Multi-factor enterprise innovation ability dynamic evaluation method and system

Abstract

The multi-factor enterprise innovation capability dynamic evaluation method and system specifically comprise the following steps: S1, inputting multi-source heterogeneous data; s2, respectively calculating subjective and objective weight distribution factors, input and output factors and industry policy factors, and outputting scores Si and uncertainty indexes UIi of the factors; s3, performing conflict detection on the data obtained in the S2, and judging whether the difference between the Si exceeds a threshold value or not; when the threshold value is exceeded, an arbitration process is carried out; if not, entering S4 to calculate a final weight score, and outputting a result; s4, inputting the score Si and the uncertainty UIi of the three factors into a multi-objective optimization algorithm for processing; and S5, carrying out visual output. The method has the beneficial effects that the quality, namely the uncertainty, of each factor in the current calculation can be automatically sensed, an optimal weight combination is decided through an optimization algorithm, an intelligent arbitration process is started when the conclusion of the factors conflicts, and finally a robust, explainable and dynamically adaptive evaluation result is output.

Description

Technical Field

[0001] This invention relates to the field of enterprise evaluation, and more specifically to a method and system for dynamic evaluation of enterprise innovation capabilities using multiple factors. Background Technology

[0002] Existing enterprises utilize dynamic evaluation systems for enterprise innovation capabilities as a tool for comprehensive and real-time assessment of their innovation capabilities. This allows for the quantification and analysis of various aspects of enterprise innovation activities, reflecting the dynamic changes in enterprise innovation capabilities.

[0003] Traditional evaluation systems employ a static weighted average fusion model, which simply uses the arithmetic average of expert weights to process factors. This approach fails to address uncertainty and, since the final weights in traditional methods are static, they cannot dynamically reflect the evaluation results. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a method and system for dynamic evaluation of multi-factor enterprise innovation capabilities.

[0005] The objective of this invention is achieved through the following technical solution.

[0006] The method for dynamic evaluation of multi-factor enterprise innovation capability includes the following steps:

[0007] S1. Input multi-source heterogeneous data;

[0008] S2. Calculate the factor score S_i and uncertainty index UI_i for each enterprise;

[0009] S3. Perform conflict detection on the data obtained in S2 to determine whether the difference between S_i exceeds the threshold.

[0010] When the threshold is exceeded, an arbitration process is initiated, and the post-arbitration analysis and diagnostic process is output.

[0011] If the threshold is not exceeded, proceed to S4 to calculate the final weighted score and output the result;

[0012] S4. Input the scores S_i and uncertainties UI_i of each factor into the multi-objective optimization algorithm for processing, and output a Pareto optimal weight combination [W1, W2, W3, W4] through the multi-objective optimization algorithm.

[0013] S5. Perform visualization output.

[0014] The factors in S2 include: subjective and objective weight allocation factors, input-output factors, and industry policy factors;

[0015] The results obtained during the calculation of the subjective and objective weight allocation factors are the weight vector W_sub and the uncertainty UI_sub;

[0016] The results obtained in the calculation of input-output factors are the efficiency value S_io and the uncertainty UI_io;

[0017] The results obtained in the calculation of industry policy factors are the policy score S_pol and uncertainty UI_pol;

[0018] The results of the subjective and objective weight allocation factors, input-output factors and industry policy factors in S2 are standardized to obtain the factor score S_i and uncertainty index UI_i for each factor.

[0019] The calculation method for subjective and objective weighting factors includes the following steps:

[0020] A1. Collect evaluation data and obtain the judgment matrix. The value of the judgment matrix represents the importance value of indicator i relative to indicator j as determined by the evaluators.

[0021] A2. Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector. After normalizing the eigenvector, obtain the weight vector.

[0022] A3. Calculate the consistency index CI and consistency ratio CR using the largest eigenvalue of the judgment matrix, and use the consistency test to judge the judgment matrix in the collected evaluation data.

[0023] When the consistency ratio CR = CI / RI is greater than the preset threshold, the consistency test is deemed to have failed; when the consistency ratio CR is less than or equal to the preset threshold, the consistency test is deemed to have passed.

[0024] A4. Use a cloud model to calibrate the collected evaluation data and output the final weight vector.

[0025] The specific steps for A4 include:

[0026] A41. For the i-th indicator, extract the weight values ​​that all evaluators assign to the indicator from the weight vector of all evaluators to form a sample dataset.

[0027] A42. For each sample dataset obtained in A41, use the inverse cloud generator algorithm to calculate three mathematical features of the opinion cloud model of the corresponding sample dataset. These mathematical features include:

[0028] Expectation Ex is used to determine the average opinion of the group of evaluators regarding the weight of this indicator.

[0029] Entropy En is used to measure the degree of disagreement among evaluators;

[0030] Hyperentropy He, the entropy of entropy, represents the stability of the degree of divergence;

[0031] A43, regarding entropy En i Superentropy He i Sample size n of evaluators i Normalization is performed using the following formula.

