Method and system for improving bid evaluation accuracy

Abstract

The invention provides a method and system for improving bid evaluation accuracy, and relates to the technical field of bid evaluation, and the method comprises the steps: bid file data preprocessing including data cleaning and feature selection, bid insensitive learning model building and fusion, and real-time fusion model performance evaluation including fairness constraint, cross validation and bid detection. Updating according to an evaluation result, substituting the feature information to calculate a bid evaluation score, adding an expert review mechanism to adjust the score, and archiving a final bid evaluation score; according to the method, the labor cost is reduced through a rule calculation engine, the scoring accuracy and generalization ability are improved through data cleaning and other technologies, expert review and fairness constraint are introduced to guarantee fairness, review traceability is guaranteed through an archiving result, project decision is supported, and the bid evaluation accuracy is improved through multiple innovation strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of bidding evaluation, and in particular to a method and system for improving the accuracy of bid evaluation. Background Art

[0002] In the field of bidding, the accuracy and fairness of bid evaluation are of vital importance. With the continuous development of information technology, bid evaluation technology is gradually moving towards intelligence and automation; currently, the development of this field is mainly reflected in the following aspects:

[0003] Ability to process a large amount of bidding document data, including data collection, organization, and analysis; more and more advanced models and algorithms are applied to the bidding evaluation process to improve the accuracy and efficiency of bidding evaluation; such as large bidding evaluation models and rule calculation engines, making the bidding evaluation process more automated and intelligent;

[0004] However, there are still some problems in the existing technology. First, there are often some biased features or data points in the bidding document data, which affect the fairness of the bid evaluation; and a single bid evaluation model is prone to bias, resulting in inaccurate scoring; at the same time, manual operation is cumbersome and time-consuming, and it is difficult to ensure that each evaluation is based on a unified and fair standard; the archiving and traceability of the evaluation results are not perfect, which is not conducive to subsequent project management and decision-making;

[0005] Therefore, the art is in urgent need of a method and system for improving the accuracy of bid evaluation to solve the above-mentioned technical problems. Summary of the invention

[0006] The present invention provides a method and system for improving the accuracy of bid evaluation, which improves the accuracy of bid evaluation and ensures the fairness and traceability of the review process by means of data cleaning, feature engineering, selecting a suitable model, introducing cross-validation and monitoring evaluation mechanisms, increasing data enhancement and fairness constraints, adopting multi-model fusion, expert review and result archiving.

[0007] In one aspect, the present invention provides a method for improving bid evaluation accuracy, comprising:

[0008] Step 1: Preprocess the bidding document data, including data cleaning and feature selection;

[0009] Step 2: Establish several bias-insensitive learning models and fuse them through multi-model fusion strategy;

[0010] Step 3, performing real-time evaluation on the performance of the fusion model, including fairness constraints, cross-validation, and bias detection, and updating the fusion model based on the evaluation results;

[0011] Step 4: Substitute the characteristic information corresponding to each bidding document data into the fusion model to calculate the evaluation score of each bidder;

[0012] Step 5: Add an expert review mechanism to adjust the bid evaluation scores;

[0013] Step six: Archive the adjusted final evaluation scores.

[0014] According to a method for improving bid evaluation accuracy provided by the present invention, the process of data cleaning and feature selection includes:

[0015] The data cleaning includes using an intelligent data cleaning algorithm based on outlier detection to calculate the local outlier factor LOF of each data point. If the LOF value of a data point is greater than a preset threshold, it is removed from the data set; the specific formula is:

[0016]

[0017] Where P is the data point to be detected, LOF(P) is the LOF value of data point P, N(P) is the set of neighborhood data points of P, and ρ(P) is the local reachability density of data point P;

[0018] The feature selection includes: using a feature selection method based on mutual information, calculating the mutual information between each feature and the bid evaluation result, selecting features with higher mutual information values, and avoiding features that cause bias; the specific formula of the mutual information value is:

[0019]

[0020] Where I(X;Y) is the mutual information value, X is the feature, x is the value of feature X, Y is the evaluation result, y is the value of evaluation result Y, p(x,y) is the joint probability distribution of X=x and Y=y, p(x) and p(y) are the marginal probability distributions of X and Y respectively.

