claim 11, wherein the first loss function is constructed based on the factor discrimination label of each factor feature combination and the factor discrimination result output by the trained first neural network model.

18. The system of claim 11, wherein the first training unit is further configured to: The parameters of the trained first neural network model are adjusted using the first loss function until the first loss function is in a convergent state, so as to obtain the factor discrimination model.

19. The system of claim 11, wherein the second loss function is constructed based on the hierarchical rating label of each factor feature combination, the factor discrimination result output by the factor discrimination model, the hierarchical rating result output by the trained second neural network model, and the reciprocal of the hierarchical rating result output by the trained second neural network model.

20. The system of claim 11, wherein the second training unit is further configured to: The parameters of the trained second neural network model are adjusted using the second loss function until the second loss function converges to obtain the hierarchical scoring model.

21. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed, perform the method for training a matrix generator according to any one of claims 1-10.

22. A method for constructing a contrast matrix, characterized in that, The method includes the following steps: Obtain a set of factor features; and The set of factor features is input into a matrix generator to obtain a comparison matrix, wherein the matrix generator is trained using the method for training a matrix generator as described in any one of claims 1-10.

23. The method of claim 22, wherein obtaining the set of factor features comprises: Obtain the semantic vector of each of the multiple factors associated with the target task; as well as The semantic vectors of each of the multiple factors are combined to obtain a set of features of the factors.

24. The method of claim 22, wherein inputting the set of factor features into a matrix generator to obtain a comparison matrix comprises: The set of factor features is input into a matrix generator to generate hierarchical scoring results for multiple factor feature pairs, representing pairwise combinations of factor features within the set of factor features; and A comparison matrix is ​​generated based on the hierarchical scoring results of the multiple factor feature pairs.

25. A system for constructing a contrast matrix, characterized in that, The system includes: Acquisition unit, configured to acquire a set of factor features; and A matrix generator configured to process the set of factor features to obtain a contrast matrix, wherein the matrix generator is trained using the method for training a matrix generator as described in any one of claims 1-10.

26. The system of claim 25, wherein the acquisition unit is further configured to: Obtain the semantic vector of each of the multiple factors associated with the target task; and The semantic vectors of each of the multiple factors are combined to obtain a set of features of the factors.

27. The system of claim 25, wherein the matrix generator is further configured to: The set of factor features is processed to generate hierarchical scoring results for multiple factor feature pairs, representing pairwise combinations of factor features within the set of factor features; and A comparison matrix is ​​generated based on the hierarchical scoring results of the multiple factor feature pairs.

28. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed, perform the method for constructing a contrast matrix according to any one of claims 22-24.

29. A hierarchical analysis method, characterized in that, The hierarchical analysis method includes the following steps: Obtain a set of factor features for multiple factors associated with the target task; The set of factor features is input into a matrix generator to obtain a comparison matrix, wherein the matrix generator is trained using the method for training a matrix generator as described in any one of claims 1-10. The comparison matrix is ​​processed to determine the priority of the multiple factors; as well as The consistency of the comparison matrix is ​​checked using the priority of the multiple factors, and the weights of the multiple factors are determined using the comparison matrix that passes the consistency check.

30. The method of claim 29, wherein obtaining a set of factor features of multiple factors associated with the target task comprises: Obtain the semantic vector of each of the multiple factors associated with the target task; as well as The semantic vectors of each of the multiple factors are combined to obtain a set of features of the factors.

31. The method of claim 29, wherein inputting the set of factor features into a matrix generator to obtain a comparison matrix comprises: The set of factor features is input into a matrix generator to generate hierarchical scoring results for multiple factor feature pairs, representing pairwise combinations of factor features within the set of factor features; and A comparison matrix is ​​obtained based on the hierarchical scoring results of the multiple factor feature pairs.

32. The method of claim 29, wherein processing the comparison matrix to determine the priority of the plurality of factors comprises: The elements of each column in the comparison matrix are summed to obtain the sum of the elements in each column; Divide each element in the comparison matrix by the sum of the elements in the corresponding column to generate a column-averaged comparison matrix; as well as The average value of each row element in the comparison matrix of the column averages is determined as the priority of the multiple factors.

33. The method of claim 29, wherein performing a consistency check on the comparison matrix using the priority of the plurality of factors comprises: The maximum eigenvalue of the comparison matrix is ​​determined based on the priority of the multiple factors; The consistency index of the comparison matrix is ​​determined based on the largest eigenvalue and the number of factors in the comparison matrix. The random index of the comparison matrix is ​​determined based on its dimensions; and The consistency of the comparison matrix is ​​checked based on the consistency index and the random index.

34. The method of claim 33, wherein the method further comprises: The consistency test is deemed passed when the ratio of the consistency index to the random index in the comparison matrix is ​​less than a predetermined threshold.

35. A hierarchical analysis system, characterized in that, The system includes: Memory; A processor coupled to the memory; and A computer program stored in the memory and running on the processor, the execution of which causes the following operations: Obtain a set of factor features for multiple factors associated with the target task; The set of factor features is input into a matrix generator to obtain a comparison matrix, wherein the matrix generator is trained using the method for training a matrix generator as described in any one of claims 1-10. Process the comparison matrix to determine the priority of the plurality of factors; and The consistency of the comparison matrix is ​​checked using the priority of the multiple factors, and the weights of the multiple factors are determined using the comparison matrix that passes the consistency check.

36. The system of claim 35, wherein execution of the computer program further results in the following operations: Obtain the semantic vector of each of the multiple factors associated with the target task; and The semantic vectors of each of the multiple factors are combined to obtain a set of features of the factors.

37. The system of claim 35, wherein execution of the computer program further results in the following operations: The set of factor features is input into a matrix generator to generate hierarchical scoring results for multiple factor feature pairs, representing pairwise combinations of factor features within the set of factor features; and A comparison matrix is ​​obtained based on the hierarchical scoring results of the multiple factor feature pairs.

38. The system of claim 35, wherein execution of the computer program further results in the following operations: The elements of each column in the comparison matrix are summed to obtain the sum of the elements in each column; Divide each element in the comparison matrix by the sum of the elements in the corresponding column to generate a column-averaged comparison matrix; as well as The average value of each row element in the comparison matrix of the column averages is determined as the priority of the multiple factors.

39. The system of claim 35, wherein execution of the computer program further results in the following operations: The maximum eigenvalue of the comparison matrix is ​​determined based on the priority of the multiple factors; The consistency index of the comparison matrix is ​​determined based on the largest eigenvalue and the number of factors in the comparison matrix. The random index of the comparison matrix is ​​determined based on its dimensions; and The consistency of the comparison matrix is ​​checked based on the consistency index and the random index.

40. The system of claim 39, wherein execution of the computer program further results in the following operations: The consistency test is deemed passed when the ratio of the consistency index to the random index in the comparison matrix is ​​less than a predetermined threshold.

41. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed, perform the hierarchical analysis method according to any one of claims 29-34.

Citation Information

Patent Citations

  • Method and device for evaluating power interruptibility of sewage treatment enterprise

    CN113095687A

  • Intra-regional energy health degree evaluation method based on analytic hierarchy process

    CN113139701A