[0032]

[0033] Among them, En norm He norm and n norm These are the normalized values ​​of entropy, hyperentropy, and the sample size of the evaluators, respectively. ref He ref n ref These represent the entropy, hyperentropy, and normalized reference values ​​for the evaluator sample size, respectively; the evaluator sample size is n. i , which is the number of evaluators who score or rate the i-th indicator;

[0034] A44. Construct the feasibility factor and define the credibility factor Cred based on the parameters obtained in A43. i Credibility factor i as follows,

[0035] Cred i =exp(-(α·En) norm +β·He norm +γ·(1-n norm ))))

[0036] Among them, α, β, and γ are adjustable trade-off parameters that reflect the importance of different sources of uncertainty;

[0037] A45. The credibility of the subjective weights is corrected, and the corrected subjective weights W sub_corr,i The formula is,

[0038]

[0039] Among them, Ax i The original subjective weights of the evaluators for indicator i; ∑ j (Ax j ×Cred j The sum of all weights after indicator adjustment is denoted as ).

[0040] A46. Calculate the data stability coefficient Stab. i Data stability coefficient Stab i The calculation formula is as follows:

[0041]

[0042] Where, σ i Let i be the standard deviation of index i; Let ε be the mean of index i; ε is a small constant to prevent the denominator from being zero; A47, calculate the data quality coefficient DS. i Data quality coefficient DS i The calculation formula is as follows:

[0043] DS i =ω1Cmpl i +ω2Stab i ω1+ω2=1

[0044] Wherein, ω1 and ω2 are preset values;

[0045] A48. Obtaining the objective weight W obj,i Based on data quality coefficient DS i Adjust the objective weights;

[0046] The weight values ​​are calculated using both the entropy method and the CRITIC method. The objective weight W is obtained by weighted averaging of the weight values ​​obtained by the two methods. obj,i ;

[0047] For objective weight W obj,i Adjustments are made to obtain the adjusted objective weight W' obj,i The formula for adjusting the objective weights is:

[0048]

[0049] Among them, W' obj,i The adjusted objective weights; ∑ j (W obj,j ×DS j The sum of the adjusted weights of all indicators is represented by A49. A comprehensive analysis of subjective and objective weights is achieved through linear fusion, using the following fusion formula:

[0050]

[0051] in, W is the combined subjective and objective weighted value after fusion. sub_corr,i W is the subjective weight value. sub_corr,j The subjective weights have been modified using the credibility factor; W obj,i For objective weighting; W obj,j The adjusted objective weights; λ∈[0,1] controls the proportion of objective weights;

[0052] A410, Output the corrected weight vector.

[0053] The calculation methods for industry policy factors specifically include:

[0054] B1. Collect the full text of policy documents related to the target industry to form an original policy text library;

[0055] B2. Construct a policy graph. Predefine a closed, hierarchical policy keyword graph. The policy graph contains multi-level classifications. Assign a basic policy intensity value l to each node in the graph. The basic policy intensity value l is solidified in the system after being reviewed by the evaluators.

[0056] B3. Use NLP technology to analyze the policy text library, calculate the semantic similarity between the policy text library and each node in the policy graph, and calculate the total strength of the original policy according to the formula I_raw=∑(fn*I);

[0057] B4. Introduce the time decay coefficient γ(t) and the scale correction coefficient β. The formula for calculating the scale correction coefficient β is as follows:

[0058] β = log 10 (1+S_i / S_avg)

[0059] In the formula, S_i represents the company's own size data, and S_avg represents the industry average data;

[0060] B5. Calculate the policy factors. The formula for calculating the policy factors is as follows:

[0061] S_p=I_raw*β*γ(t)

[0062] In the formula, S_p is the policy factor; I_raw is the original total policy strength; β is the size correction coefficient; and γ(t) is the time decay coefficient.

[0063] In B3, NLP technology uses lightweight keyword extraction and semantic matching algorithms or pre-trained deep learning models to calculate semantic similarity.

[0064] The calculation method for input-output factors includes the following specific steps:

[0065] C1. Differentiate the performance indicators and classify the resources that the company wants to minimize as input indicators;

[0066] C2. Divide the results that the company wants to maximize into output indicators;

[0067] C3. Use the SBM-DEA model to calculate and obtain the company's superefficiency value;

[0068] C4. Normalize the superefficiency values ​​of different enterprises to obtain the input-output factors of the enterprises.

[0069] The specific steps of the multi-objective optimization algorithm in S4 include:

[0070] D1. Input is the factor scores S_i and uncertainty index UI_i of N enterprises on subjective and objective weight allocation factors, input-output factors and industry policy factors;

[0071] D2. Initialization algorithm: Randomly generate multiple weight combinations that satisfy the condition that the sum of the weights is 1 to form the initial population;

[0072] D3. Evaluate the initial population and calculate the function value of the variance of the final score of all enterprises corresponding to each weight combination in the population and the function value of the weighted sum of weights and uncertainties.

[0073] D4. Calculate the non-dominated ordination and crowding degree. Based on the objective function value obtained in D3, divide the population individuals into different non-dominated levels to obtain the non-dominated ordination. Calculate the crowding degree of individuals within the same level to measure their distribution density in the target space and ensure population diversity.

[0074] D5. Based on the non-dominated ordering and crowding degree obtained in D4, select excellent individuals as parents, and generate offspring through crossover and mutation operations to form a new generation of population.

[0075] D6. Repeat D2-D4 until the preset algebra or convergence condition is reached, obtain the final Pareto optimal solution set, select the optimal weight scheme from the Pareto optimal solution set as the final weight and output it.

