Bid evaluation monitoring system and method based on network platform

Abstract

The invention discloses a bid evaluation monitoring system and method based on a network platform, and aims to optimize a bid evaluation process through a plurality of advanced technologies. The system adopts a multi-factor identity verification and behavior biological recognition technology to ensure the identity safety of bid evaluation participants; simulating the performance of the bidding file in different situations by using a mixed reality situation simulator and a generative adversarial network technology; automatically auditing the bidding file through a natural language processing technology, and carrying out stylized analysis on design elements in combination with an art cognition algorithm; and dynamically optimizing an expert scoring process by using a fuzzy logic decision tree technology, identifying a scoring deviation and providing real-time feedback. The system also adopts an adaptive learning mechanism, continuously optimizes a scoring strategy, and improves the efficiency, transparency and compliance of the bid evaluation process.

Description

Technical Field

[0001] The present invention relates to the field of information technology and network technology, and in particular to a bid evaluation monitoring system and method based on a network platform. Background Art

[0002] With the development of information technology and network technology, the bid evaluation system based on the network platform has gradually replaced the traditional offline bid evaluation method. However, the existing bid evaluation system based on the network platform still has many problems and cannot fully meet the needs of modern bid evaluation, especially in terms of information security, multi-party collaboration, and data intelligent analysis.

[0003] First, the existing online bid evaluation system has loopholes in identity authentication. Traditional online platform bid evaluation systems usually rely only on usernames and passwords for identity authentication, which is at risk of being attacked and forged, and cannot effectively guarantee the authenticity and uniqueness of the identities of evaluation experts and bidders. Although multi-factor authentication has been applied, it has not yet been combined with behavioral biometrics technology, making the system vulnerable to complex attacks.

[0004] Secondly, the existing bid evaluation system lacks effective means to simulate and analyze the performance of bid documents in complex situations. Traditional systems only provide basic text and data analysis and cannot deeply simulate the performance of bid documents in different market situations and user experiences. This lack of situational simulation analysis means that evaluation experts lack reference when evaluating the actual effect of bid documents, which may lead to deviations and misjudgments in bid evaluation results.

[0005] Third, when existing natural language processing technology is applied in the bid evaluation system, it is mainly used for basic text review and keyword matching, and cannot deeply analyze the implicit information in the bid documents, especially the identification of regulatory compliance and potential risks. In addition, the existing system has limited artistic cognition ability and cannot accurately evaluate the performance of design elements in the bid documents in terms of artistic style and user experience, resulting in a lack of comprehensive data support for evaluation experts when making decisions.

[0006] Fourth, the existing system lacks an effective dynamic adjustment mechanism in terms of scoring fairness and consistency. The existing scoring system usually uses fixed rules for scoring, and cannot perform real-time analysis and optimization based on the scoring path and logic of the review experts. Fuzzy logic decision tree technology is rarely used in existing systems, and the system cannot dynamically adjust the scoring strategy based on real-time data, resulting in the scoring results being affected by subjective bias, which reduces the fairness of the bid evaluation.

[0007] In summary, the existing bid evaluation system based on the network platform has many deficiencies in identity authentication, complex situation simulation, natural language processing, art cognition and scoring optimization, etc. Therefore, how to provide a bid evaluation monitoring system and method based on the network platform is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0008] One purpose of the present invention is to propose a bid evaluation monitoring system and method based on a network platform, which monitors and optimizes the bid evaluation process in real time, identifies potential risks and generates bid evaluation reports through multi-factor identity authentication, mixed reality scenario simulation, natural language processing, art cognitive algorithm, fuzzy logic decision tree and automated data analysis. The system not only improves the efficiency and fairness of bid evaluation, but also continuously optimizes the scoring strategy through an adaptive learning mechanism to ensure the accuracy and foresight of the bid evaluation process.

[0009] The bidding evaluation monitoring method based on a network platform according to an embodiment of the present invention comprises the following steps:

[0010] S1. Publish the bidding information on the online platform and use multi-factor authentication and behavioral biometrics technology to authenticate the bidders and evaluation experts;

[0011] S2. After the bidder submits the bid documents, a virtual bid evaluation environment is created through a mixed reality scenario simulator combined with generative adversarial network technology to simulate the performance of the bid documents in different extreme market scenarios and user interaction environments and conduct virtual user experience tests;

[0012] S3. After receiving the bidding documents, we will use natural language processing technology to automatically review the content of the documents, identify and mark possible illegal clauses or non-compliant content, and compare this information with the preset regulations and policy rule base to generate a preliminary review report;

[0013] S4. During the review process, an art recognition algorithm is implanted and combined with an artificial intelligence stylized analysis network to perform artistic style recognition and emotional response analysis on the design elements in the bidding documents and generate an artistic aesthetic evaluation report;

[0014] S5. Use fuzzy logic decision tree technology to analyze experts’ scoring paths and decision logic in real time, and dynamically adjust and optimize the uncertainty factors involved in the scoring process;

[0015] S6. Use automated data analysis tools to monitor and analyze the review experts’ scoring data in real time, identify potential scoring biases, and provide an immediate feedback mechanism to allow experts to adjust their scoring strategies based on the suggestions;

[0016] S7. After the bid evaluation is completed, the bid evaluation result report is automatically generated and the results are securely stored using data encryption technology;

[0017] S8. Comprehensively consider the evaluation experts’ scoring data, the content of the bidding documents and their performance in the virtual environment to generate a multi-dimensional bidding evaluation decision support report and publish it to relevant parties.

[0018] Optionally, the S1 includes the following steps:

[0019] S11. When posting bidding information on the online platform, the identity information of the bidder and the evaluation experts shall be initially verified through identity verification, which includes the combination of user name and password verification and dynamic verification code verification sent by SMS or email;

[0020] S12. After the initial verification is completed, the behavioral biometrics module is started to collect behavioral feature data of the bidder and the evaluation experts. The specific behavioral features include keystroke dynamics, mouse movement trajectory and touch screen gestures;

[0021] S13, comparing the collected behavior feature data with the behavior feature model stored in the database, and calculating the matching degree R, where R represents the matching degree between the behavior feature and the preset model, and the matching degree is calculated based on the behavior feature vector, and the value range of R is [0,1], and the higher the value, the higher the matching degree;

[0022] S14. According to the result of the matching degree R, if the R value is greater than or equal to the preset threshold value T, the identity verification is deemed to be passed, and the bidder and the evaluation expert are allowed to continue the operation; if the R value is lower than T, the identity verification fails, and the user is prompted to re-verify;

[0023] S15. After identity authentication is passed, a unique session identifier SID is generated for each successfully authenticated user. The SID is used to track the integrity and continuity of all operations of bidders and evaluation experts during the bid evaluation process.

