Heterogeneous information standardization processing method for massive bidding data

Abstract

This invention discloses a method for standardizing heterogeneous information processing of massive bidding data, belonging to the field of bidding data processing technology. This method collects massive amounts of multi-source heterogeneous bidding data from enterprise internal bidding documents, public resource trading platforms, electronic bidding systems, and public resource trading center result announcement platforms, and constructs a multi-dimensional basic dataset. By constructing a multi-dimensional basic dataset through multi-source data collection channels, this invention achieves comprehensive aggregation of massive heterogeneous data. Then, through targeted data preprocessing, standardization operations such as unit unification, special symbol removal, invalid data removal, and missing data supplementation are performed to achieve effective fusion and purification of multi-source heterogeneous data, reducing the interference of redundant and missing data on subsequent analysis. It also allows dispersed multi-dimensional data to form effective correlations, solving the core pain points of existing methods that cannot effectively integrate data and are difficult to support data analysis and risk assessment.

Description

Technical Field

[0001] This invention relates to the field of bidding data processing technology, specifically a method for standardizing heterogeneous information processing of massive bidding data. Background Technology

[0002] With the standardization and digital transformation of the bidding market, a massive amount of bidding data has been generated through various channels such as internal bidding by enterprises, public resource trading platforms, and electronic bidding evaluation systems. This data covers multiple dimensions, including information on the bidding party, the bidder, the project, the evaluation results, and the reasons for the rejection of bids. It also exhibits significant heterogeneity, with scattered data sources, inconsistent formats, mixed data types, a lot of redundant information, and frequent missing information.

[0003] Existing bidding data processing methods generally suffer from low data standardization and a lack of unified processing standards for multi-source heterogeneous data. This results in ineffective data integration, making it difficult to support subsequent data analysis and risk assessment. Furthermore, the accuracy of anomaly identification is insufficient. Traditional methods often rely on single indicators for judgment, failing to comprehensively consider multi-dimensional correlation features such as pricing, bidding structure, and rejected bid records, leading to misjudgments and omissions. Simultaneously, existing methods struggle to accurately identify violations such as bid rigging, fraudulent bidding, and malicious rejection, and cannot scientifically construct supplier credit profiles. Consequently, companies lack objective and accurate evaluation criteria during supplier selection, making it difficult to choose suitable suppliers. This increases bidding risks, impacts bidding efficiency and procurement quality, and fails to provide strong support for stable supply chain and compliant operations. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for standardizing heterogeneous information processing of massive bidding data, in order to solve the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for standardizing heterogeneous information processing of massive bidding data, comprising the following steps:

[0006] S1. Collect massive amounts of heterogeneous bidding data from multiple sources through internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and the public resource trading center's bid result announcement platform, and construct a multi-dimensional basic dataset;

[0007] S2. Extract the actual quotation amount from the target supplier based on a multi-dimensional basic dataset. Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) And perform preprocessing;

[0008] S3. Using a convolutional neural network, construct an AI neural network model. Input the multi-dimensional basic dataset into the AI ​​neural network model to output the analysis results of anomaly prediction in bidding data. Based on the multi-dimensional basic dataset, calculate the target supplier's price deviation coefficient. Bidding Concentration Index Correlation coefficient of rejected bids ;

[0009] S4. Price deviation coefficient based on target supplier Bidding Concentration Index Correlation coefficient of rejected bids Correlate and construct a matching index for target suppliers. And the credit profile volatility prediction coefficient W, and evaluate and optimize it. In a preferred embodiment, S1 includes:

[0010] S11. Collect data from internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and public resource trading center's bid result announcement platforms to obtain bidding party data, bidder data, bidding project data, bid evaluation data, and winning bid data, and establish a multi-dimensional basic dataset.

[0011] In a preferred embodiment, S2 includes:

[0012] S21: The tendering party's data includes the benchmark bid amount. The data was collected from internal bidding documents, which are formal bidding documents prepared by the company based on project requirements, industry standards, and cost estimates. These documents are available in both electronic and paper formats, covering all valid bidding projects. During the data collection process, electronic bidding documents were retrieved from the company's internal bidding management platform, and offline paper bidding documents were scanned and digitized. The benchmark price was located by searching for keywords such as "benchmark price," "budget amount," "control price," and "reference price," and the benchmark price amount was obtained. ;

[0013] Bidder data includes the total number of target supplier bidding regions (M), the total number of valid bids (N), the total number of products involved in the target supplier's bids (K), and the text of the reasons for the target supplier's previous bid rejections. Furthermore, data is collected through a public resource trading platform. During collection, the platform uses the target supplier's name and unique unified social credit code to filter all project records in which the target supplier participated in bidding. The bidding region information for each project is extracted to obtain the total number of bidding regions (M) for the target supplier. All bidding projects in which the target supplier participated are queried, and the number of projects is counted to obtain the total number of valid bidding projects (N). Product information for all valid bidding projects in which the target supplier participated is extracted to obtain the total number of products involved in the target supplier's bids (K). The historical bidding records of the target supplier are queried using its unique identifier, and projects marked as "rejected" are filtered out. The reason text for each rejected project is extracted to form a complete historical reason text for rejected bids of the target supplier. ;

[0014] The bidding project data includes the actual bid amount from the target supplier. The data was collected by analyzing internal bidding documents. During the data collection process, the electronic version of the corresponding bidding project was accessed. Offline paper bidding documents were converted into processable electronic files using high-definition scanning and OCR recognition technology. By searching for keywords such as "actual price," "bid price," and "supplier price," the target supplier's price was located, and the actual price amount of the target supplier was obtained. ;

[0015] The evaluation data includes the text of the reasons for the current rejection of the target supplier's bid. The data is collected by connecting to an electronic bidding platform. During the data collection process, the data interface of the electronic bidding platform is used to crawl and obtain the complete bidding records of the target supplier's current bidding projects, locate the relevant modules for bid rejection and detailed descriptions, extract the text of the reasons for bid rejection, and obtain the text of the reasons for the target supplier's current bid rejection. ;

[0016] The data information for successful bids includes a text library of common reasons for bid rejection in the industry. Data is collected through the public resource trading center's bidding result announcement platform. During the data collection process, the platform's backend data interface is connected to collect the text of reasons for bid rejection for publicly disclosed rejected projects within the platform in batches. This text is then organized and analyzed to obtain a database of common reasons for bid rejection in the industry. ;

[0017] S22. Based on the actual quoted amount from the target supplier and benchmark quote amount All collected data was uniformly converted to "yuan" as the sole unit of measurement. If the original data unit was "ten thousand yuan" or "thousand yuan," it was multiplied by 10,000 and 1,000 respectively for conversion. The conversion process was recorded, and the unit of measurement was marked after conversion. All non-numerical special symbols in the collected data were completely removed, including currency symbols, thousandths separators, spaces, and text notes, retaining only pure numerical values ​​and decimal points. After removal, the collected data underwent preliminary verification, and the monetary values ​​after removing special symbols were uniformly retained to two decimal places. Missing quotation numbers found during the collection process were also addressed. According to the data, the average price of bidding projects of the same category and in the same period is used for filling. The same category is based on the product and service categories of the bidding project, and the same period is based on the release time of the bidding project. Before filling, it is necessary to confirm that the sample size of the same category and the same period of bidding projects is ≥3. If the sample size is <3, the same period range is expanded to 3 calendar months. If the same period range is still <3 after expanding the same period range, the average price of the same type of project of the target supplier in the next 3 months is used for filling. After filling, the data is marked "Data Supplement" and the supplementary basis. The monetary data is linked with the corresponding project number, target supplier ID, collection source, and preprocessing time for archiving.

[0018] Based on the total number of target supplier bidding regions M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bidding K;

[0019] For the total number of valid bidding projects N collected, invalid bidding project records are removed, including projects that did not submit complete bidding documents, failed the bid document review, were voluntarily withdrawn, had irregularities in the bidding documents, were submitted after the deadline, or did not meet the qualification requirements. Valid bidding projects that have passed the preliminary review and meet the bidding requirements are retained for counting.

[0020] For the total number of bidding areas M collected, invalid area records are removed, including bidding areas that have not been approved, false area entries, areas corresponding to withdrawn bids, and abnormal areas in the bidding scope. The bidding areas that the target supplier actually participated in and approved are counted again.