[0021] According to a method for improving bid evaluation accuracy provided by the present invention, the several bias-insensitive learning models include a random forest model and a gradient boosting tree model.

[0022] According to a method for improving bid evaluation accuracy provided by the present invention, the process of fusing the learning models through a multi-model fusion strategy includes:

[0023] A weighted average method is used to combine several bias-insensitive learning models. Different weights are assigned to each model according to its performance on the validation set to reduce the bias of a single learning model.

[0024] The fusion model is:

[0025]

[0026] Where i is the bidder index, j is the learning model index; S i is the evaluation score after fusion, S ij is the evaluation score of the j-th learning model for the i-th bidder, ω j is the weight of the jth learning model, and n is the number of learning models.

[0027] According to a method for improving bid evaluation accuracy provided by the present invention, the implementation process of the fairness constraint is to add the fairness constraint during the model training process, which specifically includes:

[0028] A fairness indicator is defined, where the fairness indicator is the variance of the score differences between different bidders;

[0029] Establish the fairness indicator formula:

[0030]

[0031] Among them, F is the fairness index, m is the number of bidders, S i and S k are the evaluation scores of the i-th bidder and the k-th bidder respectively;

[0032] A fairness index threshold is preset, and the fairness index F is compared with the fairness index threshold; if the fairness index F is greater than the fairness index threshold, it is judged that the fusion model is unfair;

[0033] The parameters of the fusion model are adjusted based on the fairness index F until the fairness index F is less than or equal to the fairness index threshold.

[0034] According to a method for improving bid evaluation accuracy provided by the present invention, the implementation process of the cross-validation is to use K-fold cross-validation to evaluate the performance stability of the fusion model, which specifically includes:

[0035] The evaluation scores of each bidder in history are used as training data, and the training data set is divided into K subsets of equal size. One subset is selected as the validation set in turn, and the remaining K-1 subsets are used as the training set for model training and validation;

[0036] Calculate the mean and standard deviation of K validation results;

[0037] A standard deviation threshold is set in advance, and the standard deviation of the verification result is compared with the preset standard deviation threshold; if the standard deviation of the verification result is greater than the standard deviation threshold, it is judged that the fusion model is unstable;

[0038] The parameters of the fusion model are adjusted based on the standard deviation of the verification results until the standard deviation of the verification results is less than or equal to the standard deviation threshold.

[0039] According to a method for improving bid evaluation accuracy provided by the present invention, the implementation process of bias detection includes:

[0040] Establish a real-time monitoring system to continuously monitor the performance and scoring results of the fusion model;

[0041] Regularly evaluate the model with new data and calculate bias indicators;

[0042] The calculation formula of the bias index is:

[0043]

[0044] Among them, B is the bias index, S G1 and S G2 are the average scores of different bidder combinations, is the average score of all bidders;

[0045] A bias threshold is set in advance, and the bias index B is compared with the preset bias threshold; if the bias index B is greater than the bias threshold, it is determined that the fusion model has bias;

[0046] The parameters of the fusion model are adjusted based on the bias index B until the bias index B is less than or equal to the bias threshold.

[0047] According to a method for improving bid evaluation accuracy provided by the present invention, the process of updating the fusion model based on the evaluation result includes:

[0048] Based on the results of fairness constraints, cross-validation, and bias detection, adjust the model parameters or select a new model for fusion to improve the performance and accuracy of the model.