[0024] Optionally, S2 includes the following steps:

[0025] S21. After the bidder submits the bid document, extract the multi-dimensional element data in the bid document, including technical indicators T j 、Design parameters D k Functional module F l and user interaction logic U m , convert it into a feature vector set X;

[0026] S22, generate a series of market variables M through generative adversarial network technology i Each market scenario is formed by coupling market demand fluctuation V1, competitor strategy change V2, and economic environment fluctuation V3 to form a composite scenario model C i , specifically:

[0027]

[0028] Among them, λ p1 , q2 , r3 and λ s4 are the coupling coefficients related to technical indicators, design parameters, functional modules and user interaction logic in the market context, respectively. i Represents a composite scenario model under market scenarios;

[0029] S23, composite situation model C i Introducing the mixed reality scenario simulator, which combines virtual reality and augmented reality technologies to build an immersive virtual bid evaluation environment that includes elements of the real world and the virtual world, allowing evaluation experts to interact with the bid proposals in a highly simulated environment;

[0030] S24. In the mixed reality scenario simulator, according to the composite scenario model C i Conduct user interaction simulation on the bidding scheme, including the user's behavior path P u 、Interactive feedback u , and emotional response E u , and transform these data into the interaction behavior matrix B i ;

[0031] S25, the interaction behavior matrix B is analyzed through a recurrent neural network i Analyze and identify the temporal characteristics of user behavior and calculate the user experience score S u At the same time, according to the composite situation model C i Calculate the market adaptability score S m ;

[0032] S26. In the mixed reality scenario simulator, the user experience score S in different market scenarios u and market adaptability score S m , calculate the comprehensive score result R s :

[0033]

[0034] Among them, α i and β i is the weight factor in each situation, max(S m ) and max(S u ) is the maximum value of the score, which is used for standardization;

[0035] S27, according to the comprehensive scoring results R s, automatically generates a virtual user experience test report, which includes interactive behavior analysis, emotional response summary, market adaptability assessment and potential risk warning, and gives optimization suggestions, which are then fed back to the bidder for adjustment;

[0036] S28. Update and optimize the market scenarios in the generative adversarial network model and mixed reality scenario simulator by continuously accumulating and analyzing bid evaluation data.

[0037] Optionally, S3 includes the following steps:

[0038] S31. After receiving the bidding documents, the text content in the bidding documents is structured and key content elements are extracted, including technical description T d 、Contract Clauses C t 、Legal Statement s And the bidder's self-description materials a Etc., and convert it into the corresponding text feature vector set X t ;

[0039] S32. Use natural language processing technology to process the text feature vector set X t Perform semantic analysis and convert text content into a high-dimensional vector space representation V through word embedding technology t , calculate the similarity S between the text content and the terms in the preset violation and non-compliance rule base sim ;

[0040] S33, according to the similarity score S sim The results of the bidding documents are used to identify and mark the possible illegal clauses or non-compliant contents in the bidding documents, and the similarity scores exceeding the preset threshold T sim The clauses are automatically marked as high-risk content;

[0041] S34. Compare the marked high-risk content with the preset regulations and policy rule base R p Compare and determine whether it complies with relevant regulations through the rule matching algorithm, and generate the rule matching degree M r :

[0042]

[0043] Among them, γ i is the importance weight of each regulation or policy rule, S sim (i) is the similarity score of the i-th rule, M r Indicates the overall rule matching degree;

[0044] S35, according to the rule matching degree M rBased on the results, a preliminary audit report is generated, which includes the marked violations and non-compliant clauses, rule matching analysis and recommended modification measures, and the report is sent to the review experts for further evaluation;

[0045] S36. After generating the preliminary audit report, continuously optimize and update the rule base R based on feedback and historical data p , and adjust the similarity calculation and rule matching algorithms.

[0046] Optionally, S4 includes the following steps:

[0047] S41. Extract design elements from bidding documents X d , including graphic design d , Color scheme C s , Layout L o and brand logo B i , converted into a multidimensional feature vector x i , each element represents the characteristic value of the design attribute in different dimensions;

[0048] S42, using the art recognition algorithm, through the multi-layer perceptron and self-attention mechanism, the feature vector x i For in-depth processing:

[0049] S43, identify the importance of design elements in various dimensions through the self-attention mechanism and generate a weighted feature vector a i :

[0050]

[0051] Among them, Q i =W q ·x i , K i =W k ·x i 、V i =W v ·x i , respectively, the linear transformation matrices of query, key, and value, d k is the dimension of the key vector, W q , W k , W v is the weight matrix of the self-attention mechanism, Softmax is used to calculate the attention weight, a i represents the weighted feature vector;

[0052] S45. Use a multi-layer perceptron to perform nonlinear mapping on the weighted feature vector to obtain a preliminary style feature vector z i :

[0053] z i=ReLU(W1·a i +b1);

[0054] Among them, W1 and b1 are the weight and bias of MLP, and ReLU is the activation function, which maps the weighted features nonlinearly;

[0055] S46, the style feature vector z i With known art style vector S j Compare and calculate the similarity R between each design element and the art style sij :

[0056]

[0057] Among them, S fj represents the vector of the jth known artistic style, R sij Indicates the design element z i With art style S j The similarity is calculated by dot product and normalized;

[0058] S47. Use the generative adversarial network to further adjust the style similarity matrix R s , generate the optimized style feature matrix F adj :

[0059]

[0060] Among them, G j is the generator function, η is the style adjustment coefficient, z j is the introduced noise vector, tanh represents the hyperbolic tangent function;

[0061] S48, the optimized style feature matrix F adj Input the emotional computing model to calculate the emotional response vector e of each design element in the user's emotional space i :

[0062]

[0063] Among them, θ j is the emotional weight coefficient, reflecting the influence of different artistic styles on emotional response; σ j is a nonlinear sentiment activation function;

[0064] S49, combined with style consistency index C s and the emotional response vector e i , calculate the sentiment consistency score E of the bidding document c :

[0065]

[0066] in, is the emotional vector of the user experience standard, reflecting the user's expected emotional response; λ is the emotional consistency adjustment factor, adjusting the balance between style and emotion; C s It is a style consistency index, which measures the coordination of design elements in the dimension of artistic style;

[0067] S410, dynamically adjust the weight W1 and bias b1 of the multilayer perceptron and the generator parameter G in the generative adversarial network according to the data feedback after each review j and the sentiment activation function σ j .

