Method and system for identifying risk of bid enclosing and bid stringing, electronic equipment and medium

Abstract

The invention provides a bid invitation risk identification method and system, electronic equipment and a storage medium, and aims to solve the problems of single bid invitation form and low judgment accuracy in consideration of bid invitation risk, and the method comprises the steps: building a bidding relation matrix for bid invitation projects according to bid invitation activities, bid winners and participant information; historical bidding and tendering data is cleaned and combed, and project risk evaluation information is obtained based on the bidding relation matrix and comprises a bid rate, a bid winning rate, a quotation gap and an enterprise information consistency result; performing unsupervised learning on the project data, calculating a project risk value according to the risk evaluation information, and judging whether the project needs text analysis according to the project risk value; similarity analysis is carried out on the technical part and the price part of the bidding document for the project needing to be analyzed; and for the high-risk-value project, combining the technical part similarity and the price part similarity of the bidding file to judge whether the bidding and confusion bidding exist or not. According to the invention, the accuracy and comprehensiveness of bid enclosing and bid stringing analysis can be improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of information data processing, and in particular to a method for identifying risks of bid rigging and collusion, a system for identifying risks of bid rigging and collusion, an electronic device and a computer-readable storage medium. Background Art

[0002] The phenomenon of bid rigging and bid collusion is common in bidding and various procurement bidding practices and is becoming more and more serious. Since the evidence required to determine the facts of bid rigging and bid collusion is difficult to collect, and the means used by bid rigging and bid collusion are relatively covert and low-risk, the risk-benefit-risk-cost ratio is magnified, and bid rigging and bid collusion are rampant.

[0003] "Bid rigging" refers to the means, behavior and phenomenon of several bidders (suppliers) agreeing with each other to raise or lower their bid prices in order to restrict competition, exclude other bidders and make a certain stakeholder win the bid, thereby seeking benefits. The initiator of bid rigging is called the bid rigging person, and the bidders who participate in bid rigging are called the bidders who accompany the bid. In a broad sense, "bid collusion" means bidding collusion, that is, bidders collude with each other to win the bid and damage the interests of the project owner, or bidders collude with the bid issuer or agency to damage the interests of other bidders or project owners in an attempt to illegally win the bid. There is no clear distinction between bid rigging and bid collusion. Bid rigging uses bid collusion as a means, while bid collusion uses bid rigging as an end. Although the existing technology can identify bid rigging and bid collusion by analyzing the bid documents, considering that bid rigging and bid collusion take only a single form, the existing technology only analyzes one scenario of bid rigging and bid collusion, such as consistent corporate information, regular quotations, and similarity in bid document contents, without comprehensive consideration. Moreover, the judgment accuracy is low, the means of bid rigging and bid collusion are becoming increasingly diverse and concealed, and traditional judgment methods cannot be used for efficient verification. The risk projects have a low correlation with whether there is actually a problem of bid rigging and bid collusion. Summary of the invention

[0004] In order to at least solve the problem in the prior art that bid colluding and bid collusion have only a single form and the accuracy of bid colluding and bid collusion determination is low, the present disclosure provides a bid colluding and bid collusion risk identification method, a bid colluding and bid collusion risk identification system, an electronic device and a computer-readable storage medium, which can effectively prevent bid colluding and bid collusion from circumventing review by collusion among multiple parties with weak or no connection, further improve the accuracy of bid colluding and bid collusion analysis, improve the efficiency of reviewing bid colluding and bid collusion in bid documents, and improve the quality and comprehensiveness of analysis.

[0005] In a first aspect, the present disclosure provides a method for identifying bid rigging and collusion risks, the method comprising:

[0006] Establish a bidding relationship matrix for bidding projects based on bidding activities, winning bidders, and participating party information;

[0007] Clean and sort out historical bidding data, and obtain project risk assessment information based on the bidding relationship matrix, including bid participation rate, bid winning rate, quotation gap and enterprise information consistency results;

[0008] Perform unsupervised learning on project data, calculate the risk value of project bid rigging based on project risk assessment information, and use the risk value to determine whether the project needs further text analysis;

[0009] For projects that require text analysis, we analyze the similarity of the technical part of the bidding documents and the similarity of the price part of the bidding documents;

[0010] For high-risk projects obtained based on the project collusion risk value, the similarity of the technical part and the price part of the bidding documents are combined to determine whether there is collusion in bidding, and an analysis report is output.

[0011] Furthermore, the bidding relationship matrix is ​​established for the bidding project based on the bidding activities, the winning bidder, and the information of the participating parties, including:

[0012] Based on the existing bidding data, one-to-one matching is performed between projects and winning bidders, and a two-dimensional matrix with bidding projects and winning bidders as axes is established;

[0013] The data of multiple participants in the same bidding project in the bidding data are combined with a two-dimensional matrix to be converted into a multi-dimensional matrix, including: taking at least one participant as the main body, merging the names of multiple participants containing the main body into a unique identifier, and taking the identifiers of the multiple participant information as the axis, upgrading the two-dimensional matrix to a multi-dimensional matrix to achieve one-to-many matching of bidding activities, successful bidders, and multiple participants.