[0021] For the total number K of products involved in the bidding collected, product categories with abnormal bidding requirements are removed, including product categories that are outside the scope of bidding products, have mismatched specifications and models, or do not meet the qualification requirements. Product categories that meet the bidding requirements and have passed the review are retained for counting.

[0022] After counting is completed, the total number of target supplier bidding areas M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bids K are verified to confirm that the data logic after removing invalid and duplicate records is reasonable. After verification, the valid counting basis is marked, including the counting range, details of invalid and duplicate records removed, verification personnel and verification time. For missing counting data found during the collection process, the average data of the corresponding similar bidding projects of the target supplier within the past 3 months is used to supplement it. When supplementing, it is necessary to confirm that there are similar bidding projects in the past 3 months. If there are no similar bidding projects in the past 3 months, the average data of similar projects of suppliers of the same industry and scale during the same period is used to supplement it. The supplemented data and the basis for supplementation are marked.

[0023] Based on the text of the current reasons for the target supplier's rejection of the bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The text containing the reasons for the current rejection of bids by the target suppliers was collected. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) A comprehensive process is performed to remove invalid characters from the text, including garbled text, special symbols, irrelevant notes, blank characters, and line breaks. Valid text content related to the reason for rejection is retained. Typos and grammatical errors in the text are corrected. The preprocessed text data is then uniformly converted to UTF-8 encoding format.

[0024] In a preferred embodiment, S3 includes:

[0025] S31 is used to construct an AI neural network model using a convolutional neural network, train and test the AI ​​neural network model with a multi-dimensional basic dataset, and use the trained AI neural network model as a bidding processing model. At the same time, the intermediate layer output of the bidding processing model is used as a feature vector to identify the information features in the multi-dimensional basic dataset, and the trained bidding processing model is used as a data prediction tool.

[0026] In a preferred embodiment, S3 further includes:

[0027] S32. Extract the actual quoted amount from the target supplier. Benchmark quotation amount Combined and normalized, the deviation of a single quote from the target supplier is calculated using the following formula. :

[0028] ;

[0029] S33. Extract the total number of valid bids N, and output the average deviation of the target supplier's bid. It can be calculated using the following formula:

[0030] ;

[0031] In the formula, N represents the total number of valid bids submitted by the target supplier within the statistical period. This represents the bid amount from the target supplier in the i-th bidding project;

[0032] S34. Deviation of single quotation from the target supplier Average deviation from the target supplier's quote After normalization, the deviation coefficient of the target supplier's quote is calculated using the following formula. :

[0033] ;

[0034] In the formula, Kb represents the deviation of the quoted price from the tolerance threshold, which is an empirical constant. This represents the maximum deviation amplification factor, which is an empirical constant.

[0035] By setting a preset deviation threshold Q, and calculating the deviation coefficient of the target supplier's quotation... Compare with the deviation threshold Q, including:

[0036] when When the value is less than Q, it indicates that the target supplier's quoted price is reasonable.

[0037] When Q≤ A price less than 2Q indicates a Level 1 deviation from the target supplier's quote, suggesting price fluctuations. This necessitates monitoring the target supplier, highlighting the actual quote data, correlating it with historical average quotes and the reasons for deviations, and archiving it in the supplier quote anomaly file, based on industry benchmark quote amounts. Provide suppliers with targeted suggestions for optimizing their bids, and explicitly require that the deviation between their subsequent bids and historical averages be reduced by more than 30%, thereby guiding suppliers to optimize their bidding strategies and standardize cost accounting.

[0038] when When the deviation is ≥2Q, it indicates that the target supplier's quotation has a second-level deviation. The second level is greater than the first level, and the quotation is abnormal. The quotation data should be marked as abnormal, the validity of the quotation should be suspended, and the quotation data and deviation details should be synchronized to the procurement management department and the bid evaluation committee to trigger a quotation abnormality risk warning. The supplier should be required to submit a special rectification plan and quotation optimization plan within a time frame. After the rectification is completed, the quotation data should be recalculated. The quotation validity should be restored after the quotation deviation drops below Q.

[0039] In a preferred embodiment, S3 further includes:

[0040] S35. By extracting the total number of bidding regions M of the target suppliers, the regional bidding concentration is calculated using the following formula. :

[0041] ;

[0042] In the formula, This represents the proportion of the target supplier's bids in the j-th region to the total number of bids, where j represents the region number.

[0043] S36. By extracting the total number K of products involved in the target supplier's bids, the category bidding concentration is calculated using the following formula. :

[0044] ;

[0045] In the formula, This represents the proportion of the target supplier's bids in the k-th category out of the total number of bids, where k represents the category number;

[0046] S37. Concentration of regional bids from target suppliers Category concentration of bids with target suppliers The bid concentration index of the target supplier is obtained by correlation and normalization using the following formula. ;

[0047] ;

[0048] In the formula, and Represents the weighting coefficients, satisfying + =1;

[0049] S39. By setting a concentration threshold R, the bid concentration coefficient of the target supplier is... Compare with the concentration threshold R, including:

[0050] when When <R, it indicates that the target supplier matches the business layout;

[0051] If R≤ When the deviation is less than 1.5R, it indicates that the target supplier's own business has a first-level deviation, and the target supplier needs to be controlled, verified, and the risks mitigated.

[0052] when When the deviation is ≥1.5R, it indicates that the target supplier's own business has a second-level deviation. The second level is greater than the first level. Optimization strategies will be implemented, including simultaneously sending early warning instructions to the procurement management department and the bid evaluation committee, suspending the bidding qualification for this tender, and including the supplier in the key risk monitoring list for key investigation of bid rigging and collusion.

[0053] In a preferred embodiment, S3 further includes:

[0054] S310. Extract the text explaining the current reasons for bid rejection from the target supplier. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The BERT model is used to extract the text of the reasons for the current rejection of the target supplier's bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The numerical feature vectors obtained by converting the text into the reasons for the current rejection by the target supplier are then processed into vectors. The numerical feature vector is obtained by vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry :

[0055] ;

[0056] ;

[0057] ;

[0058] In the formula, (.) represents the text vectorization function;

[0059] S311. Numerical feature vector obtained by vectorizing the text of the reason for the current rejection of the target supplier's bid. The numerical feature vector obtained after vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry By combining these factors, the similarity between the target supplier's current bid rejection behavior and its own historical bid rejection behavior can be calculated. Similarity to current rejected bids and common industry-wide rejected bid standards :

[0060] ;

[0061] ;

[0062] In the formula, Represented as a cosine similarity function, it measures the cosine of the angle between two vectors, reflecting the relevance of their content.

[0063] S312. Compare the current rejection of bids by the target supplier with the similarity of the target supplier's past rejection behavior. Similarity to current rejected bids and common industry-wide rejected bid standards After combining and normalizing the data, the correlation coefficient of the rejected standards is calculated. :

[0064] ;

[0065] In the formula, and Represents the weighting coefficients, satisfying =1;

[0066] By setting a preset rejection correlation threshold T, and setting the rejection correlation coefficient of the target supplier... Compare with the rejection threshold T, including:

[0067] when When T < T, it indicates that the risk of the target supplier's bid being rejected is low;

[0068] When T≤ When the value is less than 1.5T, it indicates that the target supplier faces a Level 1 risk of bid rejection. The target supplier is required to submit an explanation of the reasonableness of the reasons for bid rejection within the specified period, focusing on clarifying the specific causes of the reasons for bid rejection.

[0069] when When the value is ≥1.5T, it indicates that the target supplier has a level 2 risk of bid rejection. Level 2 is greater than level 1, and the risk of bid rejection is high. It is necessary to combine the supplier's historical bid rejection records, qualification documents, and compliance status to separately verify the root cause of the abnormal bid rejection association, focus on analyzing the reasons for strong bid rejection association, and investigate illegal bidding, false bidding, and malicious bid rejection.

[0070] In a preferred embodiment, S4 includes:

[0071] S41. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids After normalization, the matching index of the target supplier is calculated using the following formula. :

[0072] ;

[0073] S41. By setting a preset matching threshold X, and setting the matching index of the target supplier... Compare with the matching threshold X, including:

[0074] When the matching index of the target supplier > When the matching threshold X is reached, it indicates that the target supplier's bidding data is normal, the overall performance of price reasonableness, bid dispersion, and bid rejection compliance meets the standards, and the target supplier's integrity level and procurement suitability meet the requirements.