[0049] According to a method for improving bid evaluation accuracy provided by the present invention, the process of adding an expert review mechanism and adjusting the bid evaluation score includes:

[0050] The experts adjust the evaluation scores automatically calculated by the fusion model and give reasons for the adjustments. The final evaluation score is obtained by weighting the evaluation score automatically calculated by the fusion model and the evaluation score adjusted by the experts. The specific formula is:

[0051] FS i =α*S i +(1-α)*ES i

[0052] Among them, FS i is the final evaluation score of the i-th bidder, Si is the evaluation score of the i-th bidder automatically calculated based on the fusion model, ES i is the evaluation score of the ith bidder based on expert adjustment, α is the scoring weight of the fusion model, and its value range is {0, 1}.

[0053] In another aspect, the present invention provides a system for improving bid evaluation accuracy, comprising:

[0054] Data preprocessing module, used to preprocess bidding document data, including data cleaning and feature selection;

[0055] A model fusion module, which is connected to the data preprocessing module, establishes several bias-insensitive learning models, and fuses the learning models through a multi-model fusion strategy;

[0056] A model evaluation module, which is connected to the model fusion module and is used to evaluate the performance of the fusion model in real time, including fairness constraints, cross-validation and bias detection, and update the fusion model based on the evaluation results;

[0057] A bid evaluation calculation module, which is connected to the data preprocessing module and the model fusion module, substitutes the characteristic information corresponding to each bid document data into the fusion model, and calculates the bid evaluation score of each bidder;

[0058] An expert review module, which is connected to the bid evaluation calculation module and is used to add an expert review mechanism to adjust the bid evaluation scores;

[0059] A result archiving module is connected to the expert review module to archive the adjusted final bid evaluation scores.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] Through the innovative rule calculation engine, the bidder's evaluation score is automatically and efficiently calculated, greatly reducing the tediousness and time cost of manual operation;

[0062] The accuracy and generalization of scoring have been significantly improved by adopting techniques such as data cleaning, feature engineering, and selecting models that are insensitive to bias;

[0063] Introducing an expert review mechanism and fairness constraints to ensure that each review is based on unified and fair standards, reducing the impact of bias and ensuring the fairness of the bid evaluation process;

[0064] Submitting the review results to the bid evaluation brain for archiving ensures the integrity and traceability of the entire review process, providing strong support for subsequent project management and decision-making;

[0065] The accuracy of bid evaluation can be improved through data cleaning algorithms for outlier detection, feature selection methods for mutual information, and multi-model fusion strategies.

[0066] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0067] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] 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:

[0069] Figure 1 It is a flowchart of a method for improving bid evaluation accuracy provided by an embodiment of the present invention;

[0070] Figure 2 It is a structural diagram of a system for improving bid evaluation accuracy provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0072] Embodiment 1:

[0073] This embodiment provides a method for improving the accuracy of bid evaluation. Figure 1 ,include:

[0074] A method for improving the accuracy of bid evaluation, characterized by comprising:

[0075] Step 1: Preprocess the bidding document data, including data cleaning and feature selection;

[0076] Step 2: Establish several bias-insensitive learning models and fuse them through multi-model fusion strategy;

[0077] Step 3: Perform real-time evaluation of the performance of the fusion model, including fairness constraints, cross-validation, and bias detection, and update the fusion model based on the evaluation results;

[0078] Step 4: Substitute the characteristic information corresponding to each bidding document data into the fusion model to calculate the evaluation score of each bidder;

[0079] Step 5: Add an expert review mechanism to adjust the bid evaluation scores;

[0080] Step six: Archive the adjusted final evaluation scores.

[0081] The principle of the above embodiment is: by preprocessing the bidding document data, removing data points that may be biased and selecting appropriate features, establishing multiple bias-insensitive learning models and fusing them, and then performing real-time evaluation of the fusion model, including fairness constraints, cross-validation and bias detection, to ensure the accuracy and fairness of the model; finally, substituting the characteristic information of the bidding document data into the fusion model to calculate the evaluation score, and adjusting it through the expert review mechanism, and finally archiving the adjusted final evaluation score to ensure the traceability of the entire review process.