[0014] Furthermore, the cleaning and sorting of historical bidding data includes:

[0015] Initial deletion includes: deleting data with the same project identifier "result" and the number of participating suppliers "count" ≥ the preset threshold, deleting data with incomplete project identifiers and corresponding supplier information, and deleting data with empty or all zero bid amounts;

[0016] Screening and merging, including: extracting projects with similar last digits of project identifier "result", identical bidding suppliers and bidding amounts, treating them as multiple bidding records, and retaining only the project corresponding to the latest "result";

[0017] Data restoration includes: extracting all items with bidding amounts between 0 and 100, and restoring all bidding prices according to the winning bid price and bidding document price score conversion rules.

[0018] Furthermore, the project risk assessment information is obtained based on the bidding relationship matrix, including:

[0019] According to the information of the winning bidder and the information of multiple participants in the multidimensional matrix, the number of times each supplier participated in the bidding is obtained, and the participation rate is determined based on a certain period of time, where the participation rate = the number of current participations / the total number of participations;

[0020] According to the multidimensional matrix, taking the bidding project as the benchmark, the number of winning bids of each winning bidder is traversed to determine the winning rate of each winning bidder, wherein the winning rate = the number of winning bids in the current period / the number of participations in the current period;

[0021] Based on the bidding projects, extract the winning bid amount of each project and the bid amount of each bidder in turn, calculate the price difference between each bidder amount and the winning bid amount, and get the quotation gap;

[0022] In the enterprise information database, extract all supplier enterprise information, conduct consistency comparison on all participating enterprise information of a single project, and obtain the enterprise information consistency result.

[0023] Furthermore, the unsupervised learning of the project data and the calculation of the risk value of the project bid rigging based on the project risk assessment information include:

[0024] Obtain risk dimension indicators of projects and suppliers based on project risk assessment information of projects and suppliers, wherein the risk dimension indicators include price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate;

[0025] Obtain data containing existing bid-rigging cases as a training set, use risk dimension indicators as input, use the Xgboost (eXtreme Gradient Boosting) algorithm to train the training set, and perform unsupervised learning on the data to obtain a risk prediction model;

[0026] The risk value of bid rigging for all projects is calculated through the risk prediction model.

[0027] Furthermore, the analysis of the similarity of the technical parts of the bidding documents includes:

[0028] By identifying the relevant information of the keywords "contact person" and "telephone", the contact information of the bidding document is extracted, wherein the contact information includes the bidding contact person and contact information;

[0029] The extracted contact information is compared one-to-one between suppliers. If one or more pieces of information are consistent, the text similarity is determined to be similar;

[0030] If it does not exist, the technical solution content in the bidding document is extracted by identifying the relevant information of the keywords "solution" and "technical solution";

[0031] Use the BBPE (Byte-level Byte Pair Encoding) algorithm at the subword granularity to segment the technical solution content;

[0032] The dual-tower SBERT (Siamese network) model is used to calculate the final risk value of the similarity of the technical part after word segmentation.

[0033] Furthermore, the analysis of the similarity of the price part of the bidding documents includes:

[0034] Calculate the price difference of each bidder's itemized quotations, the price difference of the total price quotations, and whether there is any regularity in the quotations;

[0035] Calculate the price difference between the itemized bids of each bidder that participated but failed to win the bid and the winning bidder to see whether there is a similar price difference or a regular pattern in the overall bids;

[0036] If it exists, the final risk value of the price part similarity is determined as the highest value;

[0037] If it does not exist, the total price difference percentage and the itemized price difference percentage between all bidders shall be calculated based on the itemized price difference and the total price difference;

[0038] Output the final risk value of the price similarity based on the weighted average of the three dimensions of itemized quotation repetition rate, itemized quotation risk value, and total price quotation risk value;

[0039] Among them, the risk value of each bidder's sub-item quotation = 1-min{corresponding sub-item quotation price difference percentage}, the total price quotation risk = 1-min{corresponding total price quotation price difference percentage}, and the sub-item quotation repetition rate = the number of sub-items with consistent sub-item quotation price differences / the number of sub-items in the sub-item quotation table.

[0040] Furthermore, judging whether there is bid rigging or bid collusion for the high risk projects obtained according to the risk value of bid rigging or bid collusion in the project by combining the similarity of the technical part and the similarity of the price part of the bidding documents, includes:

[0041] If the text similarity is determined to be similar, it is directly determined that there is a string of bids;

[0042] If they are not determined to be similar, the risk value of each supplier will be output according to the project's bid-rigging risk value, the final risk value of the technical similarity, and the final risk value of the price similarity using the random forest method;

[0043] For each bidding project, the risk values ​​of all participating bidders are calculated and the risk level of the bidding project is output. The risk level is divided into: high risk, medium risk and low risk.