[0075] when <X indicates that the target supplier's bidding data is abnormal, generating the first strategy, which includes directly suspending the bidder's eligibility for this bidding, adding the supplier to the key risk monitoring list, focusing on investigating suspected illegal bidding, and separately verifying the root cause of abnormal bidding associations based on the target supplier's historical bid rejection records, qualification documents, and compliance status, focusing on analyzing the reasons for strong bid rejection associations, investigating illegal bidding, false bidding, and malicious bid rejection violations, recording them in the supplier's credit file, and simultaneously restricting their subsequent bidding rights.

[0076] In a preferred embodiment, S4 further includes:

[0077] S46. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids After normalization, the credit profile fluctuation prediction coefficient W of the target supplier is calculated using the following related formula:

[0078] .

[0079] In a preferred embodiment, S4 further includes:

[0080] S47. A preset fluctuation threshold Y is used to compare the predicted fluctuation coefficient W of the target supplier's credit profile with the fluctuation threshold Y to obtain the prediction results, including:

[0081] If W≤Y, it means that the target supplier's credit profile is stable, its bidding behavior is regular, and its performance risk is controllable. The current target supplier rating and cooperation strategy will be maintained.

[0082] If W > Y, it indicates that the target supplier's credit profile is fluctuating drastically, its bidding behavior is abnormal, and its potential risk of default is high. This triggers a credit fluctuation warning instruction and generates a second strategy, which includes implementing key monitoring, review of bidding qualifications, performance risk warnings, or phased access restriction control measures on the target supplier.

[0083] The method for standardizing heterogeneous information processing of massive bidding data provided by this invention has the following beneficial effects:

[0084] 1. By collecting data from multiple sources, a multi-dimensional basic dataset is constructed to achieve comprehensive aggregation of massive heterogeneous data; then, through targeted data preprocessing, the actual quotation amount of the target supplier is analyzed. Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) Standardized operations such as unifying units, removing special symbols, removing invalid data, and supplementing missing data are performed to establish unified data processing specifications. This enables the effective fusion and purification of multi-source heterogeneous data, which not only significantly improves the standardization and data quality of bidding data and reduces the interference of redundant and missing data on subsequent analysis, but also allows scattered multi-dimensional data to form effective correlations. This provides high-quality and reusable data support for subsequent AI neural network model training and various coefficient calculations, breaks down data silos, improves data utilization, and solves the core pain points of existing methods that cannot effectively integrate data and support data analysis and risk assessment.

[0085] 2. Construct an AI neural network model using convolutional neural networks, combine it with a multi-dimensional basic dataset to output anomaly prediction results for bidding data, and simultaneously calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids Three core indicators enable comprehensive analysis of multi-dimensional characteristics. By constructing a matching index S and a credit profile fluctuation prediction coefficient W, a comprehensive evaluation and optimization of suppliers is conducted. This breaks through the limitations of traditional single-indicator judgment and achieves multi-dimensional and all-round assessment of supplier bidding behavior. It can accurately identify anomalies in bidding data, improve the accuracy of anomaly identification, and effectively avoid the risks of misjudgment and omission of violations such as bid rigging, false bidding, and malicious rejection. It can also scientifically construct supplier credit profiles, providing objective and accurate evaluation basis for enterprises to select suppliers during bidding, helping enterprises to select suitable suppliers, reduce bidding risks, improve bidding efficiency and procurement quality, and provide strong support for the stability and compliant operation of enterprise supply chains. Attached Figure Description

[0086] Figure 1 This is a schematic diagram illustrating the steps of the heterogeneous information standardization processing method for massive bidding data according to the present invention. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] Example 1

[0089] Please see Figure 1 This invention provides a method for standardizing heterogeneous information processing of massive bidding data, comprising the following steps:

[0090] S1. Collect massive amounts of heterogeneous bidding data from multiple sources through internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and the public resource trading center's bid result announcement platform, and construct a multi-dimensional basic dataset;

[0091] S2. Extract the actual quotation amount from the target supplier based on a multi-dimensional basic dataset. Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) And perform preprocessing;

[0092] S3. Using a convolutional neural network, construct an AI neural network model. Input the multi-dimensional basic dataset into the AI ​​neural network model to output the analysis results of anomaly prediction in bidding data. Based on the multi-dimensional basic dataset, calculate the target supplier's price deviation coefficient. Bidding Concentration Index Correlation coefficient of rejected bids ;

[0093] S4. Price deviation coefficient based on target supplier Bidding Concentration Index Correlation coefficient of rejected bids Correlate and construct a matching index for target suppliers. And the credit profile volatility prediction coefficient W, and evaluate and optimize it.

[0094] In this embodiment, a multi-dimensional basic dataset is constructed through multiple data acquisition channels to achieve comprehensive aggregation of massive heterogeneous data. Then, targeted data preprocessing is carried out to analyze the actual quotation amount of the target supplier. Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) Standardized operations such as unifying units, removing special symbols, removing invalid data, and supplementing missing data are performed to establish unified data processing specifications. This enables the effective fusion and purification of multi-source heterogeneous data, which not only significantly improves the standardization and data quality of bidding data and reduces the interference of redundant and missing data on subsequent analysis, but also allows scattered multi-dimensional data to form effective correlations. This provides high-quality and reusable data support for subsequent AI neural network model training and various coefficient calculations, breaks down data silos, improves data utilization, and solves the core pain points of existing methods that cannot effectively integrate data and support data analysis and risk assessment.

[0095] An AI neural network model is constructed using convolutional neural networks. Combined with a multi-dimensional basic dataset, it outputs anomaly prediction results for bidding data and calculates the deviation coefficient of the target supplier's bid. Bidding Concentration Index Correlation coefficient of rejected bids Three core indicators enable comprehensive analysis of multi-dimensional features;

[0096] Furthermore, by constructing a matching index S and a credit profile fluctuation prediction coefficient W, a comprehensive evaluation and optimization of suppliers is conducted. This breaks through the limitations of traditional single-indicator judgment and achieves multi-dimensional and all-round analysis of supplier bidding behavior. It can accurately identify anomalies in bidding data, improve the accuracy of anomaly identification, and effectively avoid the risks of misjudgment and omission of violations such as bid rigging, false bidding, and malicious rejection. It can also scientifically construct supplier credit profiles, providing objective and accurate evaluation basis for enterprises to select suppliers during bidding, helping enterprises to select suitable suppliers, reduce bidding risks, improve bidding efficiency and procurement quality, and provide strong support for the stability and compliant operation of enterprise supply chains.

[0097] Example 2

[0098] This embodiment is an explanation of Embodiment 1; please refer to it. Figure 1 Specifically, S1 includes:

[0099] S11. Collect data from internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and public resource trading center's bid result announcement platforms to obtain bidding party data, bidder data, bidding project data, bid evaluation data, and winning bid data, and establish a multi-dimensional basic dataset.

[0100] In this embodiment, the multi-dimensional basic dataset covers the entire bidding process, avoiding analytical bias caused by a single data source. It covers the entire lifecycle of bidding, from "bidding, tendering, evaluation, to acceptance," ensuring that the dataset is complete and seamless, providing a complete input foundation for subsequent AI analysis and quantitative calculations.

[0101] Example 3

[0102] This embodiment is an explanation of Embodiment 1. Specifically, S2 includes:

[0103] S21: The tendering party's data includes the benchmark bid amount. The data was collected from internal bidding documents, which are formal bidding documents prepared by the company based on project requirements, industry standards, and cost estimates. These documents are available in both electronic and paper formats, covering all valid bidding projects. During the data collection process, electronic bidding documents were retrieved from the company's internal bidding management platform, and offline paper bidding documents were scanned and digitized. The benchmark price was located by searching for keywords such as "benchmark price," "budget amount," "control price," and "reference price," and the benchmark price amount was obtained. ;

[0104] Bidder data includes the total number of target supplier bidding regions (M), the total number of valid bids (N), the total number of products involved in the target supplier's bids (K), and the text of the reasons for the target supplier's previous bid rejections. Furthermore, data is collected through a public resource trading platform. During collection, the platform uses the target supplier's name and unique unified social credit code to filter all project records in which the target supplier participated in bidding. The bidding region information for each project is extracted to obtain the total number of bidding regions (M) for the target supplier. All bidding projects in which the target supplier participated are queried, and the number of projects is counted to obtain the total number of valid bidding projects (N). Product information for all valid bidding projects in which the target supplier participated is extracted to obtain the total number of products involved in the target supplier's bids (K). The historical bidding records of the target supplier are queried using its unique identifier, and projects marked as "rejected" are filtered out. The reason text for each rejected project is extracted to form a complete historical reason text for rejected bids of the target supplier. ;