[0082] The beneficial effects of the above embodiments are: through technologies such as data cleaning, feature selection and multi-model fusion, the impact of bias is reduced and the accuracy of scoring is improved; the introduction of fairness constraints and expert review mechanisms ensures that the review process is based on unified and fair standards; the evaluation scores are automatically calculated to reduce the tediousness and time cost of manual operations; the review results are archived to facilitate subsequent project management and decision-making.

[0083] In order to further optimize the above embodiment, the process of data cleaning and feature selection includes:

[0084] Data cleaning includes using an intelligent data cleaning algorithm based on outlier detection to calculate the local outlier factor LOF of each data point. If the LOF value of a data point is greater than a preset threshold, it will be removed from the data set; the specific formula is:

[0085]

[0086] Where P is the data point to be detected, LOF(P) is the LOF value of data point P, N(P) is the set of neighborhood data points of P, and ρ(P) is the local reachability density of data point P;

[0087] Feature selection includes: using the feature selection method based on mutual information, calculating the mutual information between each feature and the bid evaluation result, selecting features with higher mutual information values, and avoiding features that cause bias; the specific formula for the mutual information value is:

[0088]

[0089] Where I(X;Y) is the mutual information value, X is the feature, x is the value of feature X, Y is the evaluation result, y is the value of evaluation result Y, p(x,y) is the joint probability distribution of X=x and Y=y, p(x) and p(y) are the marginal probability distributions of X and Y respectively.

[0090] It should be noted that, assuming we have a set of bidding document data, including multiple features and corresponding evaluation results.

[0091] 1. Data cleaning part

[0092] Assume there are 10 data points and we calculate the Local Outlier Factor (LOF) for each data point.

[0093] 1. First, determine the set of neighborhood data points N(P). Assume that for data point p1, we select the 5 data points closest to it as its neighborhood, that is, N(P1) = {P2, P3, P4, P5, P6}.

[0094] 2. Calculate the local reachable density ρ(P). Assume that the distance between data points is calculated using the Euclidean distance, and the reachable distance is determined based on the distance. For example, the reachable distance from data point p1 to p2 is the larger value of d(P1, P2) and a constant k, where k can be set according to the specific situation. Then calculate the local reachable density as

[0095]

[0096] 3. Calculate LOF(P).

[0097] For data point p1,

[0098] Calculate the LOF values ​​of other data points in turn. Assuming the threshold is set to 1.5, if the LOF value of a data point is greater than 1.5, it will be removed from the data set.

[0099] 2. Feature Selection

[0100] Suppose we have three features X1, X2, X3 and the evaluation result Y.

[0101] 1. Calculate the joint probability distribution p(x, y). Assume that we have 50 data samples, of which there are 10 samples where the feature X1 takes the value x1 and the bid evaluation result Y takes the value y1, then p(x1, y1) = 10 / 50 = 0.2. Similarly, calculate the joint probability distribution of other combinations.

[0102] 2. Calculate the marginal probability distribution p(x) and p(y). For example, if there are 20 samples with the value of X1 being x1, then p(x1) = 20 / 50 = 0.4. Similarly, calculate the marginal probability distribution of the bid evaluation result Y.

[0103] 3. Calculate mutual information I(X;Y). Taking feature X1 as an example, The mutual information values ​​of X2 and X3 are calculated in turn, and the features with higher mutual information values ​​are selected for subsequent bid evaluation.

[0104] Through this process of data cleaning and feature selection, noise and bias in the data can be effectively removed, and the quality and reliability of the data can be improved, thereby providing more accurate and effective data support for the subsequent bid evaluation process.

[0105] To further optimize the above embodiment, several bias-insensitive learning models include a random forest model and a gradient boosting tree model.

[0106] It should be noted that the random forest model can effectively reduce the bias of a single model and improve the accuracy and stability of scoring by combining multiple decision trees; the gradient boosting tree model can automatically adjust the model parameters through continuous iterative training to make it more adaptable to the distribution of data, thereby improving the generalization ability and accuracy of the model; both models have good bias insensitivity and can provide reliable scoring results in the bid evaluation process.