[0076] The specific steps for selecting the optimal weight scheme from the Pareto optimal solution set in D6 include:

[0077] Three strategies are set;

[0078] The first strategy is to directly select the weight combination that maximizes the variance of the company's score;

[0079] The second strategy is to directly select the weight combination that minimizes the weighted sum of uncertainty;

[0080] The third strategy is to adopt a compromise planning method;

[0081] Determine the positive and negative ideal solutions, calculate the Euclidean distance between each solution in the Pareto set and the positive and negative ideal solutions, calculate the relative proximity of each solution, and select the solution with the largest relative proximity as the final weight.

[0082] The specific steps of S3 include:

[0083] S31. Conflict detection: Calculate the absolute difference between each pair of factor scores of the same enterprise, compare the absolute difference with the dynamic threshold obtained by calculating the overall dispersion of the scores of all enterprises this time, determine the enterprise whose score difference exceeds the dynamic threshold as a conflict state, trigger the arbitration signal, lock the corresponding factor whose score difference exceeds the dynamic threshold, and enter S32.

[0084] S32, Level 1 arbitration: Among the factors whose score difference exceeds the dynamic threshold, the factor with the highest uncertainty index is obtained. The factor with the highest uncertainty index is weighted and penalized to obtain the weight combination of the weight penalty. The total score of the enterprise after the weight penalty is recalculated. The conflict monitoring in S31 is re-performed to determine whether the score difference of the factor corresponding to the conflict state in S1 has been reduced to within the dynamic threshold.

[0085] When the dynamic threshold is reached, the process ends and the result is output.

[0086] If the dynamic threshold is not reached, proceed to step S33 to carry out the second-level arbitration process.

[0087] S33, Level 2 Arbitration: The data of the enterprises in conflict is output as a diagnostic report, which is then handed over to a human for final adjudication to adjust the data. The final score is manually entered, and this data case is recorded as training data for the Level 1 Arbitration model.

[0088] The specific steps for applying a weight penalty to the factor with the highest uncertainty index in S32 include: calculating the penalty strength using the penalty model formula, obtaining the penalized weight factors, and redistributing the weights based on the penalized weight factors. The penalty model formula is as follows:

[0089] P = k * (1 - exp(-c * UI_i))

[0090] Where P represents the penalty intensity, UI_i represents the uncertainty value of the factor, k represents the maximum penalty intensity, and c represents the penalty sensitivity.

[0091] A system for dynamic evaluation of multi-factor firm innovation capabilities, and methods for conducting such evaluations, including:

[0092] The preprocessing module is used to preprocess multi-source heterogeneous data;

[0093] The subjective and objective weight allocation factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the subjective and objective weight allocation factor.

[0094] The input-output factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the input-output factors;

[0095] The industry policy factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the industry policy factors.

[0096] The dynamic weight optimization module is used to run multi-objective optimization algorithms to obtain the Pareto optimal weight combination;

[0097] The conflict arbitration module is used to run the arbitration process.

[0098] The visualization module is used to visualize the data.

[0099] The beneficial effects of this invention are as follows: This solution, based on a multi-objective dynamic optimization and three-level conflict arbitration multi-source factor collaborative decision-making mechanism, breaks through the traditional static weighted average fusion mode, upgrading the existing evaluation system from a calculator to an intelligent decision-making body, and constructing an intelligent evaluation method and system with "perception-decision-arbitration" capabilities. It can automatically perceive the quality (i.e., uncertainty) of each factor's current calculation, determine the optimal weight combination through optimization algorithms, and initiate an intelligent arbitration process when factor conclusions conflict, ultimately outputting a robust, interpretable, and dynamically adaptable evaluation result. Attached Figure Description

[0100] Figure 1 This is a schematic diagram illustrating the calculation process of the subjective and objective weight allocation factors;

[0101] Figure 2 This is a schematic diagram of the calculation process for input-output factors;

[0102] Figure 3 This is a schematic diagram of the structure of a multi-factor dynamic evaluation system for enterprise innovation capabilities. Detailed Implementation

[0103] The technical solution of the present invention will be further described below through specific embodiments.

[0104] Example

[0105] The method for dynamic evaluation of multi-factor enterprise innovation capability includes the following steps:

[0106] S1. Input multi-source heterogeneous data;

[0107] S2. Calculate the factor score S_i and uncertainty index UI_i for each enterprise;

[0108] S3. Perform conflict detection on the data obtained in S2 to determine whether the difference between S_i exceeds the threshold.

[0109] When the threshold is exceeded, an arbitration process is initiated, and the post-arbitration analysis and diagnostic process is output.

[0110] If the threshold is not exceeded, proceed to S4 to calculate the final weighted score and output the result;

[0111] S4. Input the scores S_i and uncertainties UI_i of each factor into the multi-objective optimization algorithm for processing, and output a Pareto optimal weight combination [W1, W2, W3, W4] through the multi-objective optimization algorithm.

[0112] S5. Perform visualization output.

[0113] The factors in S2 include: subjective and objective weight allocation factors, input-output factors, and industry policy factors;

[0114] The results obtained during the calculation of the subjective and objective weight allocation factors are the weight vector W_sub and the uncertainty UI_sub;

[0115] The results obtained in the calculation of input-output factors are the efficiency value S_io and the uncertainty UI_io;

[0116] The results obtained in the calculation of industry policy factors are the policy score S_pol and uncertainty UI_pol;

[0117] The results of the subjective and objective weight allocation factors, input-output factors and industry policy factors in S2 are standardized to obtain the factor score S_i and uncertainty index UI_i for each factor.