[0105] The bidding project data includes the actual bid amount from the target supplier. The data was collected by analyzing internal bidding documents. During the data collection process, the electronic version of the corresponding bidding project was accessed. Offline paper bidding documents were converted into processable electronic files using high-definition scanning and OCR recognition technology. By searching for keywords such as "actual price," "bid price," and "supplier price," the target supplier's price was located, and the actual price amount of the target supplier was obtained. ;

[0106] The evaluation data includes the text of the reasons for the current rejection of the target supplier's bid. The data is collected by connecting to an electronic bidding platform. During the data collection process, the data interface of the electronic bidding platform is used to crawl and obtain the complete bidding records of the target supplier's current bidding projects, locate the relevant modules for bid rejection and detailed descriptions, extract the text of the reasons for bid rejection, and obtain the text of the reasons for the target supplier's current bid rejection. ;

[0107] The data information for successful bids includes a text library of common reasons for bid rejection in the industry. Data is collected through the public resource trading center's bidding result announcement platform. During the data collection process, the platform's backend data interface is connected to collect the text of reasons for bid rejection for publicly disclosed rejected projects within the platform in batches. This text is then organized and analyzed to obtain a database of common reasons for bid rejection in the industry. ;

[0108] S22. Based on the actual quoted amount from the target supplier and benchmark quote amount All collected data was uniformly converted to "yuan" as the sole unit of measurement. If the original data unit was "ten thousand yuan" or "thousand yuan," it was multiplied by 10,000 and 1,000 respectively for conversion. The conversion process was recorded, and the unit of measurement was marked after conversion. All non-numerical special symbols in the collected data were completely removed, including currency symbols, thousandths separators, spaces, and text notes, retaining only pure numerical values ​​and decimal points. After removal, the collected data underwent preliminary verification, and the monetary values ​​after removing special symbols were uniformly retained to two decimal places. Missing quotation numbers found during the collection process were also addressed. According to the data, the average price of bidding projects of the same category and in the same period is used for filling. The same category is based on the product and service categories of the bidding project, and the same period is based on the release time of the bidding project. Before filling, it is necessary to confirm that the sample size of the same category and the same period of bidding projects is ≥3. If the sample size is <3, the same period range is expanded to 3 calendar months. If the same period range is still <3 after expanding the same period range, the average price of the same type of project of the target supplier in the next 3 months is used for filling. After filling, the data is marked "Data Supplement" and the supplementary basis. The monetary data is linked with the corresponding project number, target supplier ID, collection source, and preprocessing time for archiving.

[0109] Based on the total number of target supplier bidding regions M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bidding K;

[0110] For the total number of valid bidding projects N collected, invalid bidding project records are removed, including projects that did not submit complete bidding documents, failed the bid document review, were voluntarily withdrawn, had irregularities in the bidding documents, were submitted after the deadline, or did not meet the qualification requirements. Valid bidding projects that have passed the preliminary review and meet the bidding requirements are retained for counting.

[0111] For the total number of bidding areas M collected, invalid area records are removed, including bidding areas that have not been approved, false area entries, areas corresponding to withdrawn bids, and abnormal areas in the bidding scope. The bidding areas that the target supplier actually participated in and approved are counted again.

[0112] For the total number K of products involved in the bidding collected, product categories with abnormal bidding requirements are removed, including product categories that are outside the scope of bidding products, have mismatched specifications and models, or do not meet the qualification requirements. Product categories that meet the bidding requirements and have passed the review are retained for counting.

[0113] After counting is completed, the total number of target supplier bidding areas M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bids K are verified to confirm that the data logic after removing invalid and duplicate records is reasonable. After verification, the valid counting basis is marked, including the counting range, details of invalid and duplicate records removed, verification personnel and verification time. For missing counting data found during the collection process, the average data of the corresponding similar bidding projects of the target supplier within the past 3 months is used to supplement it. When supplementing, it is necessary to confirm that there are similar bidding projects in the past 3 months. If there are no similar bidding projects in the past 3 months, the average data of similar projects of suppliers of the same industry and scale during the same period is used to supplement it. The supplemented data and the basis for supplementation are marked.

[0114] Based on the text of the current reasons for the target supplier's rejection of the bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The text containing the reasons for the current rejection of bids by the target suppliers was collected. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) A comprehensive process is performed to remove invalid characters from the text, including garbled text, special symbols, irrelevant notes, blank characters, and line breaks. Valid text content related to the reason for rejection is retained. Typos and grammatical errors in the text are corrected. The preprocessed text data is then uniformly converted to UTF-8 encoding format.

[0115] In this embodiment, the actual quoted amount from the target supplier is... Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) Three types of heterogeneous data are processed using differentiated rules;

[0116] Regarding the actual quoted amount from the target supplier and benchmark quote amount Data preprocessing: Standardize units, format, and precision; remove special symbols; fill missing values ​​with the average value of the same product category during the same period to avoid outliers or missing values ​​affecting the price deviation coefficient. calculate.

[0117] For the total number of target supplier bidding areas M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bidding K, the data preprocessing includes: deduplication, removal of invalid records, and filling in missing data with the average of similar data from the past 3 months to ensure the authenticity of the bidding area, project, and product counts.

[0118] The text regarding the reasons for the current rejection of the target supplier's bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) Data preprocessing: Unify UTF-8 encoding, remove invalid characters (spaces, special symbols), and archive the data with supplier ID, time, and project to make the text of the reasons for bid rejection traceable, searchable, and comparable;

[0119] Completely eliminate heterogeneous differences in multi-source data, transform "messy data" into "standard data", adapt to subsequent AI models and mathematical calculations, classify and preprocess the data to achieve unified format and improved quality of heterogeneous data, eliminate data noise and missing data, and adapt to the needs of subsequent AI models and quantitative calculations.

[0120] Example 4

[0121] This embodiment is an explanation of embodiment 1. Specifically, S3 includes:

[0122] S31 is used to construct an AI neural network model using a convolutional neural network, train and test the AI ​​neural network model with a multi-dimensional basic dataset, and use the trained AI neural network model as a bidding processing model. At the same time, the intermediate layer output of the bidding processing model is used as a feature vector to identify the information features in the multi-dimensional basic dataset, and the trained bidding processing model is used as a data prediction tool.

[0123] In this embodiment, a convolutional neural network is used, which is adept at feature extraction from structured data. It is suitable for mining structured features such as amount and count in bidding data. A multi-dimensional basic dataset is used as the training and testing set to allow the model to learn the data patterns of normal and abnormal bidding, forming a dedicated model for bidding processing. The output of the intermediate layer of the model is extracted as a feature vector to automatically capture hidden features that cannot be discovered by humans (such as price fluctuation patterns and abnormal bidding layout patterns), improve the accuracy of anomaly identification, replace manual review, and increase the processing speed by more than 100 times. It is suitable for the real-time analysis needs of massive bidding data and avoids human subjective error.

[0124] This embodiment is an explanation of embodiment 1. Specifically, S3 further includes:

[0125] S32. Extract the actual quoted amount from the target supplier. Benchmark quotation amount Combined and normalized, the deviation of a single quote from the target supplier is calculated using the following formula. :

[0126] ;

[0127] S33. Extract the total number of valid bids N, and output the average deviation of the target supplier's bid. It can be calculated using the following formula:

[0128] ;

[0129] In the formula, N represents the total number of valid bids submitted by the target supplier within the statistical period. This represents the bid amount from the target supplier in the i-th bidding project;

[0130] S34. Deviation of single quotation from the target supplier Average deviation from the target supplier's quote After normalization, the deviation coefficient of the target supplier's quote is calculated using the following formula. :

[0131] ;

[0132] In the formula, Kb represents the deviation of the quoted price from the tolerance threshold, which is an empirical constant. This represents the maximum deviation amplification factor, which is an empirical constant. Kb is a fixed value preset by the bidding party and evaluation experts based on the project budget and industry profit margin, obtained by collecting data from general bidding control standards in the bidding industry and combining the historical average of price deviations from similar bidding projects. The values ​​are based on the hyperparameter tuning results during the training process of the convolutional neural network model and are fixed empirical constants;

[0133] The following is the deviation of the target supplier's quote. Example table, see Table 1 below;

[0134] Supplier ID Actual price quote (RMB) Benchmark Price (RMB) Single quote deviation Number of valid bids N Mean of price deviation Price deviation coefficient GYS001 98000 100000 0.02 5 0.03 0.026 GYS002 125000 100000 0.25 6 0.18 0.218 GYS003 72000 100000 0.28 4 0.22 0.256 GYS004 180000 100000 0.80 5 0.35 0.575 GYS005 55000 100000 0.45 3 0.32 0.389

[0135] Table 1

[0136] By setting a preset deviation threshold Q, and calculating the deviation coefficient of the target supplier's quotation... Compare with the deviation threshold Q, including:

[0137] when When the value is less than Q, it indicates that the target supplier's quoted price is reasonable.