[0107] In order to further optimize the above embodiment, the process of fusing the learning models through the multi-model fusion strategy includes:

[0108] A weighted average method is used to combine several bias-insensitive learning models. Different weights are assigned to each model according to its performance on the validation set to reduce the bias of a single learning model.

[0109] The fusion model is:

[0110]

[0111] Where i is the bidder index, j is the learning model index; S i is the evaluation score after fusion, S ij is the evaluation score of the j-th learning model for the i-th bidder, ω j is the weight of the jth learning model, and n is the number of learning models.

[0112] It should be noted that through this multi-model fusion strategy, the advantages of each learning model can be fully utilized, while reducing the bias of a single model and improving the accuracy and reliability of the evaluation results; by assigning different weights according to the performance of each model on the validation set, the fused model can be more adapted to the distribution of actual data, thereby improving the generalization ability of the model.

[0113] In order to further optimize the above embodiment, the implementation process of fairness constraint is to add fairness constraint in the model training process, which specifically includes:

[0114] Define a fairness indicator, which is the variance of the score differences between different bidders;

[0115] Establish the fairness indicator formula:

[0116]

[0117] Among them, F is the fairness index, m is the number of bidders, S i and S k are the evaluation scores of the i-th bidder and the k-th bidder respectively;

[0118] A fairness index threshold is preset, and the fairness index F is compared with the fairness index threshold; if the fairness index F is greater than the fairness index threshold, it is judged that the fusion model is unfair;

[0119] The parameters of the fusion model are adjusted based on the fairness index F until the fairness index F is less than or equal to the fairness index threshold.

[0120] It should be noted that when the fairness index F is large, it means that the score differences between different bidders are large, and some bidders may be treated unfairly.

[0121] By adjusting the parameters of the model, the behavior of the model can be changed to make it more fair in scoring different bidders. For example, the model's weights, thresholds, and other parameters can be adjusted to reduce the variance of scoring differences.

[0122] Through this fairness constraint method, it can be ensured that the fusion model remains fair to all bidders during the bid evaluation process and avoid unfair results caused by model bias.

[0123] In order to further optimize the above embodiment, it is characterized in that the implementation process of cross validation is to use K-fold cross validation to evaluate the performance stability of the fusion model, which specifically includes:

[0124] The evaluation scores of each bidder in history are used as training data, and the training data set is divided into K subsets of equal size. One subset is selected as the validation set in turn, and the remaining K-1 subsets are used as the training set for model training and validation;

[0125] Calculate the mean and standard deviation of K validation results;

[0126] A standard deviation threshold is set in advance, and the standard deviation of the verification result is compared with the preset standard deviation threshold; if the standard deviation of the verification result is greater than the standard deviation threshold, it is judged that the fusion model is unstable;

[0127] The parameters of the fusion model are adjusted based on the standard deviation of the verification results until the standard deviation of the verification results is less than or equal to the standard deviation threshold.

[0128] It should be noted that the average value can reflect the overall performance level of the model, while the standard deviation can reflect the stability of the model performance; if the standard deviation is small, it means that the model is relatively stable on different data sets; if the standard deviation is large, it means that the model is less stable; if the standard deviation of the verification result is greater than the standard deviation threshold, it is judged that the fusion model is unstable; this is because a large standard deviation means that the model has large performance differences on different data sets and may not be reliably applied in actual bid evaluation;

[0129] Adjust the parameters of the fusion model based on the standard deviation of the verification results until the standard deviation of the verification results is less than or equal to the standard deviation threshold. The stability of the model can be improved by adjusting the model's hyperparameters, training algorithms, etc. For example, the amount of training data can be increased, the learning rate can be adjusted, and regularization techniques can be used.

[0130] K-fold cross-validation can be used to more comprehensively evaluate the performance stability of the fusion model, promptly identify problems with the model, and improve the reliability and stability of the model by adjusting parameters, thereby improving the accuracy and fairness of the bid evaluation.