[0118] like Figure 1 As shown, the calculation method for the subjective and objective weight allocation factor includes the following steps:

[0119] A1. Collect evaluation data and obtain the judgment matrix. The value of the judgment matrix represents the importance value of indicator i relative to indicator j as determined by the evaluators.

[0120] A2. Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector. After normalizing the eigenvector, obtain the weight vector.

[0121] A3. Calculate the consistency index CI and consistency ratio CR using the largest eigenvalue of the judgment matrix, and use the consistency test to judge the judgment matrix in the collected evaluation data.

[0122] When the consistency ratio CR = CI / RI is greater than the preset threshold, the consistency test is deemed to have failed; when the consistency ratio CR is less than or equal to the preset threshold, the consistency test is deemed to have passed.

[0123] A4. Use a cloud model to calibrate the collected evaluation data and output the final weight vector.

[0124] Preferably, in this embodiment, when collecting evaluation data in A1, the indicators in the enterprise technological innovation evaluation indicator system are combined in pairs, and experts are asked to use the 1-9 scale method to determine their relative importance.

[0125] For example, regarding "R&D investment" and "number of patents," experts need to answer: "How important do you consider 'R&D investment' to be relative to 'number of patents'?" The values ​​are: 1 (equally important), 3 (slightly important), 5 (significantly important), 7 (very important), 9 (extremely important), with 2, 4, 6, and 8 being intermediate values. The final output is a judgment matrix A for each expert on the entire indicator system. The values ​​in the judgment matrix represent the importance that experts consider indicator i to be relative to indicator j.

[0126] The specific steps for A4 include:

[0127] A41. For the i-th indicator, extract the weight values ​​that all evaluators assign to the indicator from the weight vector of all evaluators to form a sample dataset.

[0128] A42. For each sample dataset obtained in A41, use the inverse cloud generator algorithm to calculate three mathematical features of the opinion cloud model of the corresponding sample dataset. These mathematical features include:

[0129] Expectation Ex is used to determine the average opinion of the group of evaluators regarding the weight of this indicator.

[0130] Entropy En is used to measure the degree of disagreement among evaluators;

[0131] Hyperentropy He, the entropy of entropy, represents the stability of the degree of divergence;

[0132] A43, regarding entropy En i Superentropy He i Sample size n of evaluators i Normalization is performed using the following formula.

[0133]

[0134] Among them, En norm He norm and n norm These are the normalized values ​​of entropy, hyperentropy, and the sample size of the evaluators, respectively. ref He ref n ref These are normalized reference values ​​for entropy, hyperentropy, and evaluator sample size, respectively, which can be set based on historical samples or industry benchmarks; the evaluator sample size is n. i , which is the number of evaluators who score or rate the i-th indicator;

[0135] A44. Construct the feasibility factor and define the credibility factor Cred based on the parameters obtained in A43. i Credibility factor i as follows,

[0136] Cred i =exp(-(α·En) norm +β·He norm +γ·(1-n norm ))))

[0137] Wherein, α, β, and γ are adjustable trade-off parameters reflecting the importance of different sources of uncertainty; preferably, the parameters are set to α = 1.0, β = 1.0, and γ = 1.0 by default, with values ​​ranging from [0.5, 3.0]. When En or He increases, Cred... i Decreases; as the sample size n increases, Cred... i The function range is increasing; the range of the function is (0,1], continuous and numerically stable.

[0138] A45. The credibility of the subjective weights is corrected, and the corrected subjective weights W sub_corr,i The formula is,

[0139]

[0140] Among them, Ax i The original subjective weights of the evaluators for indicator i; ∑ j (Ax j ×Cred j ) is the sum of all weights after the indicator is adjusted.

[0141] This step corrects for the bias in expert subjective weighting under high uncertainty conditions, using a method based on Cred. i The multiplicative modulation mechanism, let Ax i This represents the original subjective weights of indicator i assigned by experts. This correction process can effectively suppress the influence of high-entropy, low-sample indicators and improve the robustness of the weight distribution.

[0142] A46. Calculate the data stability coefficient Stab. i Data stability coefficient Stab i The calculation formula is as follows:

[0143]

[0144] Where, σ i Let i be the standard deviation of index i; Let be the mean of index i; ε is a small constant to prevent the denominator from being zero.

[0145] A47. Calculate the data quality coefficient DSi Data quality coefficient DS i The calculation formula is as follows:

[0146] DS i =ω1Cmpl i +ω2Stab i ω1+ω2=1

[0147] Wherein, ω1 and ω2 are preset values.

[0148] A48. Obtaining the objective weight W obj,i Based on data quality coefficient DS i Adjust the objective weights;

[0149] The weight values ​​are calculated using both the entropy method and the CRITIC method. The objective weight W is obtained by weighted averaging of the weight values ​​obtained by the two methods. obj,i ;

[0150] For objective weight W obj,i Adjustments are made to obtain the adjusted objective weight W' obj,i The formula for adjusting the objective weights is:

[0151]

[0152] Among them, W' obj,i The adjusted objective weights; ∑ j (W obj,j ×DS j The sum of the adjusted weights of all indicators is represented by A49. A comprehensive analysis of subjective and objective weights is achieved through linear fusion, using the following fusion formula:

[0153]

[0154] in, W is the combined subjective and objective weighted value after fusion. sub_corr,i W is the subjective weight value. sub_corr,j The subjective weights have been modified using the credibility factor; W obj,j For objective weighting; W' obj,j The adjusted objective weights; λ∈[0,1] controls the proportion of objective weights; in this embodiment, λ=0.6 is taken by default.