[0068] Optionally, S5 includes the following steps:

[0069] S51, extract the scoring path data from the review experts' scoring process in real time, and record the input feature X of each scoring node p =[x p1 ,x p2 ,...,x pn ] and the corresponding scoring decision D p , where x pi represents the i-th scoring feature value, D p represents the output of the scoring decision;

[0070] S52. For each scoring feature value x pi Fuzzy processing is performed to convert it into a fuzzy set μ pi (x pi ), membership function of fuzzy set:

[0071]

[0072] Among them, c pi is the center of the eigenvalue, σ pi is the expansion coefficient, and the membership function is used to describe the degree of belonging of the scoring feature value in the fuzzy set;

[0073] S53, expansion coefficient σ pi The choice of affects the shape and width of the membership function, which is usually adjusted dynamically according to the distribution of the scoring features to meet the requirements of different review tasks. The expansion coefficient σ pi Optimization:

[0074]

[0075] Among them, Q represents the number of samples in the historical data, x pi,q Represents the score feature value of the qth sample

[0076] S54, based on the fuzzified eigenvalues, generate the fuzzy rule set R using fuzzy logic reasoning j :

[0077] R j :IFμ p1 (x p1 )ANDμ p2 (x p2 )AND...ANDμ pn (x pn )THEND p =f j (X p );

[0078] Among them, f j (X p ) indicates that the rule R j The associated output scoring decision, f j (X p The specific form of ) depends on the definition of fuzzy logic rules, which is usually a linear or nonlinear function. It is optimized by analyzing historical scoring data and dynamically adjusting the weight w according to historical data. i :

[0079]

[0080] Among them, D p,q For the scoring decision in the historical scoring data, the weight w of the feature is updated by weighting the results of the historical scoring data i ;

[0081] S55, based on fuzzy rule set R j Construct fuzzy decision tree T f In the decision tree, each node represents a fuzzy rule, and each leaf node corresponds to a decision output D fj ;

[0082] S56. Further optimize the fuzzy decision tree T f The output of the rule is adjusted by adjusting the confidence C j , the influence of dynamic optimization fuzzy rules:

[0083]

[0084] Among them, w i To adjust the feature weights based on historical scoring data, the weights w are dynamically optimized using historical scoring data. i , the final score result S is generated by confidence weighting f :

[0085]

[0086] S57, each scoring process, according to the optimized fuzzy decision tree T f 'Adjust the evaluation experts' scores in real time and generate the final score result S f ;

[0087] S58. After each review, the rule base and membership function parameter c of the fuzzy logic decision tree are updated by analyzing historical scoring data and expert feedback. pi and σ pi , and continuously optimize the fuzzy rules and confidence adjustment strategies.

[0088] Optionally, the S6 includes the following steps:

[0089] S61. Collect scoring data from the review experts in real time during their scoring process s ={D p1 ,D p2 ,...,D pn}, where D pi Indicates the scoring result of the i-th scoring node, collects scoring data, and simultaneously collects multi-dimensional feature data X related to scoring s ={X p1 ,X p2 ,...,X pn}, including the behavioral characteristics of review experts, scoring time characteristics, and historical scoring trends;

[0090] S62. Use automated data analysis tools and real-time stream processing technology to analyze the scoring data D s and multidimensional feature data X s Perform dynamic monitoring, for each scoring node D pi , calculate the dynamic mean D pi (t) and the time-weighted standard deviation

[0091]

[0092] Among them, α is the smoothing factor, β and γ are the time decay coefficients, and Δt is the time interval;

[0093] S63. Using multi-dimensional feature data X s and rating data D s , combined with the deep learning model to conduct in-depth analysis of the scoring deviation, the deviation analysis model calculates the scoring deviation vector ΔD pi (t) and its contribution rate;

[0094] S64, based on the scoring deviation vector ΔD pi (t), combined with the scoring feature data X pi(t)), identify the source of the deviation through the causal inference model and calculate the attribution weight w of the deviation k :

[0095]

[0096] Among them, g k (X pi,k (t)) is the characteristic response function related to the deviation, and the characteristic response function is optimized through the training data set;

[0097] S65. Generate instant feedback report F r (t), including deviation nodes, deviation sources and adjustment suggestions, calculate the deviation adjustment value A for each scoring node pi (t):

[0098]

[0099] Among them, λ is the adjustment strength coefficient, m is the feature dimension, and the feedback adjustment value A pi (t) Provide score adjustment suggestions and allow review experts to make real-time modifications;

[0100] S66. Dynamically adjust the scoring deviation detection and feedback model during the scoring process, and adjust the parameters α, β, γ and the attribution weight w of the monitoring model through continuous learning and optimization. k , and dynamically adjust the threshold τ'(t) for scoring deviation detection:

[0101]

[0102] Among them, η is the learning rate, n is the number of scoring nodes, and the scoring strategy is continuously optimized through the adaptive learning mechanism.

[0103] The bidding evaluation monitoring system based on the network platform includes:

[0104] An identity verification module, which is used to verify the identity of bidders and review experts using multi-factor identity verification and behavioral biometrics when posting bidding information on the online platform;

[0105] The mixed reality scenario simulation module is used to create a virtual bid evaluation environment after the bidder submits the bid document, combining the generative adversarial network technology to simulate the performance of the bid document in different extreme market scenarios and user interaction environments and conduct virtual user experience testing;

[0106] The automatic review module is used to automatically review the content of the bidding documents using natural language processing technology, identify and mark possible illegal clauses or non-compliant content, and compare them with the preset regulations and policy rule base to generate a preliminary review report;

[0107] Artistic cognition and style analysis module, which is used to identify the artistic style and analyze the emotional response of the design elements in the bidding documents in combination with the stylized analysis network of artificial intelligence during the review process, and generate an artistic aesthetic evaluation report for the review experts' reference;

[0108] The scoring optimization module is used to use fuzzy logic decision tree technology to analyze the expert's scoring path and decision logic in real time when the review experts are scoring, and dynamically adjust and optimize the uncertainty factors involved in the scoring process; the real-time monitoring module is used to monitor and analyze the review experts' scoring data in real time through automated data analysis tools, identify potential scoring deviations, and provide an instant feedback mechanism to adjust the scoring strategy;

[0109] The result generation and storage module is used to generate a bid evaluation result report after the bid evaluation is completed, and to store the results securely through data encryption technology.

[0110] The beneficial effects of the present invention are:

[0111] (1) The present invention proposes a bid evaluation monitoring system and method based on a network platform. Through multi-factor identity authentication and behavioral biometrics technology, the authenticity of the identities of bidders and evaluation experts is effectively guaranteed, avoiding the risk of false identities participating in bid evaluation. By introducing a mixed reality scenario simulator and generative adversarial network technology, the system can simulate the performance of bidding documents in different market scenarios and user interaction environments in a virtual environment, providing evaluation experts with more comprehensive evaluation data and significantly improving the accuracy and scientificity of bid evaluation results.

[0112] (2) The present invention proposes a technical method for automatically reviewing bidding documents using natural language processing technology, effectively identifying and marking illegal clauses and non-compliant content, ensuring the legality and compliance of the bid evaluation process, while reducing the time and error rate of manual review, and further improving the efficiency of bid evaluation.

[0113] (3) The present invention proposes a method for generating an artistic aesthetic evaluation report by combining an artistic cognitive algorithm to identify the artistic style and analyze the emotional response of the design elements in the bidding documents, thereby improving the artistic evaluation dimension in the bidding evaluation process and ensuring that the decision-making of the evaluation experts in terms of design aesthetics is more scientific and reasonable.