[0044] In a second aspect, the present disclosure provides a bid-rigging and bid-collusion risk identification system, the system comprising:

[0045] A matrix establishment module is configured to establish a bidding relationship matrix for a bidding project based on information on bidding activities, successful bidders, and participating parties;

[0046] A processing module is configured to clean and sort out historical bidding data, and obtain project risk assessment information based on a bidding relationship matrix, wherein the project risk assessment information includes a bid participation rate, a bid winning rate, a bid gap, and a result of consistency of enterprise information;

[0047] The project risk assessment module is configured to perform unsupervised learning on project data, calculate the risk value of bid rigging based on the project risk assessment information, and determine whether the project needs further text analysis based on the risk value of bid rigging;

[0048] A text analysis module, which is configured to analyze the similarity of the technical part of the bidding documents and the similarity of the price part of the bidding documents for the projects requiring text analysis;

[0049] The bid rigging and collusion identification module is configured to determine whether there is bid rigging and collusion for high-risk projects obtained based on the project bid rigging and collusion risk values, combined with the similarity of the technical part and the similarity of the price part of the bidding documents, and output an analysis report.

[0050] In a third aspect, the present disclosure provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for identifying bid rigging and collusion risks as described in any one of the first aspects.

[0051] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for identifying the risk of bid rigging and collusion described in any one of the above-mentioned first aspects is implemented.

[0052] Beneficial effects:

[0053] The method for identifying the risk of bid rigging and collusion, the system for identifying the risk of bid rigging and collusion, the electronic device and the storage medium provided by the present disclosure construct a bidding combination relationship matrix, clearly sort out the relationship between the bidding activities, the successful bidders and the participating parties, conduct behavioral analysis on the bidding activities, and analyze the four aspects of information including participation rate, successful bid rate, quotation gap and consistency of enterprise information. The bidding combination is determined by analyzing the behavior of participating in the bidding activities, and then the weight selection of various types of information is simulated and trained through an unsupervised learning algorithm to calculate the risk value of bid rigging and collusion for the project and then judge the possibility of bid rigging and collusion. For projects with a high possibility of bid rigging and collusion, the final target projects with high risk of bid rigging and collusion are determined by further conducting text analysis on the bidding documents of projects with the possibility of bid rigging and collusion. It can effectively avoid bid rigging and collusion from evading review by collusion between multiple parties with weak or no association, further improve the accuracy of bid rigging and collusion analysis, the efficiency of reviewing bid rigging and collusion in bid documents, and improve the quality and comprehensiveness of analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of a method for identifying bid rigging and collusion risks provided in the first embodiment of the present disclosure;

[0055] Figure 2 A flowchart of a method for identifying bid rigging and collusion risks provided in the second embodiment of the present disclosure;

[0056] Figure 3 This is an architecture diagram of a bid rigging and collusion risk identification system provided in Embodiment 3 of the present disclosure;

[0057] Figure 4 This is an architecture diagram of an electronic device provided in Embodiment 4 of the present disclosure. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, the present disclosure is further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention, rather than to limit the present invention.

[0059] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence; and, in the absence of conflict, the embodiments in the present disclosure and the features in the embodiments can be arbitrarily combined with each other.

[0060] The terms used in the embodiments of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. The singular forms of "a", "said" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings.

[0061] In the subsequent description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present disclosure, and have no specific meanings. Therefore, "module", "component" or "unit" can be used in a mixed manner.

[0062] The following is a detailed description of the technical solution of the present invention and how the technical solution of the present invention solves the problems of the single form of bid rigging and bid collusion in the prior art and the low accuracy of bid rigging and bid collusion determination with specific embodiments. It can be understood that in the embodiments of the present application, the execution subject can perform some or all of the steps in the embodiments of the present application, and these steps or operations are only examples. The embodiments of the present application can also perform other operations or variations of various operations. In addition, the various steps can be performed in different orders presented in the embodiments of the present application, and it may not be necessary to perform all the operations in the embodiments of the present application. Moreover, the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0063] Figure 1 A flowchart of a method for identifying bid rigging and collusion risks provided in the first embodiment of the present disclosure is shown in FIG. Figure 1 As shown, the method includes:

[0064] Step S101: Establish a bidding relationship matrix for the bidding project based on the bidding activities, successful bidders, and participating party information;

[0065] Step S102: Clean and sort out historical bidding data, and obtain project risk assessment information based on the bidding relationship matrix, wherein the project risk assessment information includes bid participation rate, bid winning rate, bid price gap, and enterprise information consistency results;

[0066] Step S103: Perform unsupervised learning on the project data, calculate the risk value of bid rigging for the project based on the project risk assessment information, and determine whether the project needs further text analysis based on the risk value of bid rigging for the project;

[0067] Step S104: for the project requiring text analysis, the similarity of the technical part of the bidding document is analyzed, and the similarity of the price part of the bidding document is analyzed;

[0068] Step S105: for the high-risk projects obtained according to the risk value of bid rigging, determine whether there is bid rigging or bid collusion by combining the similarity of the technical part and the similarity of the price part of the bidding documents, and output an analysis report.