[0138] When Q≤ A price less than 2Q indicates a Level 1 deviation from the target supplier's quote, suggesting price fluctuations. This necessitates monitoring the target supplier, highlighting the actual quote data, correlating it with historical average quotes and the reasons for deviations, and archiving it in the supplier quote anomaly file, based on industry benchmark quote amounts. Provide suppliers with targeted suggestions for optimizing their bids, and explicitly require that the deviation between their subsequent bids and historical averages be reduced by more than 30%, thereby guiding suppliers to optimize their bidding strategies and standardize cost accounting.

[0139] when When the deviation is ≥2Q, it indicates that the target supplier's quotation has a second-level deviation. The second level is greater than the first level, and the quotation is abnormal. The quotation data should be marked as abnormal, the validity of the quotation should be suspended, and the quotation data and deviation details should be synchronized to the procurement management department and the bid evaluation committee to trigger a quotation abnormality risk warning. The supplier should be required to submit a special rectification plan and quotation optimization plan within a time frame. After the rectification is completed, the quotation data should be recalculated. The quotation validity should be restored after the quotation deviation drops below Q.

[0140] The deviation threshold Q is trained using a convolutional neural network model. During the model training phase, iterative learning is performed on relevant data samples such as supplier historical quotation data, industry benchmark quotations, and quotation deviation characteristics. The model hyperparameters are systematically tuned and analyzed. Finally, the deviation threshold Q is extracted and determined from the results of the hyperparameter tuning of the convolutional neural network model.

[0141] S3 also includes:

[0142] S35. By extracting the total number of bidding regions M of the target suppliers, the regional bidding concentration is calculated using the following formula. :

[0143] ;

[0144] In the formula, This represents the proportion of the target supplier's bids in the j-th region to the total number of bids, where j represents the region number.

[0145] S36. By extracting the total number K of products involved in the target supplier's bids, the category bidding concentration is calculated using the following formula. :

[0146] ;

[0147] In the formula, This represents the proportion of the target supplier's bids in the k-th category out of the total number of bids, where k represents the category number;

[0148] S37. Concentration of regional bids from target suppliers Category concentration of bids with target suppliers The bid concentration index of the target supplier is obtained by correlation and normalization using the following formula. ;

[0149] ;

[0150] In the formula, and Represents the weighting coefficients, satisfying + =1, The tendering party, combining its industry experience in tendering management with its actual procurement control needs, and based on its own level of concern regarding regional layout risks, obtains the necessary approvals. The results are determined by the bidding party based on its industry bidding management experience and actual procurement control needs, taking into account its own level of concern regarding the risks of product category layout.

[0151] The following is the bid concentration index of the target supplier. See Table 2 below for an example table;

[0152] Supplier ID Total number of bidding areas M Regional Bidding Concentration Total number of products in the tender K Category Bidding Concentration Bid Concentration Index GYS001 8 0.12 10 0.10 0.112 GYS002 3 0.35 4 0.32 0.338 GYS003 2 0.48 3 0.45 0.468 GYS004 2 0.52 2 0.50 0.512 GYS005 6 0.22 7 0.19 0.208

[0153] Table 2

[0154] S39. By setting a concentration threshold R, the bid concentration coefficient of the target supplier is... Compare with the concentration threshold R, including:

[0155] when When <R, it indicates that the target supplier matches the business layout;

[0156] If R≤ When the deviation is less than 1.5R, it indicates that the target supplier's own business has a first-level deviation, and the target supplier needs to be controlled, verified, and the risks mitigated.

[0157] when When the deviation is ≥1.5R, it indicates that the target supplier's own business has a secondary deviation. If the secondary deviation is greater than the primary deviation, an optimization strategy will be implemented, including simultaneously sending early warning instructions to the procurement management department and the bid evaluation committee, suspending the bidding qualification for this tender, and including the supplier in the key risk monitoring list for a focused investigation of bid rigging and collusion.

[0158] Among them, the concentration threshold R is based on the full historical bidding data, supplier business layout characteristic data, and the distribution pattern of bidding concentration of similar projects in the industry. It combines the convolutional neural network model to perform deep learning, feature extraction and risk pattern mining on massive historical bidding data. During the model training process, multiple rounds of hyperparameter iteration optimization and fitting verification are carried out around the reasonable range of bidding concentration. At the same time, it takes into account the bidding party's own procurement control standards, project risk level classification requirements and industry bidding supervision norms. Through quantitative differentiation analysis of supplier normal business layout concentration and abnormal concentration, it is finally determined from the model hyperparameter optimization results.

[0159] S3 also includes: Under the premise of a unified calculation method for bid concentration coefficient, selecting suppliers with stable business structures and no abnormal or irregular behavior in recent years as a normal sample set, and conducting statistical analysis on their bid concentration coefficients. Through quantile analysis or the mean and standard deviation method, extracting an upper limit value of concentration that reflects the normal business layout characteristics of the suppliers, and defining this value as the concentration threshold R.

[0160] S310. Extract the text explaining the current reasons for bid rejection from the target supplier. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The BERT model is used to extract the text of the reasons for the current rejection of the target supplier's bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The numerical feature vectors obtained by converting the text into the reasons for the current rejection by the target supplier are then processed into vectors. The numerical feature vector is obtained by vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry :

[0161] ;

[0162] ;

[0163] ;

[0164] In the formula, (.) represents the text vectorization function;

[0165] S311. Numerical feature vector obtained by vectorizing the text of the reason for the current rejection of the target supplier's bid. The numerical feature vector obtained after vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry By combining these factors, the similarity between the target supplier's current bid rejection behavior and its own historical bid rejection behavior can be calculated. Similarity to current rejected bids and common industry-wide rejected bid standards :

[0166] ;

[0167] ;

[0168] In the formula, Represented as a cosine similarity function, it measures the cosine of the angle between two vectors, reflecting the relevance of their content.

[0169] S312. Compare the current rejection of bids by the target supplier with the similarity of the target supplier's past rejection behavior. Similarity to current rejected bids and common industry-wide rejected bid standards After combining and normalizing the data, the correlation coefficient of the rejected standards is calculated. :

[0170] ;

[0171] In the formula, and Represents the weighting coefficients, satisfying =1, where, Based on the degree of risk assessment that the bidding party and procurement management department place on the supplier's past bid rejections, the need to control the recurrence of past bid rejections, and in conjunction with industry bidding supervision requirements and the supplier performance risk assessment system, this was obtained. Based on the bidding party's enforcement of common industry-wide bid rejection standards and the priority of managing industry-standard bid rejection risks, and with reference to similar project evaluation management standards and risk prevention guidelines;

[0172] The following are the rejection correlation coefficients of the target supplier. See Table 3 below for an example table;

[0173] Supplier ID Similarity between current bid rejection and its own past bid rejection behavior Similarity between current rejected bids and common industry-wide rejected bid standards Correlation coefficient of rejected bids GYS001 0.12 0.08 0.104 GYS002 0.35 0.28 0.322 GYS003 0.48 0.42 0.456 GYS004 0.62 0.58 0.604 GYS005 0.75 0.68 0.722

[0174] Table 3

[0175] By setting a preset rejection correlation threshold T, and setting the rejection correlation coefficient of the target supplier... Compare with the rejection threshold T, including:

[0176] when When T < T, it indicates that the risk of the target supplier's bid being rejected is low;

[0177] When T≤ When the value is less than 1.5T, it indicates that the target supplier faces a Level 1 risk of bid rejection. The target supplier is required to submit an explanation of the reasonableness of the reasons for bid rejection within the specified period, focusing on clarifying the specific causes of the reasons for bid rejection.