[0114] (4) The present invention proposes to use fuzzy logic decision tree technology to analyze the scoring path and decision logic of review experts in real time, dynamically adjust and optimize the uncertainty factors involved in the scoring process, further improve the fairness and consistency of the scoring, and ensure the objectivity and reliability of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0116] Figure 1 A flowchart of a bidding evaluation monitoring system and method based on a network platform proposed by the present invention;

[0117] Figure 2 A schematic diagram of a simulation of a bid evaluation environment in which a mixed reality scenario simulator is combined with a generative adversarial network technology in the present invention;

[0118] Figure 3 This is a schematic diagram of combining the art recognition algorithm to perform artistic style recognition and emotional response analysis in the present invention. DETAILED DESCRIPTION

[0119] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0120] refer to Figure 1-3 , the bidding evaluation monitoring method based on the network platform includes the following steps:

[0121] S1. Publish the bidding information on the online platform and use multi-factor authentication and behavioral biometrics technology to authenticate the bidders and evaluation experts;

[0122] S2. After the bidder submits the bid documents, a virtual bid evaluation environment is created through a mixed reality scenario simulator combined with generative adversarial network technology to simulate the performance of the bid documents in different extreme market scenarios and user interaction environments and conduct virtual user experience tests;

[0123] S3. After receiving the bidding documents, we will use natural language processing technology to automatically review the content of the documents, identify and mark possible illegal clauses or non-compliant content, and compare this information with the preset regulations and policy rule base to generate a preliminary review report;

[0124] S4. During the review process, an art recognition algorithm is implanted and combined with an artificial intelligence stylized analysis network to perform artistic style recognition and emotional response analysis on the design elements in the bidding documents and generate an artistic aesthetic evaluation report;

[0125] S5. Use fuzzy logic decision tree technology to analyze experts’ scoring paths and decision logic in real time, and dynamically adjust and optimize the uncertainty factors involved in the scoring process;

[0126] S6. Use automated data analysis tools to monitor and analyze the review experts’ scoring data in real time, identify potential scoring biases, and provide an immediate feedback mechanism to allow experts to adjust their scoring strategies based on the suggestions;

[0127] S7. After the bid evaluation is completed, the bid evaluation result report is automatically generated and the results are securely stored using data encryption technology;

[0128] S8. Comprehensively consider the evaluation experts’ scoring data, the content of the bidding documents and their performance in the virtual environment to generate a multi-dimensional bidding evaluation decision support report and publish it to relevant parties.

[0129] In this implementation, S1 includes the following steps:

[0130] S11. When posting bidding information on the online platform, the identity information of the bidder and the evaluation experts shall be initially verified through identity verification, which includes the combination of user name and password verification and dynamic verification code verification sent by SMS or email;

[0131] S12. After the initial verification is completed, the behavioral biometrics module is started to collect behavioral feature data of the bidder and the evaluation experts. The specific behavioral features include keystroke dynamics, mouse movement trajectory and touch screen gestures;

[0132] S13, comparing the collected behavior feature data with the behavior feature model stored in the database, and calculating the matching degree R, where R represents the matching degree between the behavior feature and the preset model, and the matching degree is calculated based on the behavior feature vector, and the value range of R is [0,1], and the higher the value, the higher the matching degree;

[0133] S14. According to the result of the matching degree R, if the R value is greater than or equal to the preset threshold value T, the identity verification is deemed to be passed, and the bidder and the evaluation expert are allowed to continue the operation; if the R value is lower than T, the identity verification fails, and the user is prompted to re-verify;

[0134] S15. After identity authentication is passed, a unique session identifier SID is generated for each successfully authenticated user. The SID is used to track the integrity and continuity of all operations of bidders and evaluation experts during the bid evaluation process.

[0135] In this implementation, S2 includes the following steps:

[0136] S21. After the bidder submits the bid document, extract the multi-dimensional element data in the bid document, including technical indicators T j 、Design parameters D k Functional module F l and user interaction logic U m , convert it into a feature vector set X;

[0137] S22, generate a series of market variables M through generative adversarial network technology i Each market scenario is formed by coupling market demand fluctuation V1, competitor strategy change V2, and economic environment fluctuation V3 to form a composite scenario model C i , specifically:

[0138]

[0139] Among them, λ p1 , q2 , r3 and λ s4 are the coupling coefficients related to technical indicators, design parameters, functional modules and user interaction logic in the market context, respectively. i Represents a composite scenario model under market scenarios;

[0140] S23, composite situation model C i Introducing the mixed reality scenario simulator, which combines virtual reality and augmented reality technologies to build an immersive virtual bid evaluation environment that includes elements of the real world and the virtual world, allowing evaluation experts to interact with the bid proposals in a highly simulated environment;

[0141] S24. In the mixed reality scenario simulator, according to the composite scenario model C i Conduct user interaction simulation on the bidding scheme, including the user's behavior path P u 、Interactive feedback u , and emotional response E u , and transform these data into the interaction behavior matrix B i ;

[0142] S25, the interaction behavior matrix B is analyzed through a recurrent neural network i Analyze and identify the temporal characteristics of user behavior and calculate the user experience score S u At the same time, according to the composite situation model C i Calculate the market adaptability score S m ;

[0143] S26. In the mixed reality scenario simulator, the user experience score S in different market scenarios u and market adaptability score S m , calculate the comprehensive score result R s :

[0144]

[0145] Among them, α iand β i is the weight factor in each situation, max(S m ) and max(S u ) is the maximum value of the score, which is used for standardization;

[0146] S27, according to the comprehensive scoring results R s , automatically generates a virtual user experience test report, which includes interactive behavior analysis, emotional response summary, market adaptability assessment and potential risk warning, and gives optimization suggestions, which are then fed back to the bidder for adjustment;

[0147] S28. Update and optimize the market scenarios in the generative adversarial network model and mixed reality scenario simulator by continuously accumulating and analyzing bid evaluation data.

[0148] In this implementation, S3 includes the following steps:

[0149] S31. After receiving the bidding documents, the text content in the bidding documents is structured and key content elements are extracted, including technical description T d 、Contract Clauses C t 、Legal Statement s And the bidder's self-description materials a Etc., and convert it into the corresponding text feature vector set X t ;

[0150] S32. Use natural language processing technology to process the text feature vector set X t Perform semantic analysis and convert text content into a high-dimensional vector space representation V through word embedding technology t , calculate the similarity S between the text content and the terms in the preset violation and non-compliance rule base sim ;

[0151] S33, according to the similarity score S sim The results of the bidding documents are used to identify and mark the possible illegal clauses or non-compliant contents in the bidding documents, and the similarity scores exceeding the preset threshold T sim The clauses are automatically marked as high-risk content;

[0152] S34. Compare the marked high-risk content with the preset regulations and policy rule base R p Compare and determine whether it complies with relevant regulations through the rule matching algorithm, and generate the rule matching degree M r :

[0153]

[0154] Among them, γ i is the importance weight of each regulation or policy rule, S sim(i) is the similarity score of the i-th rule, M r Indicates the overall rule matching degree;

[0155] S35, according to the rule matching degree M r Based on the results, a preliminary audit report is generated, which includes the marked violations and non-compliant clauses, rule matching analysis and recommended modification measures, and the report is sent to the review experts for further evaluation;

[0156] S36. After generating the preliminary audit report, continuously optimize and update the rule base R based on feedback and historical data p , and adjust the similarity calculation and rule matching algorithms.