[0069] The purpose of the disclosed embodiment is to comprehensively consider multiple scenarios of bidding projects by using multiple technologies and algorithms such as data analysis and text analysis, involving relatively comprehensive dimensions, so as to effectively avoid bid rigging and collusion by using weakly related or unrelated collusion to evade review, and further improve the accuracy of bid rigging and collusion analysis; the disclosed embodiment not only analyzes from the data perspective, but also combines text analysis and uses natural language processing technology to analyze bidding documents, thereby improving the review efficiency, analysis quality and comprehensiveness of bid rigging and collusion. In addition, by matching similarities from the perspective of similar risks of various materials, the bid rigging and collusion risk value of the bid document is calculated by weighting various aspects such as operator data and external network enterprise personnel telephone comparison, quotation comparison, scoring detail ratio, bidding combination, etc., and then the possibility of bid rigging and collusion is judged.

[0070] The disclosure does not restrict the type of bidding and can be applied to various bidding activities, such as procurement, project construction or maintenance. First, a bidding relationship matrix is ​​established for the bidding project based on the information of the bidding activities, the successful bidders and the participating parties; the relationship matrix is ​​used to clearly sort out the relationship between the bidding activities, the successful bidders and the participating parties (non-successful bidders). Since the bidding combination relationship matrix functions provided by various systems are black boxes, it is impossible to filter data as needed; this solution generates combination relationships through manual programming in Python: the procurement data is placed in the database, and the procurement data and winning order data that need to be analyzed are accessed through the first-level cursor loop. The second-level cursor searches for all participating manufacturers in the same winning order, and builds a matrix (program array) of participating manufacturers and purchase orders in the program and database. Through relationship sorting, a one-to-many match between bidding activities, successful bidders and multiple participating parties is established.

[0071] Then, the specific detailed data is cleaned, and based on the bidding relationship matrix, the project risk assessment information is obtained, such as the participation rate and winning rate of each participant, the quotation gap of each participant in the project and the consistency results of enterprise information. In addition, the combination relationship between the participants can be found, such as the number of times multiple participants participate in a project at the same time, the winning relationship, etc.; using the project risk assessment information, all project data are unsupervisedly learned by using algorithms such as XGboost, and the risk value of bid rigging and collusion in each project is calculated, and the need for further text analysis is determined by the relevant threshold; the setting of the relevant threshold can be determined by empirical values ​​or analysis of some known data. For high-risk projects, the similarity of the text is determined by further analyzing the bidding documents, including the similarity of the technical part and the similarity of the price part. Finally, based on the risk value of bid rigging and collusion in the project and the similarity of the text, it is determined whether the project has bid rigging and collusion, and an analysis report is output for the high-risk target projects of bid rigging and collusion that are finally determined.

[0072] The disclosed embodiment constructs a bidding combination relationship matrix to clearly sort out the relationship between the bidding activities, the successful bidders, and the participating parties, and conducts behavioral analysis on the bidding activities. The analysis dimensions include four aspects of information: participation rate, successful bid rate, quotation gap, and enterprise information consistency. The accompanying bidding combination is determined through the analysis of the bidding activities, and then the weight selection of various types of information is simulated and trained through an unsupervised learning algorithm to calculate the risk value of bid rigging and collusion for the project, and then the possibility of bid rigging and collusion is determined. For projects with a high possibility of bid rigging and collusion, the final target projects with high risk of bid rigging and collusion are determined by further text analysis of the bidding documents of projects with the possibility of bid rigging and collusion. It can effectively prevent bid rigging and collusion from evading review by collusion between multiple weakly related or unrelated parties, further improve the accuracy of bid rigging and collusion analysis, and the efficiency of reviewing bid rigging and collusion in bid documents, and improve the quality and comprehensiveness of analysis.

[0073] Furthermore, the bidding relationship matrix is ​​established for the bidding project based on the bidding activities, the winning bidder, and the information of the participating parties, including:

[0074] Based on the existing bidding data, one-to-one matching is performed between projects and winning bidders, and a two-dimensional matrix with bidding projects and winning bidders as axes is established;

[0075] The data of multiple participants in the same bidding project in the bidding data are combined with a two-dimensional matrix to be converted into a multi-dimensional matrix, including: taking at least one participant as the main body, merging the names of multiple participants containing the main body into a unique identifier, and taking the identifiers of the multiple participant information as the axis, upgrading the two-dimensional matrix to a multi-dimensional matrix to achieve one-to-many matching of bidding activities, successful bidders, and multiple participants.

[0076] Based on the existing data, the KMP algorithm (Knuth-Morris-Pratt algorithm) is used to perform one-to-one matching between projects and project bidders, and a two-dimensional matrix is ​​established with the bidding activity as the X-axis and the bidder as the Y-axis. Then, multiple manufacturers participating in the same procurement are converted into a multi-dimensional matrix to achieve one-to-many matching of procurement activities, bidders, and multiple participants. The names of multiple participants are merged with a parameter party as the main body to become a unique identifier, and this information is used as the Z-axis to upgrade the aforementioned two-dimensional matrix to a three-dimensional matrix. According to the number of participants, each participant information is used as an axis to establish an N-dimensional matrix. For example, for project 001, there are 4 participants, A, B, C, and D. With A as the main body, the names of A and other participants are merged, such as AB (A and B have participated once), AC, and AD. With B as the main body, the combination includes BA, BC, BD, and so on. All combination relationships are exhaustively listed. After all projects are processed, all combination relationships that have participated in A can be determined. In general, in order to reduce the amount of data, two participants can be combined to analyze which participants have participated in more projects at the same time and have a higher winning rate, or the other party is just a bidder with a particularly low winning rate or has never won a bid. If necessary, 3 or 4 participants can also be combined for analysis.