[0155] A410, Output the corrected weight vector.

[0156] Compared with existing technologies, the calculation method of subjective and objective weight allocation factors in this scheme has the following technical improvements.

[0157] 1. This solution adopts a deep integration of AHP, consistency verification, and cloud model, filling the gap in uncertainty handling of traditional methods.

[0158] 2. Traditional AHP only addresses the logical consistency of individual expert judgments, i.e., the CR test, while this solution further addresses the issues of consensus and reliability of expert group opinions.

[0159] 3. Quantitative modeling of uncertainty is used. Unlike existing technologies, which simply take the arithmetic average of expert weights, this method uses three numerical features (Ex, En, He) of the cloud model to accurately quantify the central tendency (Ex), dispersion (En), and stability of dispersion (He) of group opinions.

[0160] The calculation methods for industry policy factors specifically include:

[0161] B1. Collect the full text of policy documents related to the target industry to form an original policy text library;

[0162] B2. Construct a policy graph. Predefine a closed, hierarchical policy keyword graph. The policy graph contains multi-level classifications. Assign a basic policy intensity value l to each node in the graph. The basic policy intensity value l is solidified in the system after being reviewed by the evaluators.

[0163] B3. Use NLP technology to analyze the policy text library, calculate the semantic similarity between the policy text library and each node in the policy graph, and calculate the total strength of the original policy according to the formula I_raw=∑(fn*I);

[0164] B4. Introduce the time decay coefficient γ(t) and the scale correction coefficient β. The time decay coefficient γ(t) is used to characterize the degree to which the policy's influence decays over time. Its calculation formula is: γ(t) = exp(-λ·Δt), where λ is the decay rate constant, which can be set according to the policy's legal effect level; Δt is the time span (in years) from the policy's issuance to the present. The calculation formula for the scale correction coefficient β is as follows...

[0165] β = log 10 (1+S_i / S_avg)

[0166] In the formula, S_i represents the company's own size data, and S_avg represents the industry average data;

[0167] B5. Calculate the policy factors. The formula for calculating the policy factors is as follows:

[0168] S_p=I_raw*β*γ(t)

[0169] In the formula, S_p is the policy factor; I_raw is the original total policy strength; β is the size correction coefficient; and γ(t) is the time decay coefficient.

[0170] In B3, NLP technology uses lightweight keyword extraction and semantic matching algorithms or pre-trained deep learning models to calculate semantic similarity.

[0171] Furthermore, the time decay coefficient γ(t) is used to characterize the degree of decay of policy influence over time, and its calculation adopts an exponential decay model: γ(t) = exp(-λ·Δt). Here, Δt is the time span since the policy's release (in years), and λ is the decay rate constant, which can be determined by system presets or expert review based on the policy's legal validity level, planning period, etc. This coefficient ensures that the evaluation model better reflects the latest policy direction. In this embodiment, the B2 policy map includes primary, secondary, and tertiary classifications.

[0172] like Figure 2 As shown, the calculation method for input-output factors includes the following specific steps:

[0173] C1. Differentiate the performance indicators and classify the resources that the company wants to minimize as input indicators;

[0174] C2. Divide the results that the company wants to maximize into output indicators;

[0175] C3. Use the SBM-DEA model to calculate and obtain the company's superefficiency value;

[0176] C4. Normalize the superefficiency values ​​of different enterprises to obtain the input-output factors of the enterprises.

[0177] The specific steps of the multi-objective optimization algorithm in S4 include:

[0178] D1. Input is the factor scores S_i and uncertainty index UI_i of N enterprises on subjective and objective weight allocation factors, input-output factors and industry policy factors;

[0179] D2. Initialization algorithm: Randomly generate multiple weight combinations that satisfy the condition that the sum of the weights is 1 to form the initial population;

[0180] D3. Evaluate the initial population and calculate the function value of the variance of the final score of all enterprises corresponding to each weight combination in the population and the function value of the weighted sum of weights and uncertainties.

[0181] D4. Calculate the non-dominated ordination and crowding degree. Based on the objective function value obtained in D3, divide the population individuals into different non-dominated levels to obtain the non-dominated ordination. Calculate the crowding degree of individuals within the same level to measure their distribution density in the target space and ensure population diversity.

[0182] D5. Based on the non-dominated ordering and crowding degree obtained in D4, select excellent individuals as parents, and generate offspring through crossover and mutation operations to form a new generation of population.

[0183] D6. Repeat D2-D4 until the preset algebra or convergence condition is reached, obtain the final Pareto optimal solution set, select the optimal weight scheme from the Pareto optimal solution set as the final weight and output it.