[0131] In order to further optimize the above embodiment, it is characterized in that the implementation process of bias detection includes:

[0132] Establish a real-time monitoring system to continuously monitor the performance and scoring results of the fusion model;

[0133] Regularly evaluate the model with new data and calculate bias indicators;

[0134] The calculation formula of bias index is:

[0135]

[0136] Among them, B is the bias index, S G1 and S G2 are the average scores of different bidder combinations, is the average score of all bidders;

[0137] A bias threshold is set in advance, and the bias index B is compared with the preset bias threshold; if the bias index B is greater than the bias threshold, it is determined that the fusion model has bias;

[0138] The parameters of the fusion model are adjusted based on the bias index B until the bias index B is less than or equal to the bias threshold.

[0139] It should be noted that if the bias index is greater than the bias threshold, the fusion model is judged to be biased. This is because when the bias index is large, it means that the average scores of different bidders combined are quite different, and some bidders may be treated unfairly.

[0140] Through this bias detection method, possible bias problems in the fusion model can be discovered and resolved in a timely manner, thereby improving the fairness and reliability of bid evaluation.

[0141] In order to further optimize the above embodiment, the process of updating the fusion model based on the evaluation result includes:

[0142] Based on the results of fairness constraints, cross-validation, and bias detection, adjust the model parameters or select a new model for fusion to improve the performance and accuracy of the model.

[0143] It should be noted that if the results of the fairness constraint show that the model is unfair, it may be necessary to adjust the model parameters to reduce the variance of the score differences between different bidders; for example, the weight distribution of the model can be adjusted or a regularization term can be introduced to constrain the behavior of the model;

[0144] For the cross-validation results, if the standard deviation of the validation results is greater than the standard deviation threshold, it indicates that the model is not stable enough. At this time, you can try to adjust the model's hyperparameters, such as the learning rate, number of layers, etc., or choose a more appropriate model architecture to improve the model's stability.

[0145] In terms of bias detection, if the bias indicator is greater than the bias threshold, it means that the model is biased; the impact of bias can be reduced by retraining the model, adjusting the data preprocessing method, or introducing adversarial training and other technologies;

[0146] In addition, new models can be selected for fusion based on the evaluation results; for example, if a model performs well in a specific aspect, its weight in the fusion model can be increased or the original poorly performing model can be replaced;

[0147] By continuously adjusting and optimizing according to the evaluation results, the performance and accuracy of the fusion model can be improved, making it better suited to the needs of the bid evaluation task;

[0148] This process is an iterative process that requires continuous monitoring and evaluation of the model's performance to ensure that the model always performs well.

[0149] In order to further optimize the above embodiment, an expert review mechanism is added, and the process of adjusting the bid evaluation score includes:

[0150] The experts adjust the evaluation scores automatically calculated by the fusion model and give reasons for the adjustments. The final evaluation score is obtained by weighting the evaluation score automatically calculated by the fusion model and the evaluation score adjusted by the experts. The specific formula is:

[0151] FS i =α*S i +(1-α)*ES i

[0152] Among them, FS i is the final evaluation score of the i-th bidder, S i is the evaluation score of the i-th bidder automatically calculated based on the fusion model, ES i is the evaluation score of the ith bidder based on expert adjustment, α is the scoring weight of the fusion model, and its value range is {0, 1}.

[0153] It should be noted that this weighting method can balance the automatic calculation results of the fusion model and the judgment of experts to a certain extent; if the performance of the fusion model is good and the reliability is high, the value of α can be appropriately increased to make the final evaluation score more inclined to the calculation results of the fusion model; if the experts believe that in some cases they need to rely more on their professional judgment, the value of α can be reduced to increase the weight of the evaluation score adjusted by the experts;

[0154] The introduction of an expert review mechanism can improve the accuracy and reliability of bid evaluation results and avoid possible deviations that may arise from relying solely on model calculations. At the same time, clear records of adjustment reasons can also help improve the transparency and traceability of the bid evaluation process.