[0077] Furthermore, the cleaning and sorting of historical bidding data includes:

[0078] Initial deletion includes: deleting data with the same project identifier "result" and the number of participating suppliers "count" ≥ the preset threshold, deleting data with incomplete project identifiers and corresponding supplier information, and deleting data with empty or all zero bid amounts;

[0079] Screening and merging, including: extracting projects with similar last digits of project identifier "result", identical bidding suppliers and bidding amounts, treating them as multiple bidding records, and retaining only the project corresponding to the latest "result";

[0080] Data restoration includes: extracting all items with bidding amounts between 0 and 100, and restoring all bidding prices according to the winning bid price and bidding document price score conversion rules.

[0081] The preset threshold can be set according to actual conditions, such as 15, 20, etc.; according to actual procurement conditions, the number of suppliers participating in various normal procurement activities is generally not more than 15; by cleaning and combing the data, the quality and accuracy of the data can be ensured, laying a good foundation for subsequent data analysis.

[0082] Furthermore, the project risk assessment information is obtained based on the bidding relationship matrix, including:

[0083] According to the information of the winning bidder and the information of multiple participants in the multidimensional matrix, the number of times each supplier participated in the bidding is obtained, and the participation rate is determined based on a certain period of time, where the participation rate = the number of current participations / the total number of participations;

[0084] According to the multidimensional matrix, taking the bidding project as the benchmark, the number of winning bids of each winning bidder is traversed to determine the winning rate of each winning bidder, wherein the winning rate = the number of winning bids in the current period / the number of participations in the current period;

[0085] Based on the bidding projects, extract the winning bid amount of each project and the bid amount of each bidder in turn, calculate the price difference between each bidder amount and the winning bid amount, and get the quotation gap;

[0086] In the enterprise information database, extract all supplier enterprise information, conduct consistency comparison on all participating enterprise information of a single project, and obtain the enterprise information consistency result.

[0087] The suppliers are participants, including the successful bidder, and the certain period of time can be set independently, such as one quarter, several quarters or one year;

[0088] When obtaining enterprise information, the enterprise information of each participant in each project is compared one by one to obtain the consistency result of enterprise information.

[0089] Preferably, for operators, all supplier enterprise information in the system can be extracted from the operator's enterprise information database, including but not limited to enterprise telephone, email, address, registered legal person information, and enterprise change information; and compared with the enterprise information in the project to determine whether the enterprise information is consistent.

[0090] Furthermore, the unsupervised learning of the project data and the calculation of the risk value of the project bid rigging based on the project risk assessment information include:

[0091] Obtain risk dimension indicators of projects and suppliers based on project risk assessment information of projects and suppliers, wherein the risk dimension indicators include price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate;

[0092] Obtain data containing existing bid-rigging cases as a training set, use risk dimension indicators as input, use the XGboost algorithm to train the training set, and perform unsupervised learning on the data to obtain a risk prediction model;

[0093] The risk value of bid rigging for all projects is calculated through the risk prediction model.

[0094] The calculation of the risk value of bid rigging for a project is done through a trained model. The input value is set as various data of the project and supplier, and the output value is a value between 0 and 1. The specific process is to use the XGboost algorithm for a training set (a training set with known results, where the risk value of problematic cases is 1). XGboost first uses the CART tree to train a model, and generates a deviation value for each sample with known results; then the sample deviation value is used as a new training set, and the CART tree (Classification and Regression Tree) is used to train a new model; this is repeated until a risk value of ≥ 0.9 is output for all projects with known results.

[0095] There are five risk dimensions to consider, including price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate. The price is calculated based on the difference ratio with the bid winning price; enterprise information is compared based on the legal person, contact person, phone number, address, email address, etc. If one of them is consistent, it is 1, otherwise it is 0. Initially, the five risk dimensions are weighted by a coefficient of 0.2 each, and then trained and adjusted with a step size of 0.01;

[0096] Take the existing cases as the result set, and use indicators such as price, enterprise information, bid winning rate, bid participation rate, and co-occurrence rate to conduct unsupervised learning on specific data in some years; verify the obtained result set. If the correct proportion in the result set is high and no marked projects are missed, further update in this direction, and continuously perform business verification and adjust parameter weights through the obtained result set. Use different data verification results to update the result set by continuously adjusting the weights until the model meets the requirements, and then calculate the risk value of bid rigging for all projects. Determine whether further text analysis is needed based on whether the risk value of bid rigging for the project exceeds the relevant threshold. The relevant threshold is set according to the model training process and effect. It is necessary to ensure that all the existing marked data are selected in the result set, and at least 50% of the projects in the result set can be determined as bid rigging after analysis.