[0178] when When the value is ≥1.5T, it indicates that the target supplier has a level 2 risk of bid rejection. Level 2 is greater than level 1, and the risk of bid rejection is high. It is necessary to combine the supplier's historical bid rejection records, qualification documents, and compliance status to separately verify the root cause of the abnormal bid rejection association, focus on analyzing the reasons for strong bid rejection association, and investigate illegal bidding, false bidding, and malicious bid rejection.

[0179] The threshold T for associated rejection bids is taken from the statistical value of the normal range of text similarity for rejection bids in the industry.

[0180] In this embodiment, the deviation of a supplier's single bid and the long-term average bid are calculated by coordinating the actual bid and the benchmark bid. This constructs a quantitative evaluation mechanism for the rationality of bids, which effectively improves the system's accuracy in identifying abnormal bid behavior of suppliers. By comparing the bid data of the same supplier in a single bid and in historical bids, the bid deviation range and average deviation level are calculated to form a standardized bid deviation coefficient. This helps to unify judgment and subsequent hierarchical management. The system can automatically identify abnormal bid situations based on the deviation threshold Q. It can still maintain high-precision bid risk identification in multi-project and multi-category bidding scenarios, and improve the standardization and reliability of bidding data processing.

[0181] By collaboratively analyzing the bidding layout characteristics of suppliers through regional and category-based bidding distribution, a bidding concentration risk assessment mechanism is constructed. This effectively improves the system's accurate perception of abnormal bidding layouts by suppliers. By comparing the number of bidding projects of the same supplier in different regions and categories, the system calculates the regional distribution uniformity and category distribution uniformity to form a standardized bidding concentration index. This helps to unify judgments and subsequent risk warnings. The system can automatically identify abnormal bidding layouts based on the concentration threshold R. It maintains high-precision identification of bid rigging and collusion risks even in multi-regional and multi-category bidding scenarios, thereby improving the stability and rigor of bidding risk control management.

[0182] By conducting collaborative semantic analysis of current, historical, and industry-common reasons for bid rejection, a quantitative assessment mechanism for bid rejection risk is constructed. This effectively improves the system's accuracy in identifying abnormal bid rejection behavior by suppliers. By comparing the semantic text of the same supplier in current, historical, and industry-standard bid rejections, the system calculates its own historical similarity and industry standard similarity, forming a standardized bid rejection correlation coefficient. This helps to unify judgments and subsequent risk verification. The system can automatically identify abnormal bid rejection risk based on the bid rejection correlation threshold T. It maintains high-precision malicious bid rejection identification even in multi-project, multi-frequency bid rejection scenarios, improving the stability and rigor of bid rejection risk control in bidding.

[0183] Example 5

[0184] This embodiment is an explanation of embodiment 1. Specifically, S4 includes:

[0185] S41. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids After normalization, the matching index of the target supplier is calculated using the following formula. :

[0186] ;

[0187] The following is the matching index of the target supplier. Example table, see Table 4 below;

[0188] Supplier ID Price deviation coefficient Bid Concentration Index Correlation coefficient of rejected bids Matching Index GYS001 0.026 0.112 0.104 0.079 GYS002 0.218 0.338 0.322 0.289 GYS003 0.256 0.468 0.456 0.382 GYS004 0.575 0.512 0.604 0.562 GYS005 0.389 0.208 0.722 0.411

[0189] Table 4

[0190] S41. By setting a preset matching threshold X, and setting the matching index of the target supplier... Compare with the matching threshold X, including:

[0191] When the matching index of the target supplier > When the matching threshold X is reached, it indicates that the target supplier's bidding data is normal, the overall performance of price reasonableness, bid dispersion, and bid rejection compliance meets the standards, and the target supplier's integrity level and procurement suitability meet the requirements.

[0192] when <X indicates that the target supplier's bidding data is abnormal. The first strategy is generated, which includes directly suspending the bidder's qualification for this bidding, adding the supplier to the key risk monitoring list, focusing on investigating suspected illegal bidding, and separately verifying the root cause of the abnormal bidding relationship based on the target supplier's historical bid rejection records, qualification documents, and compliance status. The analysis focuses on the causes of strong bid rejection correlations, and the investigation of illegal bidding, false bidding, and malicious bid rejection violations are recorded in the supplier's credit file and their subsequent bidding rights are restricted.

[0193] The matching threshold X is determined by first screening and aggregating a group of high-quality suppliers across the industry who are compliant with regulations, have good performance records, and have standardized bidding practices. This involves comprehensively collecting matching index data from these high-quality suppliers, calculated across dimensions such as price reasonableness, bid dispersion, and compliance with bid rejection regulations. Then, the matching index data of all these high-quality suppliers is uniformly summarized, organized, and statistically analyzed. The arithmetic mean of this batch of data is calculated to obtain the statistical mean of the matching degree of high-quality suppliers in the industry. Finally, this statistical mean is used as the specific value of the matching threshold X.

[0194] In this embodiment, a comprehensive matching and evaluation mechanism for bidding data is constructed by synergistically integrating the supplier's bidding performance with the price deviation coefficient, bid concentration index, and bid rejection correlation coefficient. This effectively improves the system's accurate judgment of the supplier's bid compliance and procurement suitability. By comparing the quantitative indicators of the same supplier in three dimensions—price, bidding layout, and bid rejection behavior—a weighted fusion value of multiple indicators is calculated to form a standardized supplier matching index. This helps to unify judgment and subsequent comprehensive management. The system can automatically identify comprehensive anomalies in bidding data based on the matching threshold X, maintaining high-precision comprehensive risk identification even in multi-supplier, multi-project bidding scenarios, thereby improving the standardization and reliability of bidding data entry review.

[0195] Example 6

[0196] This embodiment is an explanation of embodiment 1. Specifically, S4 further includes:

[0197] S46. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids Dimensionless processing is used to calculate the target supplier's credit profile fluctuation prediction coefficient W using the following related formula:

[0198] ;

[0199] The following is a sample table of the fluctuation prediction coefficient W for the credit profile of the target supplier, as shown in Table 5 below;

[0200] Supplier ID Price deviation coefficient Bid Concentration Index Correlation coefficient of rejected bids Credit profile volatility prediction coefficient W GYS001 0.026 0.112 0.104 0.079 GYS002 0.218 0.338 0.322 0.289 GYS003 0.256 0.468 0.456 0.382 GYS004 0.575 0.512 0.604 0.562 GYS005 0.389 0.208 0.722 0.411

[0201] Table 5

[0202] S4 also includes:

[0203] S47. A preset fluctuation threshold Y is used to compare the predicted fluctuation coefficient W of the target supplier's credit profile with the fluctuation threshold Y to obtain the prediction results, including:

[0204] If W≤Y, it means that the target supplier's credit profile is stable, its bidding behavior is regular, and its performance risk is controllable. The current target supplier rating and cooperation strategy will be maintained.

[0205] If W > Y, it indicates that the target supplier's credit profile is fluctuating drastically, its bidding behavior is abnormal, and its potential risk of default is high, triggering a credit fluctuation warning and generating a second strategy, including implementing key monitoring, review of bidding qualifications, performance risk warnings, or phased access restriction control measures for the target supplier.

[0206] The fluctuation threshold Y is determined by collecting credit profile data from suppliers within the same industry that are operating normally, have good credit records, and have no major breaches of trust or illegal bidding behavior. The stability index of each supplier's credit profile is calculated. After collecting and organizing this data and removing extreme outliers, statistical analysis is performed to calculate the overall statistical mean of the stability of the supplier credit profile within the same industry. This value is then verified and optimized by combining the bidding and procurement party's own credit risk management standards with industry bidding and tendering performance supervision requirements. Finally, this calibrated statistical mean of the stability of the supplier credit profile within the same industry is used as the specific value of the fluctuation threshold Y.

[0207] In this embodiment, a supplier credit profile fluctuation prediction mechanism is constructed by coordinating dimensionless processing of the price deviation coefficient, bid concentration index, and bid rejection correlation coefficient. This effectively improves the system's accurate prediction of the long-term credit stability of suppliers. By comparing quantitative indicators of the same supplier's bidding behavior in multiple dimensions, the credit fluctuation amplitude and abnormal change degree are calculated to form a standardized credit profile fluctuation prediction coefficient. This helps to unify judgment and subsequent credit management. The system can automatically identify abnormal credit change trends based on fluctuation characteristics and maintain high-precision credit risk prediction even in long-term, multi-round bidding scenarios, thereby improving the scientificity and effectiveness of the supplier's full-cycle management.