[0155] Example 2

[0156] This embodiment provides a system for improving the accuracy of bid evaluation. Figure 2 ,include:

[0157] Data preprocessing module, used to preprocess bidding document data, including data cleaning and feature selection;

[0158] The model fusion module is connected with the data preprocessing module to establish several bias-insensitive learning models and fuse the learning models through a multi-model fusion strategy;

[0159] The model evaluation module is connected to the model fusion module and is used to evaluate the performance of the fusion model in real time, including fairness constraints, cross-validation, and bias detection, and to update the fusion model based on the evaluation results;

[0160] The bid evaluation calculation module is connected with the data preprocessing module and the model fusion module, and substitutes the characteristic information corresponding to the data of each bidding document into the fusion model to calculate the bid evaluation score of each bidder;

[0161] An expert review module, which is connected to the bid evaluation calculation module, is used to add an expert review mechanism to adjust the bid evaluation scores;

[0162] The result archiving module is connected with the expert review module to archive the adjusted final bid evaluation scores.

[0163] It should be noted that this embodiment can more accurately calculate the bidder's evaluation score and reduce errors and biases through data preprocessing, multi-model fusion and real-time evaluation; functions such as fairness constraints, cross-validation and bias detection ensure the fairness of the evaluation process and avoid unfair treatment of bidders; the addition of the expert review mechanism enables the experience and judgment of experts to be fully utilized, further improving the quality of bid evaluation; the existence of the result archiving module enables the evaluation results to be easily traced and queried, increasing the transparency and credibility of the evaluation process; the system's automated processing capabilities greatly reduce the time and cost of manual operations and improve the efficiency of bid evaluation.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for improving the accuracy of bid evaluation, characterized in that: include: Step 1: Preprocess the bidding document data, including data cleaning and feature selection; Step 2: Establish several bias-insensitive learning models and fuse them through multi-model fusion strategy; Step 3, performing real-time evaluation on the performance of the fusion model, including fairness constraints, cross-validation, and bias detection, and updating the fusion model based on the evaluation results; Step 4: Substitute the characteristic information corresponding to each bidding document data into the fusion model to calculate the bid evaluation score of each bidder; Step 5: Add an expert review mechanism to adjust the bid evaluation scores; Step six: Archive the adjusted final evaluation scores.

2. A method for improving bid evaluation accuracy according to claim 1, characterized in that: The process of data cleaning and feature selection includes: The data cleaning includes using an intelligent data cleaning algorithm based on outlier detection to calculate the local outlier factor LOF of each data point. If the LOF value of a data point is greater than a preset threshold, it is removed from the data set; the specific formula is: Where P is the data point to be detected, LOF(P) is the LOF value of data point P, N(P) is the set of neighborhood data points of P, and ρ(P) is the local reachability density of data point P; The feature selection includes: using a feature selection method based on mutual information, calculating the mutual information between each feature and the bid evaluation result, selecting features with higher mutual information values, and avoiding features that cause bias; the specific formula of the mutual information value is: Where I(X;Y) is the mutual information value, X is the feature, x is the value of feature X, Y is the evaluation result, y is the value of evaluation result Y, p(x,y) is the joint probability distribution of X=x and Y=y, p(x) and p(y) are the marginal probability distributions of X and Y respectively.

3. A method for improving bid evaluation accuracy according to claim 2, characterized in that: The bias-insensitive learning models include a random forest model and a gradient boosting tree model.

4. A method for improving bid evaluation accuracy according to claim 3, characterized in that: The process of fusing the learning models through the multi-model fusion strategy includes: A weighted average method is used to combine several bias-insensitive learning models. Different weights are assigned to each model according to its performance on the validation set to reduce the bias of a single learning model. The fusion model is: Where i is the bidder index, j is the learning model index; S i is the evaluation score after fusion, S ij is the evaluation score of the j-th learning model for the i-th bidder, ω j is the weight of the jth learning model, and n is the number of learning models.