[0097] Furthermore, the analysis of the similarity of the technical parts of the bidding documents includes:

[0098] By identifying the relevant information of the keywords "contact person" and "telephone", the contact information of the bidding document is extracted, wherein the contact information includes the bidding contact person and contact information;

[0099] The extracted contact information is compared one-to-one between suppliers. If one or more pieces of information are consistent, the text similarity is determined to be similar;

[0100] If it does not exist, the technical solution content in the bidding document is extracted by identifying the relevant information of the keywords "solution" and "technical solution";

[0101] Use the BBPE (Byte-level BPE) algorithm at the subword granularity to segment the technical solution content;

[0102] The dual-tower SBERT model is used to calculate the final risk value of the similarity of the technical part after word segmentation.

[0103] In addition to comparing the extracted contact information one-to-one between suppliers, it can also be compared with all the supplier enterprise information extracted from the enterprise information database; if one or more pieces of information are consistent between different participants, the project bid-rigging risk value is directly adjusted to 1, and the text similarity here is also output as 1;

[0104] The final risk value of the technical part similarity is output as a value between 0 and 1.

[0105] Since the contact information determines that the text similarity is similar, the final risk value output value is directly adjusted to 1 and is specially marked.

[0106] Furthermore, the analysis of the similarity of the price part of the bidding documents includes:

[0107] Calculate the price difference of each bidder's itemized quotations, the price difference of the total price quotations, and whether there is any regularity in the quotations;

[0108] Calculate the price difference between the itemized bids of each bidder that participated but failed to win the bid and the winning bidder to see whether there is a similar price difference or a regular pattern in the overall bids;

[0109] If it exists, the final risk value of the price part similarity is determined as the highest value;

[0110] If it does not exist, the total price difference percentage and the itemized price difference percentage between all bidders shall be calculated based on the itemized price difference and the total price difference;

[0111] Output the final risk value of the price similarity based on the weighted average of the three dimensions of itemized quotation repetition rate, itemized quotation risk value, and total price quotation risk value;

[0112] Among them, the risk value of each bidder's sub-item quotation = 1-min{corresponding sub-item quotation price difference percentage}, the total price quotation risk = 1-min{corresponding total price quotation price difference percentage}, and the sub-item quotation repetition rate = the number of sub-items with consistent sub-item quotation price differences / the number of sub-items in the sub-item quotation table.

[0113] Similarly, the final risk value of the price part similarity is output as a value between 0 and 1.

[0114] When judging whether the winning bidder and the participating parties have the same price difference or generally show regular quotations, the initial value risk value = 0. If there is the same price difference and regular quotations, such as an overall decreasing regularity, the risk value is set to = 1.

[0115] Furthermore, judging whether there is bid rigging or bid collusion for the high risk projects obtained according to the risk value of bid rigging or bid collusion in the project by combining the similarity of the technical part and the similarity of the price part of the bidding documents, includes:

[0116] If the text similarity is determined to be similar, it is directly determined that there is a string of bids;

[0117] If they are not determined to be similar, the risk value of each supplier will be output according to the project's bid-rigging risk value, the final risk value of the technical similarity, and the final risk value of the price similarity using the random forest method;

[0118] For each bidding project, the risk values ​​of all participating bidders are calculated and the risk level of the bidding project is output. The risk level is divided into: high risk, medium risk and low risk.

[0119] When outputting the risk value of each supplier using the random forest method, weights can be set for each value through semi-supervised learning to output the risk value of the supplier.

[0120] The disclosed embodiment constructs a bidding combination relationship matrix to clearly sort out the relationship between the bidding activities, the successful bidders, and the participating parties, and conducts behavioral analysis on the bidding activities. The analysis dimensions include four aspects of information: participation rate, successful bid rate, quotation gap, and enterprise information consistency; and determines the bidding combination through the analysis of the bidding activities, and then simulates the weight selection of various types of information through an unsupervised learning algorithm to calculate the risk value of bid rigging and collusion for the project, and then judges the possibility of bid rigging and collusion. For projects with a high possibility of bid rigging and collusion, not only is the data analysis conducted, but also the text analysis is combined with the use of natural language processing technology to analyze the bidding documents to determine the final high-risk target projects for bid rigging and collusion. It can effectively avoid bid rigging and collusion by using weakly related or unrelated collusion to evade review, further improve the accuracy of bid rigging and collusion analysis, improve the efficiency of reviewing bid rigging and collusion in bid documents, and improve the quality and comprehensiveness of analysis.

[0121] like Figure 2As shown, the second embodiment of the present disclosure also provides a method for identifying the risk of bid rigging and bid collusion. By constructing a bidding combination relationship matrix, the relationship between procurement activities, successful bidders, and participating parties (unsuccessful bidders) is clearly sorted out, and the behavior of procurement activities is analyzed. The analysis dimensions include four aspects of information: participation rate, successful bid rate, quotation gap, and enterprise information consistency. The behavior of participating in procurement activities is analyzed to determine the bid combination, and then the XGboost algorithm in the decision tree is used to simulate the weight selection of various types of information, calculate the risk value of bid rigging and bid collusion for the project, and then judge the possibility of bid rigging and bid collusion. For projects with a high possibility of bid rigging and bid collusion, further text analysis is performed on the bidding documents of projects with the possibility of bid rigging and bid collusion to determine the final target projects with high risk of bid rigging and bid collusion. In the text analysis process, OCR text recognition is mainly used to analyze from three aspects: the content of the bid, the bidding contact person, and the similarity of the contact information. Combine NLP technology to analyze the quotation details table, technical parameters, etc.