[0208] By using the credit profile fluctuation prediction coefficient and fluctuation threshold Y to collaboratively determine the credit stability of suppliers, a bidding credit fluctuation early warning and dynamic control mechanism is constructed. This effectively improves the system's accurate prevention and control of potential supplier default risks. By comparing the credit fluctuation prediction coefficient of the same supplier with the benchmark threshold, the system distinguishes between credit stability and abnormal fluctuation levels, forming a standardized basis for credit risk handling. This helps to unify judgment and subsequent hierarchical control. The system can automatically trigger credit early warning and control measures based on the fluctuation threshold Y, maintaining high-precision risk handling even in high-risk, multi-abnormal supplier scenarios, thus improving the stability and rigor of bidding credit management.

Claims

1. A method for standardizing heterogeneous information processing of massive bidding data, characterized in that: Includes the following steps: S1. Collect massive amounts of heterogeneous bidding data from multiple sources through internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and the public resource trading center's bid result announcement platform, and construct a multi-dimensional basic dataset; S2. Extract the actual quotation amount from the target supplier based on a multi-dimensional basic dataset. Benchmark quotation amount The target supplier's bid area total M, the total number of valid bids N, the total number of products involved in the target supplier's bids K, and the reason for the target supplier's current bid rejection text. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) And perform preprocessing; S3. Using a convolutional neural network, construct an AI neural network model. Input the multi-dimensional basic dataset into the AI ​​neural network model to output the analysis results of anomaly prediction in bidding data. Based on the multi-dimensional basic dataset, calculate the target supplier's price deviation coefficient. Bidding Concentration Index Correlation coefficient of rejected bids ; S4. Price deviation coefficient based on target supplier Bidding Concentration Index Correlation coefficient of rejected bids Correlate and construct a matching index for target suppliers. And the credit profile volatility prediction coefficient W, and evaluate and optimize it.

2. The method for standardizing heterogeneous information processing of massive bidding data according to claim 1, characterized in that, S1 includes: S11. Collect data from internal enterprise bidding documents, public resource trading platforms, electronic bidding systems, and public resource trading center's bid result announcement platforms to obtain bidding party data, bidder data, bidding project data, bid evaluation data, and winning bid data, and establish a multi-dimensional basic dataset.

3. The method for standardizing heterogeneous information processing of massive bidding data according to claim 2, characterized in that, S2 includes: S21: The tendering party's data includes the benchmark bid amount. The data was collected from internal bidding documents, which are formal bidding documents prepared by the company based on project requirements, industry standards, and cost estimates. These documents are available in both electronic and paper formats, covering all valid bidding projects. During the data collection process, electronic bidding documents were retrieved from the company's internal bidding management platform, and offline paper bidding documents were scanned and digitized. The benchmark price was located by searching for keywords such as "benchmark price," "budget amount," "control price," and "reference price," and the benchmark price amount was obtained. ; Bidder data includes the total number of target supplier bidding regions (M), the total number of valid bids (N), the total number of products involved in the target supplier's bids (K), and the text of the reasons for the target supplier's previous bid rejections. Furthermore, data is collected through a public resource trading platform. During collection, the platform uses the target supplier's name and unique unified social credit code to filter all project records in which the target supplier participated in bidding. The bidding region information for each project is extracted to obtain the total number of bidding regions (M) for the target supplier. All bidding projects in which the target supplier participated are queried, and the number of projects is counted to obtain the total number of valid bidding projects (N). Product information for all valid bidding projects in which the target supplier participated is extracted to obtain the total number of products involved in the target supplier's bids (K). The historical bidding records of the target supplier are queried using its unique identifier, and projects marked as "rejected" are filtered out. The reason text for each rejected project is extracted to form a complete historical reason text for rejected bids of the target supplier. ; The bidding project data includes the actual bid amount from the target supplier. The data was collected by analyzing internal bidding documents. During the data collection process, the electronic version of the corresponding bidding project was accessed. Offline paper bidding documents were converted into processable electronic files using high-definition scanning and OCR recognition technology. The target supplier's bid price was located by searching keywords such as "actual price," "bid price," and "supplier price," thus obtaining the target supplier's actual bid amount. ; The evaluation data includes the text of the reasons for the current rejection of the target supplier's bid. The data is collected by connecting to an electronic bidding platform. During the data collection process, the data interface of the electronic bidding platform is used to crawl and obtain the complete bidding records of the target supplier's current bidding projects, locate the relevant modules for bid rejection and detailed descriptions, extract the text of the reasons for bid rejection, and obtain the text of the reasons for the target supplier's current bid rejection. ; The data information for successful bids includes a text library of common reasons for bid rejection in the industry. Data is collected through the public resource trading center's bidding result announcement platform. During the data collection process, the platform's backend data interface is connected to collect the text of reasons for bid rejection for publicly disclosed rejected projects within the platform in batches. This text is then organized and analyzed to obtain a database of common reasons for bid rejection in the industry. ; S22. Based on the actual quotation amount from the target supplier and benchmark quotation amount All collected data was uniformly converted to "yuan" as the sole unit of measurement. If the original data unit was "ten thousand yuan" or "thousand yuan," it was multiplied by 10,000 and 1,000 respectively for conversion. The conversion process was recorded, and the unit of measurement was marked after conversion. All non-numerical special symbols in the collected data were completely removed, including currency symbols, thousandths separators, spaces, and text notes, retaining only pure numerical values ​​and decimal points. After removal, the collected data underwent preliminary verification, and the monetary values ​​after removing special symbols were uniformly retained to two decimal places. Missing quotation numbers found during the collection process were also addressed. According to the data, the average price of bidding projects of the same category and in the same period is used for filling. The same category is based on the product and service categories of the bidding project, and the same period is based on the release time of the bidding project. Before filling, it is necessary to confirm that the sample size of the same category and the same period of bidding projects is ≥3. If the sample size is <3, the same period range is expanded to 3 calendar months. If the same period range is still <3 after expanding the same period range, the average price of the same type of project of the target supplier in the next 3 months is used for filling. After filling, the data is marked "Data Supplement" and the supplementary basis. The monetary data is linked with the corresponding project number, target supplier ID, collection source, and preprocessing time for archiving. Based on the total number of target supplier bidding regions M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bidding K; For the total number of valid bidding projects N collected, invalid bidding project records are removed, including projects that did not submit complete bidding documents, failed the bid document review, were voluntarily withdrawn, had irregularities in the bidding documents, were submitted after the deadline, or did not meet the qualification requirements. Valid bidding projects that have passed the preliminary review and meet the bidding requirements are retained for counting. For the total number of bidding areas M collected, invalid area records are removed, including bidding areas that have not been approved, false area entries, areas corresponding to withdrawn bids, and abnormal areas in the bidding scope. The bidding areas that the target supplier actually participated in and approved are counted again. For the total number K of products involved in the bidding collected, product categories with abnormal bidding requirements are removed, including product categories that are outside the scope of bidding products, have mismatched specifications and models, or do not meet the qualification requirements. Product categories that meet the bidding requirements and have passed the review are retained for counting. After counting is completed, the total number of target supplier bidding areas M, the total number of valid bidding projects N, and the total number of products involved in the target supplier's bids K are verified to confirm that the data logic after removing invalid and duplicate records is reasonable. After verification, the valid counting basis is marked, including the counting range, details of invalid and duplicate records removed, verification personnel and verification time. For missing counting data found during the collection process, the average data of the corresponding similar bidding projects of the target supplier within the past 3 months is used to supplement it. When supplementing, it is necessary to confirm that there are similar bidding projects in the past 3 months. If there are no similar bidding projects in the past 3 months, the average data of similar projects of suppliers of the same industry and scale during the same period is used to supplement it. The supplemented data and the basis for supplementation are marked. Based on the text of the current reasons for the target supplier's rejection of the bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The text containing the reasons for the current rejection of bids by the target suppliers was collected. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) A comprehensive process is performed to remove invalid characters from the text, including garbled text, special symbols, irrelevant notes, whitespace characters, and line breaks. Valid text content related to the reason for rejection is retained. Typos and grammatical errors in the text are corrected. The preprocessed text data is then uniformly converted to UTF-8 encoding format.

4. The method for standardizing heterogeneous information processing of massive bidding data according to claim 1, characterized in that, S3 includes: S31 is used to construct an AI neural network model using a convolutional neural network, train and test the AI ​​neural network model with a multi-dimensional basic dataset, and use the trained AI neural network model as a bidding processing model. At the same time, the intermediate layer output of the bidding processing model is used as a feature vector to identify the information features in the multi-dimensional basic dataset, and the trained bidding processing model is used as a data prediction tool.