5. A method for improving bid evaluation accuracy according to claim 4, characterized in that: The implementation process of the fairness constraint is to add fairness constraints during the model training process, which specifically includes: A fairness indicator is defined, where the fairness indicator is the variance of the score differences between different bidders; Establish the fairness indicator formula: Among them, F is the fairness index, m is the number of bidders, S i and S k are the evaluation scores of the i-th bidder and the k-th bidder respectively; A fairness index threshold is preset, and the fairness index F is compared with the fairness index threshold; if the fairness index F is greater than the fairness index threshold, it is judged that the fusion model is unfair; The parameters of the fusion model are adjusted based on the fairness index F until the fairness index F is less than or equal to the fairness index threshold.

6. A method for improving bid evaluation accuracy according to claim 5, characterized in that: The implementation process of the cross validation is to use K-fold cross validation to evaluate the performance stability of the fusion model, which specifically includes: The evaluation scores of each bidder in history are used as training data, and the training data set is divided into K subsets of equal size. One subset is selected as the validation set in turn, and the remaining K-1 subsets are used as the training set for model training and validation; Calculate the mean and standard deviation of K validation results; A standard deviation threshold is set in advance, and the standard deviation of the verification result is compared with the preset standard deviation threshold; if the standard deviation of the verification result is greater than the standard deviation threshold, it is judged that the fusion model is unstable; The parameters of the fusion model are adjusted based on the standard deviation of the verification results until the standard deviation of the verification results is less than or equal to the standard deviation threshold.

7. A method for improving bid evaluation accuracy according to claim 6, characterized in that: The implementation process of the bias detection includes: Establish a real-time monitoring system to continuously monitor the performance and scoring results of the fusion model; Regularly evaluate the model with new data and calculate bias indicators; The calculation formula of the bias index is: Among them, B is the bias index, S G1 and S G2 are the average scores of different bidder combinations, is the average score of all bidders; A bias threshold is set in advance, and the bias index B is compared with the preset bias threshold; if the bias index B is greater than the bias threshold, it is determined that the fusion model has bias; The parameters of the fusion model are adjusted based on the bias index B until the bias index B is less than or equal to the bias threshold.

8. A method for improving bid evaluation accuracy according to claim 7, characterized in that: The process of updating the fusion model based on the evaluation result includes: Based on the results of fairness constraints, cross-validation, and bias detection, adjust the model parameters or select a new model for fusion to improve the performance and accuracy of the model.

9. A method for improving bid evaluation accuracy according to claim 8, characterized in that: The process of adding the expert review mechanism and adjusting the bid evaluation scores includes: The experts adjust the evaluation scores automatically calculated by the fusion model and give reasons for the adjustments. The final evaluation score is obtained by weighting the evaluation score automatically calculated by the fusion model and the evaluation score adjusted by the experts. The specific formula is: FS i =α*S i +(1-α)*ES i Among them, FS i is the final evaluation score of the i-th bidder, S i is the evaluation score of the i-th bidder automatically calculated based on the fusion model, ES i is the evaluation score of the ith bidder based on expert adjustment, α is the scoring weight of the fusion model, and its value range is {0, 1}.

10. A system for improving the accuracy of bid evaluation, based on the method of claims 1-9, characterized in that: include: Data preprocessing module, used to preprocess bidding document data, including data cleaning and feature selection; A model fusion module, which is connected to the data preprocessing module, establishes several bias-insensitive learning models, and fuses the learning models through a multi-model fusion strategy; A model evaluation module, which is connected to the model fusion module and is used to evaluate the performance of the fusion model in real time, including fairness constraints, cross-validation and bias detection, and update the fusion model based on the evaluation results; A bid evaluation calculation module, which is connected to the data preprocessing module and the model fusion module, substitutes the characteristic information corresponding to each bid document data into the fusion model, and calculates the bid evaluation score of each bidder; An expert review module, which is connected to the bid evaluation calculation module and is used to add an expert review mechanism to adjust the bid evaluation scores; A result archiving module is connected to the expert review module to archive the adjusted final bid evaluation scores.

Images