[0157] In this implementation, S4 includes the following steps:

[0158] S41. Extract design elements from bidding documents X d , including graphic design d , Color scheme C s , Layout L o and brand logo B i , converted into a multidimensional feature vector x i , each element represents the characteristic value of the design attribute in different dimensions;

[0159] S42, using the art recognition algorithm, through the multi-layer perceptron and self-attention mechanism, the feature vector x i For in-depth processing:

[0160] S43, identify the importance of design elements in various dimensions through the self-attention mechanism and generate a weighted feature vector a i :

[0161]

[0162] Among them, Q i =W q ·x i , K i =W k ·x i 、V i =W v ·x i , respectively, the linear transformation matrices of query, key, and value, d k is the dimension of the key vector, W q , W k , W v is the weight matrix of the self-attention mechanism, Softmax is used to calculate the attention weight, a i represents the weighted feature vector;

[0163] S45. Use a multi-layer perceptron to perform nonlinear mapping on the weighted feature vector to obtain a preliminary style feature vector z i :

[0164] z i =ReLU(W1·a i +b1);

[0165] Among them, W1 and b1 are the weight and bias of MLP, and ReLU is the activation function, which maps the weighted features nonlinearly;

[0166] S46, the style feature vector z i With known art style vector S j Compare and calculate the similarity R between each design element and the art style sij :

[0167]

[0168] Among them, S fj represents the vector of the jth known artistic style, R sij Indicates the design element z i With art style S j The similarity is calculated by dot product and normalized;

[0169] S47. Use the generative adversarial network to further adjust the style similarity matrix R s , generate the optimized style feature matrix F adj :

[0170]

[0171] Among them, G j is the generator function, η is the style adjustment coefficient, z j is the introduced noise vector, tanh represents the hyperbolic tangent function;

[0172] S48, the optimized style feature matrix F adj Input the emotional computing model to calculate the emotional response vector e of each design element in the user's emotional space i :

[0173]

[0174] Among them, θ j is the emotional weight coefficient, reflecting the influence of different artistic styles on emotional response; σ j is a nonlinear sentiment activation function;

[0175] S49, combined with style consistency index C s and the emotional response vector ei , calculate the sentiment consistency score E of the bidding document c :

[0176]

[0177] in, is the emotional vector of the user experience standard, reflecting the user's expected emotional response; λ is the emotional consistency adjustment factor, adjusting the balance between style and emotion; C s It is a style consistency index, which measures the coordination of design elements in the dimension of artistic style;

[0178] S410, dynamically adjust the weight W1 and bias b1 of the multilayer perceptron and the generator parameter G in the generative adversarial network according to the data feedback after each review j and the sentiment activation function σ j .

[0179] In this implementation, S5 includes the following steps:

[0180] S51, extract the scoring path data from the review experts' scoring process in real time, and record the input feature X of each scoring node p =[x p1 ,x p2 ,...,x pn ] and the corresponding scoring decision D p , where x pi represents the i-th scoring feature value, D p represents the output of the scoring decision;

[0181] S52. For each scoring feature value x pi Fuzzy processing is performed to convert it into a fuzzy set μ pi (x pi ), membership function of fuzzy set:

[0182]

[0183] Among them, c pi is the center of the eigenvalue, σ pi is the expansion coefficient, and the membership function is used to describe the degree of belonging of the scoring feature value in the fuzzy set;

[0184] S53, expansion coefficient σ pi The choice of affects the shape and width of the membership function, which is usually adjusted dynamically according to the distribution of the scoring features to meet the requirements of different review tasks. The expansion coefficient σ pi Optimization:

[0185]

[0186] Among them, Q represents the number of samples in the historical data, x pi,q Represents the scoring feature value of the qth sample;

[0187] S54, based on the fuzzified eigenvalues, generate the fuzzy rule set R using fuzzy logic reasoning j :

[0188] R j :IFμ p1 (x p1 )ANDμ p2 (x p2 )AND...ANDμ pn (x pn )THEND p =f j (X p );

[0189] Among them, f j (X p ) indicates that the rule R j The associated output scoring decision, f j (X p The specific form of ) depends on the definition of fuzzy logic rules, which is usually a linear or nonlinear function. It is optimized by analyzing historical scoring data and dynamically adjusting the weight w according to historical data. i :

[0190]

[0191] Among them, D p,q For the scoring decision in the historical scoring data, the weight w of the feature is updated by weighting the results of the historical scoring data i ;

[0192] S55, based on fuzzy rule set R j Construct fuzzy decision tree T f In the decision tree, each node represents a fuzzy rule, and each leaf node corresponds to a decision output D fj ;

[0193] S56. Further optimize the fuzzy decision tree T f The output of the rule is adjusted by adjusting the confidence C j , the influence of dynamic optimization fuzzy rules:

[0194]

[0195] Among them, w i To adjust the feature weights based on historical scoring data, the weights w are dynamically optimized using historical scoring data. i, the final score result S is generated by confidence weighting f :

[0196]

[0197] S57, each scoring process, according to the optimized fuzzy decision tree T f 'Adjust the evaluation experts' scores in real time and generate the final score result S f ;

[0198] S58. After each review, the rule base and membership function parameter c of the fuzzy logic decision tree are updated by analyzing historical scoring data and expert feedback. pi and σ pi , and continuously optimize the fuzzy rules and confidence adjustment strategies.

[0199] In this implementation, S6 includes the following steps:

[0200] S61. Collect scoring data from the review experts in real time during their scoring process s ={D p1 ,D p2 ,...,D pn}, where D pi Indicates the scoring result of the i-th scoring node, collects scoring data, and simultaneously collects multi-dimensional feature data X related to scoring s ={X p1 ,X p2 ,...,X pn}, including the behavioral characteristics of review experts, scoring time characteristics, and historical scoring trends;

[0201] S62. Use automated data analysis tools and real-time stream processing technology to analyze the scoring data D s and multidimensional feature data X s Perform dynamic monitoring, for each scoring node D pi , calculate the dynamic mean D pi (t) and the time-weighted standard deviation

[0202]

[0203] Among them, α is the smoothing factor, β and γ are the time decay coefficients, and Δt is the time interval;

[0204] S63. Using multi-dimensional feature data X s and rating data D s , combined with the deep learning model to conduct in-depth analysis of the scoring deviation, the deviation analysis model calculates the scoring deviation vector ΔD pi (t) and its contribution rate;