[0122] The specific steps include:

[0123] S1: Establish a bidding relationship matrix: Establish a bidding relationship matrix for the procurement project based on the procurement activities, successful bidders, and participating parties (non-successful bidders);

[0124] S2: Clean and sort the data: obtain information on participation rate, bid winning rate, quotation gap, and enterprise information consistency;

[0125] S3: Calculate project risk value: Use the XGboost algorithm to perform unsupervised learning on all project data, calculate the risk value of project bid rigging, and use relevant thresholds to determine whether further text analysis is needed;

[0126] S4: Analyze the similarity of the technical part of the bidding documents: Use text recognition technology to compare the similarity of the bidding documents, the bidding contacts and contact information;

[0127] S5: Analyze the similarity of the price part of the bidding documents: Use natural language processing technology to compare the quotation details item by item to obtain the quotation gap information;

[0128] S6: Warning output: For high-risk items obtained through data analysis, combined with text analysis similarity, determine whether there is bid rigging or collusion, and output an analysis report.

[0129] In S1, the steps to establish the bidding relationship matrix for the procurement project based on the procurement activities, the winning bidder, and the participating parties (non-winning bidders) are:

[0130] S11: Based on the existing data, the KMP algorithm is used to perform one-to-one matching between projects and project bidders, and a two-dimensional matrix is ​​established with procurement activities as the X-axis and the bidder as the Y-axis; (Since the bidding combination relationship matrix function provided by various systems is a black box, it is impossible to filter data as needed) This solution generates the combination relationship by manually writing a program in Python: the procurement data is placed in the database, and the procurement data that needs to be analyzed is accessed through the first-level cursor loop. The second-level cursor finds all participating manufacturers in the same winning bid, and builds a matrix of participating manufacturers and procurement orders in the program and database (program array)

[0131] S12: Subsequently, multiple manufacturers participating in the same procurement are converted into a multi-dimensional matrix to achieve one-to-many matching of procurement activities, successful bidders, and multiple participants; specifically, including:

[0132] S13: Taking one participant as the main body, merge the names of multiple participants to form a unique identifier, and use this information as the Z axis to upgrade the two-dimensional matrix to a three-dimensional matrix;

[0133] S14: According to the number of participants, an N-dimensional matrix is ​​established with the information of each participant as the axis one by one.

[0134] The steps for cleaning and sorting data in S2 to obtain information on participation rate, bid winning rate, quotation gap, and enterprise information consistency are:

[0135] S20: According to the data obtained in the system, ① preliminary deletion: delete the projects with the same project identifier "result" and the number of participating suppliers "count" ≥ 15 (according to the actual procurement situation, the number of participating suppliers in various normal procurement activities shall not exceed 15), delete the data with incomplete project identifiers and corresponding supplier information, delete the data with empty or all 0 bid amounts, ② screening and merging: extract the projects with similar last digits of the project identifier "result" and completely consistent information such as participating suppliers and bid amounts, and treat them as multiple bidding records, and only retain the project corresponding to the latest result; ③ data restoration: extract the projects with bid amounts between 0 and 100, and restore all bid prices according to the conversion rules of the winning bid price and the price score of the bidding document;

[0136] S21: Taking the Y axis and Z axis as the benchmark, traverse to obtain the number of participations of each supplier; taking the year or quarter as the cycle, determine the participation rate = the number of current participations / the total number of participations;

[0137] S22: Taking the X axis as a reference, traverse to obtain the number of winning bids of all winning bidders, and determine the (current) winning rate = (current) winning number of bids / (current) participation number;

[0138] S23: Taking the procurement items on the X axis as the benchmark, extract the winning bid amount, the participating bid amount, the participating bid amount… the participating bid amount of each item in rotation, and calculate the price difference between each participating bid amount and the winning bid amount;

[0139] S24: Extract all supplier enterprise information in the system from the enterprise information database, including but not limited to enterprise telephone number, email address, address, registered legal person information, and enterprise change information;

[0140] S25: Compare the consistency of information of all participating companies in a single project.

[0141] S3 uses the XGboost algorithm to perform unsupervised learning on all project data, calculates the risk value of project bid rigging, and uses the relevant thresholds to determine whether further text analysis is needed. The steps are:

[0142] S30: Set the input value to various data of projects and suppliers, and the output value to a value between 0 and 1. Using the XGboost algorithm for a training set (a training set with known results, the risk value of problematic case projects is 1), XGboost first uses CART tree training to obtain a model, and generates a deviation value for each sample with known results; then the sample deviation value is used as a new training set, and CART tree training is continued to obtain a new model; repeat this until a risk value of ≥ 0.9 is output for all projects with known results;

[0143] S31: Consider five risk dimensions: price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate. The price is calculated based on the difference ratio with the bid winning price; enterprise information is calculated based on the legal person, contact person, phone number, address, email address, etc. If one of them is consistent, it is 1, otherwise it is 0; initially, the five risk dimensions are weighted by a coefficient of 0.2 each, and then trained and adjusted with a step size of 0.01;

[0144] S32: Use existing cases as result sets, and conduct unsupervised learning on national data in some years using indicators such as price, enterprise information, bid winning rate, bid participation rate, and co-occurrence rate;

[0145] S33: Get the result set to verify key provinces;

[0146] S34: Utilize the verification results from different provinces and continuously adjust the result set to calculate the risk value of bid rigging for all projects.