[0184] The specific steps for selecting the optimal weight scheme from the Pareto optimal solution set in D6 include:

[0185] Three strategies are set;

[0186] The first strategy is to directly select the weight combination that maximizes the variance of the company's score;

[0187] The second strategy is to directly select the weight combination that minimizes the weighted sum of uncertainty;

[0188] The third strategy is to adopt a compromise planning method;

[0189] Determine the positive and negative ideal solutions, calculate the Euclidean distance between each solution in the Pareto set and the positive and negative ideal solutions, calculate the relative proximity of each solution, and select the solution with the largest relative proximity as the final weight.

[0190] The specific steps of S3 include:

[0191] S31. Conflict Detection: Calculate the absolute difference between each pair of factors scores for the same company. Simultaneously, calculate the dynamic threshold based on the overall dispersion of scores for all companies in this evaluation. Specifically, calculate the standard deviation σ or interquartile range (IQR) of the scores of all companies on the same factor in this evaluation. The dynamic threshold is T = k·σ or T = k·IQR, where k is an adjustable preset coefficient. Compare the absolute difference with the dynamic threshold T. Companies with differences exceeding the dynamic threshold are identified as being in conflict, triggering an arbitration signal. At the same time, the corresponding factor with a difference exceeding the dynamic threshold is locked, and the process proceeds to S32.

[0192] S32, Level 1 arbitration: Among the factors whose score difference exceeds the dynamic threshold, the factor with the highest uncertainty index is obtained. The factor with the highest uncertainty index is weighted and penalized to obtain the weight combination of the weight penalty. The total score of the enterprise after the weight penalty is recalculated. The conflict monitoring in S31 is re-performed to determine whether the score difference of the factor corresponding to the conflict state in S1 has been reduced to within the dynamic threshold.

[0193] When the dynamic threshold is reached, the process ends and the result is output.

[0194] If the dynamic threshold is not reached, proceed to step S33 to carry out the second-level arbitration process.

[0195] S33, Level 2 Arbitration: The data of the enterprises in conflict is output as a diagnostic report, which is then handed over to a human for final adjudication to adjust the data. The final score is manually entered, and this data case is recorded as training data for the Level 1 Arbitration model.

[0196] Furthermore, the absolute interpolation is compared with the dynamic threshold calculated based on the overall dispersion of the scores of all enterprises in this study. In this step, the dynamic threshold calculated based on the overall dispersion of the scores of all enterprises in this study can also be replaced by a threshold determined manually based on experience.

[0197] The dynamic threshold is calculated based on the overall dispersion of the scores of all enterprises. In the acquisition process, after obtaining the scores of the four factors of all factors in the database, the standard deviation of each factor score (avg-std) or interquartile range (avg-iqr) is calculated to reflect the overall difference level of all enterprises on a certain factor, i.e., dispersion. The final threshold is k*(avg-std) or k*(avg-std), where k is an adjustable coefficient that is set manually.

[0198] The specific steps for applying a weight penalty to the factor with the highest uncertainty index in S32 include: calculating the penalty strength using the penalty model formula, obtaining the penalized weight factors, and redistributing the weights based on the penalized weight factors. The penalty model formula is as follows:

[0199] P = k * (1 - exp(-c * UI_i))

[0200] Where P represents the penalty intensity, UI_i represents the uncertainty value of the factor, k represents the maximum penalty intensity, and c represents the penalty sensitivity.

[0201] The calculation example is as follows:

[0202] In this example, k is 0.4 and c is 4; these values ​​are also suitable for general calculations and are optimal settings.

[0203] Suppose that the weight of a factor is W_i = 0.3 and UI_i = 0.8.

[0204] Then P = 0.4 * (1 - exp(-4 * 0.8)) ≈ 0.38.

[0205] The new weight after the weight penalty will become 0.3*(1-0.38)=0.186.

[0206] Based on this, the weights are redistributed. The reduced weight of this factor (0.3 - 0.186 = 0.114 in the example) is proportionally added to the other three factors with lower uncertainty, ensuring that the total weight sum is 1.

[0207] A system for dynamic evaluation of multi-factor firm innovation capabilities, and methods for conducting such evaluations, including:

[0208] The preprocessing module is used to preprocess multi-source heterogeneous data;

[0209] The subjective and objective weight allocation factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the subjective and objective weight allocation factor.

[0210] The input-output factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the input-output factors;

[0211] The industry policy factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the industry policy factors.

[0212] The dynamic weight optimization module is used to run multi-objective optimization algorithms to obtain the Pareto optimal weight combination;

[0213] The conflict arbitration module is used to run the arbitration process.

[0214] The visualization module is used to visualize the data.

[0215] Preferably, the final output of the visualization module includes a science and technology innovation capability report. In this embodiment, an external large language model is used to achieve the function of directly outputting the science and technology innovation capability report, which further facilitates users to intuitively understand the data.

[0216] like Figure 3 As shown in the diagram, the data input layer of this system is used to input multi-source heterogeneous data. After processing by the data well preprocessing module, the data is then handed over to the subjective and objective weight allocation factor calculation module, the input-output factor calculation module, and the industry policy factor calculation module for individual factor calculation. The dynamic weight optimization module and the conflict arbitration module are located in the factor weight allocation layer and are used to allocate factor weights. The visualization module is located in the result output layer and is used to visualize the output results.