5. The method for standardizing heterogeneous information processing of massive bidding data according to claim 1, characterized in that, S3 further includes: S32. Extract the actual quoted amount from the target supplier. Benchmark quotation amount Combined and normalized, the deviation of a single quote from the target supplier is calculated using the following formula. : ； S33. Extract the total number of valid bids N, and output the average deviation of the target supplier's bid. It can be calculated using the following formula: ； In the formula, N represents the total number of valid bids submitted by the target supplier within the statistical period. This represents the bid amount from the target supplier in the i-th bidding project; S34. Deviation of single quotation from the target supplier Average deviation from the target supplier's quote After normalization, the deviation coefficient of the target supplier's quote is calculated using the following formula. : ； In the formula, Kb represents the deviation of the quoted price from the tolerance threshold, which is an empirical constant. This represents the maximum deviation amplification factor, which is an empirical constant. By setting a preset deviation threshold Q, and calculating the deviation coefficient of the target supplier's quotation... Compare with the deviation threshold Q, including: when When the value is less than Q, it indicates that the target supplier's quoted price is reasonable. When Q≤ A price less than 2Q indicates a Level 1 deviation from the target supplier's quote, suggesting price fluctuations. This necessitates monitoring the target supplier, highlighting the actual quote data, correlating it with historical average quotes and the reasons for deviations, and archiving it in the supplier quote anomaly file, based on industry benchmark quote amounts. Provide suppliers with targeted suggestions for optimizing their bids, and explicitly require that the deviation between their subsequent bids and historical averages be reduced by more than 30%, thereby guiding suppliers to optimize their bidding strategies and standardize cost accounting. when When the deviation is ≥2Q, it indicates that the target supplier's quotation has a second-level deviation. The second level is greater than the first level, and the quotation is abnormal. The quotation data should be marked as abnormal, the validity of the quotation should be suspended, and the quotation data and deviation details should be synchronized to the procurement management department and the bid evaluation committee to trigger a quotation abnormality risk warning. The supplier should be required to submit a special rectification plan and quotation optimization plan within a time frame. After the rectification is completed, the quotation data should be recalculated. The quotation validity should be restored after the quotation deviation drops below Q.

6. The method for standardizing heterogeneous information processing of massive bidding data according to claim 5, characterized in that, S3 further includes: S35. By extracting the total number of bidding regions M of the target suppliers, the regional bidding concentration is calculated using the following formula. : ； In the formula, This represents the proportion of the target supplier's bids in the j-th region to the total number of bids, where j represents the region number. S36. By extracting the total number K of products involved in the target supplier's bids, the category bidding concentration is calculated using the following formula. : ； In the formula, This represents the proportion of the target supplier's bids in the k-th category out of the total number of bids, where k represents the category number; S37. Concentration of regional bids from target suppliers Category concentration of bids with target suppliers The bid concentration index of the target supplier is obtained by correlation and normalization using the following formula. ; ； In the formula, and Represents the weighting coefficients, satisfying + =1; S39. By setting a concentration threshold R, the bid concentration coefficient of the target supplier is... Compare with the concentration threshold R, including: when When <R, it indicates that the target supplier matches the business layout; If R≤ When the deviation is less than 1.5R, it indicates that the target supplier's own business has a first-level deviation, and the target supplier needs to be controlled, verified, and the risks mitigated. when When the deviation is ≥1.5R, it indicates that the target supplier's own business has a second-level deviation. If the second level is greater than the first level, an optimization strategy will be implemented, including simultaneously sending early warning instructions to the procurement management department and the bid evaluation committee, suspending the bidding qualification for this tender, and including it in the key risk monitoring list for key investigation of bid rigging and collusion.

7. The method for standardizing heterogeneous information processing of massive bidding data according to claim 6, characterized in that, S3 further includes: S310. Extract the text explaining the current reasons for bid rejection from the target supplier. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The BERT model is used to extract the text of the reasons for the current rejection of the target supplier's bid. Text explaining the reasons for the target supplier's past bid rejections Common reasons for bid rejection in the industry (text library) The numerical feature vectors obtained by converting the text into the reasons for the current rejection by the target supplier are then processed into vectors. The numerical feature vector is obtained by vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry : ； ； ； In the formula, (.) represents the text vectorization function; S311. Numerical feature vector obtained by vectorizing the text of the reason for the current rejection of the target supplier's bid. The numerical feature vector obtained after vectorizing the text of the reasons for the target supplier's historical bid rejections. Benchmark vector of common reasons for bid rejection across the industry By combining these factors, the similarity between the target supplier's current bid rejection behavior and its own historical bid rejection behavior can be calculated. Similarity to current rejected bids and common industry-wide rejected bid standards : ； ； In the formula, Represented as a cosine similarity function, it measures the cosine of the angle between two vectors, reflecting the relevance of their content. S312. Compare the current rejection of bids by the target supplier with the similarity of the target supplier's past rejection behavior. Similarity to current rejected bids and common industry-wide rejected bid standards After combining and normalizing the data, the correlation coefficient of the rejected standards is calculated. : ； In the formula, and Represents the weighting coefficients, satisfying =1; By setting a preset rejection correlation threshold T, and setting the rejection correlation coefficient of the target supplier... The comparison is performed with the rejection threshold T, including: when When T < T, it indicates that the risk of the target supplier's bid being rejected is low; When T≤ When the value is less than 1.5T, it indicates that the target supplier faces a Level 1 risk of bid rejection. The target supplier is required to submit an explanation of the reasonableness of the reasons for bid rejection within the specified period, focusing on clarifying the specific causes of the reasons for bid rejection. when When the value is ≥1.5T, it indicates that the target supplier has a level 2 risk of bid rejection. Level 2 is greater than level 1, and the risk of bid rejection is high. It is necessary to check the root cause of the abnormal bid rejection in combination with the supplier's historical bid rejection records, qualification documents, and compliance status. The focus should be on analyzing the reasons for strong bid rejection correlations and investigating illegal bidding, false bidding, and malicious bid rejection.

8. The method for standardizing heterogeneous information processing of massive bidding data according to claim 1, characterized in that, S4 includes: S41. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids After normalization, the matching index of the target supplier is calculated using the following formula. : ； S41. By setting a preset matching threshold X, and setting the matching index of the target supplier... Compare with the matching threshold X, including: When the matching index of the target supplier > When the matching threshold X is reached, it indicates that the target supplier's bidding data is normal, the overall performance of price reasonableness, bid dispersion, and bid rejection compliance meets the standards, and the target supplier's integrity level and procurement suitability meet the requirements. when <X indicates that the target supplier's bidding data is abnormal, generating the first strategy, which includes directly suspending the bidder's eligibility for this bidding, adding the supplier to the key risk monitoring list, focusing on investigating suspected illegal bidding, and separately verifying the root cause of abnormal bidding associations based on the target supplier's historical bid rejection records, qualification documents, and compliance status, focusing on analyzing the reasons for strong bid rejection associations, investigating illegal bidding, false bidding, and malicious bid rejection violations, recording them in the supplier's credit file, and simultaneously restricting their subsequent bidding rights.

9. The method for standardizing heterogeneous information processing of massive bidding data according to claim 8, characterized in that, S4 further includes: S46. Calculate the deviation coefficient of the target supplier's quotation. Bidding Concentration Index Correlation coefficient of rejected bids After normalization, the credit profile fluctuation prediction coefficient W of the target supplier is calculated using the following related formula: 。 10. The method for standardizing heterogeneous information processing of massive bidding data according to claim 9, characterized in that, S4 further includes: S47. A preset fluctuation threshold Y is used to compare the predicted fluctuation coefficient W of the target supplier's credit profile with the fluctuation threshold Y to obtain the prediction results, including: If W≤Y, it means that the target supplier's credit profile is stable, its bidding behavior is regular, and its performance risk is controllable. The current target supplier rating and cooperation strategy will be maintained. If W > Y, it indicates that the target supplier's credit profile is fluctuating drastically, its bidding behavior is abnormal, and its potential risk of default is high. This triggers a credit fluctuation warning instruction and generates a second strategy, which includes implementing key monitoring, review of bidding qualifications, performance risk warnings, or phased access restriction control measures on the target supplier.

Images