[0205] S64, based on the scoring deviation vector ΔD pi (t), combined with the scoring feature data X pi (t)), identify the source of the deviation through the causal inference model and calculate the attribution weight w of the deviation k :

[0206]

[0207] Among them, g k (X pi,k (t)) is the characteristic response function related to the deviation, and the characteristic response function is optimized through the training data set;

[0208] S65. Generate instant feedback report F r (t), including deviation nodes, deviation sources and adjustment suggestions, calculate the deviation adjustment value A for each scoring node pi (t):

[0209]

[0210] Among them, λ is the adjustment strength coefficient, m is the feature dimension, and the feedback adjustment value A pi (t) Provide score adjustment suggestions and allow review experts to make real-time modifications;

[0211] S66. Dynamically adjust the scoring deviation detection and feedback model during the scoring process, and adjust the parameters α, β, γ and the attribution weight w of the monitoring model through continuous learning and optimization. k , and dynamically adjust the threshold τ'(t) for scoring deviation detection:

[0212]

[0213] Among them, η is the learning rate, n is the number of scoring nodes, and the scoring strategy is continuously optimized through the adaptive learning mechanism.

[0214] In this embodiment, it also includes:

[0215] An identity verification module, which is used to verify the identity of bidders and review experts using multi-factor identity verification and behavioral biometrics when posting bidding information on the online platform;

[0216] The mixed reality scenario simulation module is used to create a virtual bid evaluation environment after the bidder submits the bid document, combining the generative adversarial network technology to simulate the performance of the bid document in different extreme market scenarios and user interaction environments and conduct virtual user experience testing;

[0217] The automatic review module is used to automatically review the content of the bidding documents using natural language processing technology, identify and mark possible illegal clauses or non-compliant content, and compare them with the preset regulations and policy rule base to generate a preliminary review report;

[0218] Artistic cognition and style analysis module, which is used to identify the artistic style and analyze the emotional response of the design elements in the bidding documents in combination with the stylized analysis network of artificial intelligence during the review process, and generate an artistic aesthetic evaluation report for the review experts' reference;

[0219] The scoring optimization module is used to use fuzzy logic decision tree technology to analyze the expert's scoring path and decision logic in real time when the review experts are scoring, and dynamically adjust and optimize the uncertainty factors involved in the scoring process; the real-time monitoring module is used to monitor and analyze the review experts' scoring data in real time through automated data analysis tools, identify potential scoring deviations, and provide an instant feedback mechanism to adjust the scoring strategy;

[0220] The result generation and storage module is used to generate a bid evaluation result report after the bid evaluation is completed, and to store the results securely through data encryption technology.

[0221] Embodiment 1:

[0222] In the bidding and evaluation process of a provincial public project, there are many companies participating in the bidding, and the bidding evaluation process is complicated, involving multi-party collaboration and large amounts of data processing. Due to the limitations of manual operation and paper documents, the traditional bidding evaluation method has low bidding efficiency and is prone to errors, and information security and fairness cannot be effectively guaranteed. In order to solve these problems, a provincial government decided to adopt the bidding evaluation monitoring system and method based on a network platform proposed by the present invention, aiming to improve the efficiency, fairness and security of the bidding evaluation process.

[0223] During the bidding and evaluation process, the bidding evaluation system first released the bidding information through the online platform. When logging into the system, bidders and evaluation experts are required to authenticate their identities through multi-factor authentication and behavioral biometrics technology to ensure that the identity of each participant is authentic and unique, avoiding the participation of false identities.

[0224] After the bidder submits the bid document, the system uses a mixed reality scenario simulator combined with generative adversarial network technology to virtually simulate the performance of the bid document in extreme market scenarios and user interaction environments. By simulating the performance of the bid document in different scenarios, the evaluation experts can more comprehensively evaluate the actual feasibility and market adaptability of the bid. This process not only improves the scientific nature of the bid evaluation, but also provides more abundant data support for expert decision-making.

[0225] After receiving the bidding documents, the system automatically reviews the contents of the documents using natural language processing technology. This technology automatically identifies and marks possible illegal clauses or non-compliant contents by deeply analyzing the clauses and contents in the documents. This process significantly reduces the workload of manual review and the occurrence of human errors, ensuring the legality and compliance of the entire bidding evaluation process.

[0226] During the review process, the system combined the art recognition algorithm to identify the artistic style and analyze the emotional response of the design elements in the bidding documents. The system generated a detailed art aesthetic evaluation report to help the review experts better understand and evaluate the artistry and user experience of the design elements. This function is particularly suitable for the evaluation of design projects, ensuring that the review experts' decisions in terms of art aesthetics are more scientific and reasonable.

[0227] In order to improve the fairness and consistency of the scoring, the system uses fuzzy logic decision tree technology to analyze the expert's scoring path and decision logic in real time when the review experts are scoring. The system can dynamically adjust the uncertainty factors in the scoring process according to the expert's scoring, thereby optimizing the final scoring results. Through this technology, the system effectively reduces the subjective bias in the scoring and ensures the fairness and scientificity of the scoring results.

[0228] In addition, the system uses automated data analysis tools to monitor and analyze the review experts' scoring data in real time to identify possible scoring deviations. The system provides an instant feedback mechanism, allowing experts to adjust their scoring strategies in a timely manner based on the system's suggestions. This process not only improves the transparency of scoring, but also greatly improves the accuracy and consistency of scoring.

[0229] Data proves beneficial effects

[0230] During this bid evaluation process, the system recorded a large amount of data to prove the beneficial effects of the present invention. The specific data are as follows:

[0231] Table 1 Data analysis

[0232]

[0233]

[0234] The system of the invention significantly improved efficiency and accuracy in the bid evaluation process, shortened identity verification time by 75%, and increased verification accuracy to 100%. Through virtual scenario simulation, expert decision satisfaction increased by 25%. Natural language processing technology shortened document review time by 70%, and increased the accuracy of non-compliant clause identification by 30%. Fuzzy logic decision tree technology reduced scoring deviation by 40%, and increased scoring consistency and accuracy by 15%. The total time consumption of the overall bid evaluation process was reduced by 40%, greatly improving the efficiency and fairness of bid evaluation.