[0217] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for dynamic evaluation of multi-factor enterprise innovation capability, characterized in that, The specific steps include: S1. Input multi-source heterogeneous data; S2. Calculate the factor score S_i and uncertainty index UI_i for each enterprise; S3. Perform conflict detection on the data obtained in S2 to determine whether the difference between S_i exceeds the threshold. When the threshold is exceeded, an arbitration process is initiated, and the post-arbitration analysis and diagnostic process is output. If the threshold is not exceeded, proceed to S4 to calculate the final weighted score and output the result; S4. Input the scores S_i and uncertainties UI_i of each factor into the multi-objective optimization algorithm for processing, and output a Pareto optimal weight combination [W1, W2, W3, W4] through the multi-objective optimization algorithm. S5. Perform visualization output.

2. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 1, characterized in that... The factors in S2 include: subjective and objective weight allocation factors, input-output factors, and industry policy factors; The results obtained during the calculation of the subjective and objective weight allocation factors are the weight vector W_sub and the uncertainty UI_sub; The results obtained in the calculation of input-output factors are the efficiency value S_io and the uncertainty UI_io; The results obtained in the calculation of industry policy factors are the policy score S_pol and uncertainty UI_pol; The results of the subjective and objective weight allocation factors, input-output factors and industry policy factors in S2 are standardized to obtain the factor score S_i and uncertainty index UI_i for each factor.

3. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 2, characterized in that, The calculation method for subjective and objective weighting factors includes the following steps: A1. Collect evaluation data and obtain the judgment matrix. The value of the judgment matrix represents the importance value of indicator i relative to indicator j as determined by the evaluators. A2. Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector. After normalizing the eigenvector, obtain the weight vector. A3. Calculate the consistency index CI and consistency ratio CR using the largest eigenvalue of the judgment matrix, and use the consistency test to judge the judgment matrix in the collected evaluation data. When the consistency ratio CR = CI / RI is greater than the preset threshold, the consistency test is deemed to have failed; when the consistency ratio CR is less than or equal to the preset threshold, the consistency test is deemed to have passed. A4. Use a cloud model to calibrate the collected evaluation data and output the final weight vector.

4. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 3, characterized in that, The specific steps for A4 include: A41. For the i-th indicator, extract the weight values ​​that all evaluators assign to the indicator from the weight vector of all evaluators to form a sample dataset. A42. For each sample dataset obtained in A41, use the inverse cloud generator algorithm to calculate three mathematical features of the opinion cloud model of the corresponding sample dataset. These mathematical features include: Expectation Ex is used to determine the average opinion of the group of evaluators regarding the weight of this indicator. Entropy En is used to measure the degree of disagreement among evaluators; Hyperentropy He, the entropy of entropy, represents the stability of the degree of divergence; A43, regarding entropy En i Superentropy He i Sample size n of evaluators i Normalization is performed using the following formula. Among them, En norm He norm and n norm These are the normalized values ​​of entropy, hyperentropy, and the sample size of the evaluators, respectively. ref He ref n ref These represent the entropy, hyperentropy, and normalized reference values ​​for the evaluator sample size, respectively; the evaluator sample size is n. i , which is the number of evaluators who score or rate the i-th indicator; A44. Construct the feasibility factor and define the credibility factor Cred based on the parameters obtained in A43. i Credibility factor i as follows, Credit i =exp(-(α·En norm +β·He norm +γ·(1-n norm )))) Among them, α, β, and γ are adjustable trade-off parameters that reflect the importance of different sources of uncertainty; A45. The credibility of the subjective weights is corrected, and the corrected subjective weights W sub_corr,i The formula is, Among them, Ax i The original subjective weights of the evaluators for indicator i; ∑ j (Ax j ×Cred j The sum of all weights after indicator adjustment is denoted as ). A46. Calculate the data stability coefficient Stab. i Data stability coefficient Stab i The calculation formula is as follows: Where, σ i Let i be the standard deviation of index i; Let be the mean of index i; ε is a small constant to prevent the denominator from being zero. A47. Calculate the data quality coefficient DS i Data quality coefficient DS i The calculation formula is as follows: DS i =ω1Cmpl i +ω2Stab i ,ω1+ω2=1 Wherein, ω1 and ω2 are preset values; A48. Obtaining the objective weight W obj,i Based on data quality coefficient DS i Adjust the objective weights; The weight values ​​are calculated using both the entropy method and the CRITIC method. The objective weight W is obtained by weighted averaging of the weight values ​​obtained by the two methods. obj,i ; For objective weight W obj,i Adjustments are made to obtain the adjusted objective weight W' obj,i The formula for adjusting the objective weights is: Among them, W' obj,i The adjusted objective weights; ∑ j (W obj,j ×DS j ) represents the sum of all indicators after adjustment of their weights; A49. A combination of subjective and objective weights is achieved through linear fusion. The fusion formula is as follows: in, W is the combined subjective and objective weighted value after fusion. sub_corr,j The subjective weights modified by the credibility factor; W' obj,j The adjusted objective weights; λ∈[0,1] controls the proportion of objective weights; A410, Output the corrected weight vector.

5. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 3, characterized in that, The calculation methods for industry policy factors specifically include: B1. Collect the full text of policy documents related to the target industry to form an original policy text library; B2. Construct a policy graph. Predefine a closed, hierarchical policy keyword graph. The policy graph contains multi-level classifications. Assign a basic policy intensity value l to each node in the graph. The basic policy intensity value l is solidified in the system after being reviewed by the evaluators. B3. Use NLP technology to analyze the policy text library, calculate the semantic similarity between the policy text library and each node in the policy graph, and calculate the total strength of the original policy according to the formula I_raw=∑(fn*I); B4. Introduce the time decay coefficient γ(t) and the scale correction coefficient β. The formula for calculating the time decay coefficient γ(t) is: γ(t) = exp(-λ·Δt), where λ is the decay rate constant and Δt is the time span since the policy was issued. The formula for calculating the scale correction coefficient β is as follows: β=log 10 (1+S_i / S_avg) In the formula, S_i represents the company's own size data, and S_avg represents the industry average data; B5. Calculate the policy factors. The formula for calculating the policy factors is as follows: S_p=I_raw*β*γ(t) In the formula, S_p is the policy factor; I_raw is the original total policy strength; β is the size correction coefficient; and γ(t) is the time decay coefficient.

6. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 2, characterized in that, The specific steps of the multi-objective optimization algorithm in S4 include: D1. Input is the factor scores S_i and uncertainty index UI_i of N enterprises on subjective and objective weight allocation factors, input-output factors and industry policy factors; D2. Initialization algorithm: Randomly generate multiple weight combinations that satisfy the condition that the sum of the weights is 1 to form the initial population; D3. Evaluate the initial population and calculate the function value of the variance of the final score of all enterprises corresponding to each weight combination in the population and the function value of the weighted sum of weights and uncertainties. D4. Calculate the non-dominated ordination and crowding degree. Based on the objective function value obtained in D3, divide the population individuals into different non-dominated levels to obtain the non-dominated ordination. Calculate the crowding degree of individuals within the same level to measure their distribution density in the target space and ensure population diversity. D5. Based on the non-dominated ordering and crowding degree obtained in D4, select excellent individuals as parents, and generate offspring through crossover and mutation operations to form a new generation of population. D6. Repeat D2-D4 until the preset algebra or convergence condition is reached, obtain the final Pareto optimal solution set, select the optimal weight scheme from the Pareto optimal solution set as the final weight and output it.

7. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 6, characterized in that, The specific steps for selecting the optimal weight scheme from the Pareto optimal solution set in D6 include: Three strategies are set; The first strategy is to directly select the weight combination that maximizes the variance of the company's score; The second strategy is to directly select the weight combination that minimizes the weighted sum of uncertainty; The third strategy is to adopt a compromise planning method; Determine the positive and negative ideal solutions, calculate the Euclidean distance between each solution in the Pareto set and the positive and negative ideal solutions, calculate the relative proximity of each solution, and select the solution with the largest relative proximity as the final weight.

8. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 1, characterized in that, The specific steps of S3 include: S31. Conflict detection: Calculate the absolute difference between each pair of factors scores of the same enterprise. The dynamic threshold is determined as follows: Calculate the standard deviation σ of the scores of all enterprises on the same factor, and let the dynamic threshold T = k·σ, where k is a preset coefficient greater than 0; compare the absolute difference with the dynamic threshold T, determine the enterprise whose score difference exceeds the dynamic threshold as a conflict state, trigger the arbitration signal, and lock the corresponding factor whose score difference exceeds the dynamic threshold, and proceed to S32. S32, Level 1 arbitration: Among the factors whose score difference exceeds the dynamic threshold, the factor with the highest uncertainty index is obtained. The factor with the highest uncertainty index is weighted and penalized to obtain the weight combination of the weight penalty. The total score of the enterprise after the weight penalty is recalculated. The conflict monitoring in S31 is re-performed to determine whether the score difference of the factor corresponding to the conflict state in S1 has been reduced to within the dynamic threshold. When the dynamic threshold is reached, the process ends and the result is output. If the dynamic threshold is not reached, proceed to step S33 to carry out the second-level arbitration process. S33, Level 2 Arbitration: The data of the enterprises in conflict is output as a diagnostic report, which is then handed over to a human for final adjudication to adjust the data. The final score is manually entered, and this data case is recorded as training data for the Level 1 Arbitration model.

9. The method for dynamic evaluation of multi-factor enterprise innovation capability according to claim 8, characterized in that, The specific steps for applying a weight penalty to the factor with the highest uncertainty index in S32 include: calculating the penalty strength using the penalty model formula, obtaining the penalized weight factors, and redistributing the weights based on the penalized weight factors. The penalty model formula is as follows: P = k * (1 - exp(-c * UI_i)) Where P represents the penalty intensity, UI_i represents the uncertainty value of the factor, k represents the maximum penalty intensity, and c represents the penalty sensitivity.

10. A system for dynamic evaluation of multi-factor enterprise innovation capability, used to run the method for dynamic evaluation of multi-factor enterprise innovation capability as described in any one of claims 1-9, characterized in that, include: The preprocessing module is used to preprocess multi-source heterogeneous data; The subjective and objective weight allocation factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the subjective and objective weight allocation factor. The input-output factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the input-output factors; The industry policy factor calculation module is used to calculate the factor score S_i and uncertainty index UI_i of the industry policy factors. The dynamic weight optimization module is used to run multi-objective optimization algorithms to obtain the Pareto optimal weight combination; The conflict arbitration module is used to run the arbitration process. The visualization module is used to visualize the data.

Images