[0235] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The bidding evaluation monitoring method based on the network platform is characterized by: The steps include: S1. Publish the bidding information on the online platform and use multi-factor authentication and behavioral biometrics technology to authenticate the bidders and evaluation experts; S2. After the bidder submits the bid documents, a virtual bid evaluation environment is created through a mixed reality scenario simulator combined with generative adversarial network technology to simulate the performance of the bid documents in different extreme market scenarios and user interaction environments and conduct virtual user experience tests; S3. After receiving the bidding documents, we will use natural language processing technology to automatically review the content of the documents, identify and mark possible illegal clauses or non-compliant content, and compare this information with the preset regulations and policy rule base to generate a preliminary review report; S4. During the review process, an art recognition algorithm is implanted and combined with an artificial intelligence stylized analysis network to perform artistic style recognition and emotional response analysis on the design elements in the bidding documents and generate an artistic aesthetic evaluation report; S5. Use fuzzy logic decision tree technology to analyze experts’ scoring paths and decision logic in real time, and dynamically adjust and optimize the uncertainty factors involved in the scoring process; S6. Use automated data analysis tools to monitor and analyze the review experts’ scoring data in real time, identify potential scoring biases, and provide an immediate feedback mechanism to allow experts to adjust their scoring strategies based on the suggestions; S7. After the bid evaluation is completed, the bid evaluation result report is automatically generated and the results are securely stored using data encryption technology; S8. Comprehensively consider the evaluation experts’ scoring data, the content of the bidding documents and their performance in the virtual environment to generate a multi-dimensional bidding evaluation decision support report and publish it to relevant parties.

2. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S1 comprises the following steps: S11. When posting bidding information on the online platform, the identity information of the bidder and the evaluation experts shall be initially verified through identity verification, which includes the combination of user name and password verification and dynamic verification code verification sent by SMS or email; S12. After the initial verification is completed, the behavioral biometrics module is started to collect behavioral feature data of the bidder and the evaluation experts. The specific behavioral features include keystroke dynamics, mouse movement trajectory and touch screen gestures; S13, comparing the collected behavior feature data with the behavior feature model stored in the database, and calculating the matching degree R, where R represents the matching degree between the behavior feature and the preset model, and the matching degree is calculated based on the behavior feature vector, and the value range of R is [0,1], and the higher the value, the higher the matching degree; S14. According to the result of the matching degree R, if the R value is greater than or equal to the preset threshold value T, the identity verification is deemed to be passed, and the bidder and the evaluation expert are allowed to continue the operation; if the R value is lower than T, the identity verification fails, and the user is prompted to re-verify; S15. After identity authentication is passed, a unique session identifier SID is generated for each successfully authenticated user. The SID is used to track the integrity and continuity of all operations of bidders and evaluation experts during the bid evaluation process.

3. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S2 comprises the following steps: S21. After the bidder submits the bid document, extract the multi-dimensional element data in the bid document, including technical indicators T j 、Design parameters D k Functional module F l and user interaction logic U m , convert it into a feature vector set X; S22, generate a series of market variables M through generative adversarial network technology i Each market scenario is formed by coupling market demand fluctuation V1, competitor strategy change V2, and economic environment fluctuation V3 to form a composite scenario model C i , specifically: Among them, λ p1 , q2 , r3 and λ s4 are the coupling coefficients related to technical indicators, design parameters, functional modules and user interaction logic in the market context, respectively. i Represents a composite scenario model under market scenarios; S23, composite situation model C i Introducing the mixed reality scenario simulator, which combines virtual reality and augmented reality technologies to build an immersive virtual bid evaluation environment that includes elements of the real world and the virtual world, allowing evaluation experts to interact with the bid proposals in a highly simulated environment; S24. In the mixed reality scenario simulator, according to the composite scenario model C i Conduct user interaction simulation on the bidding scheme, including the user's behavior path P u 、Interactive feedback u , and emotional response E u , and transform these data into the interaction behavior matrix B i ; S25, the interaction behavior matrix B is analyzed through a recurrent neural network i Analyze and identify the temporal characteristics of user behavior and calculate the user experience score S u At the same time, according to the composite situation model C i Calculate the market adaptability score S m ; S26. In the mixed reality scenario simulator, the user experience score S in different market scenarios u and market adaptability score S m , calculate the comprehensive score result R s : Among them, α i and β i is the weight factor in each situation, max(S m ) and max(S u ) is the maximum value of the score, which is used for standardization; S27, according to the comprehensive scoring results R s , automatically generates a virtual user experience test report, which includes interactive behavior analysis, emotional response summary, market adaptability assessment and potential risk warning, and gives optimization suggestions, which are then fed back to the bidder for adjustment; S28. Update and optimize the market scenarios in the generative adversarial network model and mixed reality scenario simulator by continuously accumulating and analyzing bid evaluation data.

4. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S3 comprises the following steps: S31. After receiving the bidding documents, the text content in the bidding documents is structured and key content elements are extracted, including technical description T d 、Contract Clauses C t 、Legal Statement s And the bidder's self-description materials a Etc., and convert it into the corresponding text feature vector set X t ; S32. Use natural language processing technology to process the text feature vector set X t Perform semantic analysis and convert text content into a high-dimensional vector space representation V through word embedding technology t , calculate the similarity S between the text content and the terms in the preset violation and non-compliance rule base sim ; S33, according to the similarity score S sim The results of the bidding documents are used to identify and mark the possible illegal clauses or non-compliant contents in the bidding documents, and the similarity scores exceeding the preset threshold T sim The clauses are automatically marked as high-risk content; S34. Compare the marked high-risk content with the preset regulations and policy rule base R p Compare and determine whether it complies with relevant regulations through the rule matching algorithm, and generate the rule matching degree M r : Among them, γ i is the importance weight of each regulation or policy rule, S sim (i) is the similarity score of the i-th rule, M r Indicates the overall rule matching degree; S35, according to the rule matching degree M r Based on the results, a preliminary audit report is generated, which includes the marked violations and non-compliant clauses, rule matching analysis and recommended modification measures, and the report is sent to the review experts for further evaluation; S36. After generating the preliminary audit report, continuously optimize and update the rule base R based on feedback and historical data p , and adjust the similarity calculation and rule matching algorithms.

5. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S4 comprises the following steps: S41. Extract design elements from bidding documents X d , including graphic design d , Color scheme C s , Layout L o and brand logo B i , converted into a multidimensional feature vector x i , each element represents the characteristic value of the design attribute in different dimensions; S42, using the art recognition algorithm, through the multi-layer perceptron and self-attention mechanism, the feature vector x i For in-depth processing: S43, identify the importance of design elements in various dimensions through the self-attention mechanism and generate a weighted feature vector a i : Among them, Q i =W q ·x i , K i =W k ·x i 、V i =W v ·x i , respectively, the linear transformation matrices of query, key, and value, d k is the dimension of the key vector, W q , W k , W v is the weight matrix of the self-attention mechanism, Softmax is used to calculate the attention weight, a i represents the weighted feature vector; S45. Use a multi-layer perceptron to perform nonlinear mapping on the weighted feature vector to obtain a preliminary style feature vector z i : from i =ReLU(W1·a i +b1); Among them, W1 and b1 are the weight and bias of MLP, and ReLU is the activation function, which maps the weighted features nonlinearly; S46, the style feature vector z i With known art style vector S j Compare and calculate the similarity R between each design element and the art style sij : Among them, S fj represents the vector of the jth known artistic style, R sij Indicates the design element z i With art style S j The similarity is calculated by dot product and normalized; S47. Use the generative adversarial network to further adjust the style similarity matrix R s , generate the optimized style feature matrix F adj : Among them, G j is the generator function, η is the style adjustment coefficient, z j is the introduced noise vector, tanh represents the hyperbolic tangent function; S48, the optimized style feature matrix F adj Input the emotional computing model to calculate the emotional response vector e of each design element in the user's emotional space i : Among them, θ j is the emotional weight coefficient, reflecting the influence of different artistic styles on emotional response; σ j is a nonlinear sentiment activation function; S49, combined with style consistency index C s and the emotional response vector e i , calculate the sentiment consistency score E of the bidding document c : in, is the emotional vector of the user experience standard, reflecting the user's expected emotional response; λ is the emotional consistency adjustment factor, adjusting the balance between style and emotion; C s It is a style consistency index, which measures the coordination of design elements in the dimension of artistic style; S410, dynamically adjust the weight W1 and bias b1 of the multilayer perceptron and the generator parameter G in the generative adversarial network according to the data feedback after each review j and the sentiment activation function σ j .

6. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S5 comprises the following steps: S51, extract the scoring path data from the review experts' scoring process in real time, and record the input feature X of each scoring node p =[x p1 ,x p2 ,...,x pn ] and the corresponding scoring decision D p , where x pi represents the i-th scoring feature value, D p represents the output of the scoring decision; S52. For each scoring feature value x pi Fuzzy processing is performed to convert it into a fuzzy set μ pi (x pi ), membership function of fuzzy set: Among them, c pi is the center of the eigenvalue, σ pi is the expansion coefficient, and the membership function is used to describe the degree of belonging of the scoring feature value in the fuzzy set; S53, expansion coefficient σ pi The choice of affects the shape and width of the membership function, which is usually adjusted dynamically according to the distribution of the scoring features to meet the requirements of different review tasks. The expansion coefficient σ pi Optimization: Among them, Q represents the number of samples in the historical data, x pi,q Represents the score feature value of the qth sample; S54, based on the fuzzified eigenvalues, generate the fuzzy rule set R using fuzzy logic reasoning j : R j :IFμ p1 (x p1 )ANDμ p2 (x p2 )AND...ANDμ pn (x pn )THEND p =f j (X p ); Among them, f j (X p ) indicates that the rule R j The associated output scoring decision, f j (X p The specific form of ) depends on the definition of fuzzy logic rules, which is usually a linear or nonlinear function. It is optimized by analyzing historical scoring data and dynamically adjusting the weight w according to historical data. i : Among them, D p,q For the scoring decision in the historical scoring data, the weight w of the feature is updated by weighting the results of the historical scoring data i ; S55, based on fuzzy rule set R j Construct fuzzy decision tree T f In the decision tree, each node represents a fuzzy rule, and each leaf node corresponds to a decision output D fj ; S56. Further optimize the fuzzy decision tree T f The output of the rule is adjusted by adjusting the confidence C j , the influence of dynamic optimization fuzzy rules: Among them, w i To adjust the feature weights based on historical scoring data, the weights w are dynamically optimized using historical scoring data. i , the final score result S is generated by confidence weighting f ; S57, each scoring process, according to the optimized fuzzy decision tree T f 'Adjust the evaluation experts' scores in real time and generate the final score result S f ; S58. After each review, the rule base and membership function parameter c of the fuzzy logic decision tree are updated by analyzing historical scoring data and expert feedback. pi and σ pi , and continuously optimize the fuzzy rules and confidence adjustment strategies.

7. The bidding evaluation monitoring method based on a network platform according to claim 1 is characterized in that: The S6 comprises the following steps: S61. Collect scoring data from the review experts in real time during their scoring process s ={D p1 ,D p2 ,...,D pn }, where D pi Indicates the scoring result of the i-th scoring node, collects scoring data, and simultaneously collects multi-dimensional feature data X related to scoring s ={X p1 ,X p2 ,...,X pn }, including the behavioral characteristics of review experts, scoring time characteristics, and historical scoring trends; S62. Use automated data analysis tools and real-time stream processing technology to analyze the scoring data D s and multidimensional feature data X s Perform dynamic monitoring, for each scoring node D pi , calculate the dynamic mean D pi (t) and the time-weighted standard deviation Among them, α is the smoothing factor, β and γ are the time decay coefficients, and Δt is the time interval; S63. Using multi-dimensional feature data X s and rating data D s , combined with the deep learning model to conduct in-depth analysis of the scoring deviation, the deviation analysis model calculates the scoring deviation vector ΔD pi (t) and its contribution rate; S64, based on the scoring deviation vector ΔD pi (t), combined with the scoring feature data X pi (t)), identify the source of the deviation through the causal inference model and calculate the attribution weight w of the deviation k : Among them, g k (X pi,k (t)) is the characteristic response function related to the deviation, and the characteristic response function is optimized through the training data set; S65. Generate instant feedback report F r (t), including deviation nodes, deviation sources and adjustment suggestions, calculate the deviation adjustment value A for each scoring node pi (t): Among them, λ is the adjustment strength coefficient, m is the feature dimension, and the feedback adjustment value A pi (t) Provide score adjustment suggestions and allow review experts to make real-time modifications; S66. Dynamically adjust the scoring deviation detection and feedback model during the scoring process, and adjust the parameters α, β, γ and the attribution weight w of the monitoring model through continuous learning and optimization. k , and dynamically adjust the threshold τ'(t) for scoring deviation detection: Among them, η is the learning rate, n is the number of scoring nodes, and the scoring strategy is continuously optimized through the adaptive learning mechanism.

8. The bidding evaluation monitoring system based on the network platform is characterized by: include: An identity verification module, which is used to verify the identity of bidders and review experts using multi-factor identity verification and behavioral biometrics when posting bidding information on the online platform; The mixed reality scenario simulation module is used to create a virtual bid evaluation environment after the bidder submits the bid document, combining the generative adversarial network technology to simulate the performance of the bid document in different extreme market scenarios and user interaction environments and conduct virtual user experience testing; The automatic review module is used to automatically review the content of the bidding documents using natural language processing technology, identify and mark possible illegal clauses or non-compliant content, and compare them with the preset regulations and policy rule base to generate a preliminary review report; Artistic cognition and style analysis module, which is used to identify the artistic style and analyze the emotional response of the design elements in the bidding documents in combination with the stylized analysis network of artificial intelligence during the review process, and generate an artistic aesthetic evaluation report for the review experts' reference; The scoring optimization module is used to use fuzzy logic decision tree technology to analyze the expert's scoring path and decision logic in real time when the review expert scores, and dynamically adjust and optimize the uncertainty factors involved in the scoring process; Real-time monitoring module, which uses automated data analysis tools to monitor and analyze the reviewers' scoring data in real time, identify potential scoring biases, and provide an immediate feedback mechanism to adjust the scoring strategy; The result generation and storage module is used to generate a bid evaluation result report after the bid evaluation is completed, and to store the results securely through data encryption technology.

Images