[0147] S4 uses text recognition technology to compare the similarity of the tender content, the tender contact person, and the contact information, and the steps of outputting a value between 0 and 1 are:

[0148] S41: extracting information such as the bidding contact person and contact information by identifying information related to keywords such as "contact person" and "telephone number";

[0149] S42: The extracted contact information is compared one-to-one between suppliers, and is also compared with the enterprise information obtained in S2. If one or more pieces of information are consistent, the enterprise information risk value in S2 is directly output as 1, and the text similarity here is also output as 1;

[0150] S43: extracting the technical solution content in the bidding document by identifying relevant information of keywords such as “solution” and “technical solution”;

[0151] S44: Using the BBPE (Byte-level BPE) algorithm in the subword granularity, the technical solution is segmented into words;

[0152] S45: Use the dual-tower SBERT model to calculate text similarity, and the output is a value between 0 and 1;

[0153] S46: The text similarity is used as the output value; if the value obtained in S42 is 1, the output value is directly adjusted to 1 and a special mark is made.

[0154] The steps for S5 to compare the quotation details item by item using natural language processing technology are:

[0155] S51: extracting specific keywords from the bidding document using the TF-IDF algorithm based on the keywords in the quotation details table in the bidding document;

[0156] S52: Calculate the price difference of each bidder's itemized quotation, the price difference of the total price quotation, and whether there is a regular quotation;

[0157] S521: Item-by-item price difference and total price price difference, calculate the total price and item-by-item price difference percentages between all the participating companies;

[0158] S522: Calculate the price difference of each bid item between each bidder that participated but failed to win and the bidder that won, and check whether the price difference is the same and whether the price difference is generally equal and decreasing;

[0159] S53: Output the final risk value of the price similarity module according to the weighted average of the three dimensions of item quotation repetition, item quotation risk value, and total price quotation risk value;

[0160] S531: The risk value of each participating supplier's itemized quotation = 1-min{the price difference percentage of the itemized quotation obtained in S521}, and the risk value of the total price quotation = 1-min{the price difference percentage of the total price quotation obtained in S521};

[0161] S532: Initial value risk value = 0. If there is a regularity that the price difference is the same and the regular quotations decrease overall, the risk value = 1.

[0162] S533: Sub-item quotation repetition rate = number of sub-items with consistent price differences in sub-item quotations / number of sub-items in the sub-item quotation table.

[0163] S6 analyzes the high-risk items obtained through data analysis and combines the text analysis similarity to determine whether there is bid rigging or collusion. The steps for outputting the analysis report are:

[0164] S61: Based on the comparison of the enterprise information outputted in S42, if they are consistent, directly determine that there is bid rigging; if they are inconsistent, go to S62;

[0165] S62: Based on the text similarity output by S46, the quotation similarity or whether there is a regular quotation output by S53, and the bid-rigging risk value output by S34, the three are used to output the risk value of each supplier in a random forest manner;

[0166] S621: When ensuring that the risk level of the known result set output in S34 is high risk, perform semi-supervised learning and output the risk value of the supplier.

[0167] S63: For each procurement project, calculate the risk value of all participating bidders and output the risk level of the procurement project: high risk, medium risk, low risk.

[0168] The disclosed embodiment constructs a bidding combination relationship matrix, matches the similarity from the perspective of similar risk of various materials of the same project in the matrix combination, and calculates the risk value of bid rigging and collusion by weighting the four aspects of operator data and external network enterprise personnel telephone comparison, quotation comparison, scoring detail ratio, and bidding combination relationship, and then determines the possibility of bid rigging and collusion. Then, the possibility of bid rigging and collusion in bids is analyzed by text analysis for risky projects, which improves the review efficiency, analysis quality and comprehensiveness of bid rigging and collusion. The analysis focuses on the similarity of the contact persons and contact information of the bidding companies obtained from the operator and external enterprise information, effectively avoiding bid rigging and collusion by colluding with multiple weakly related or unrelated parties to evade review, and further improves the accuracy of bid rigging and collusion analysis.

[0169] The third embodiment of the present disclosure also provides a bid-rigging and bid-collusion risk identification system, such as Figure 3 As shown, the system comprises:

[0170] A matrix establishment module 11, which is configured to establish a bidding relationship matrix for the bidding project based on the bidding activities, the winning bidder, and the information of the participating parties;

[0171] Processing module 12, which is configured to clean and sort out historical bidding data, and obtain project risk assessment information based on the bidding relationship matrix, wherein the project risk assessment information includes bid participation rate, bid winning rate, quotation gap, and enterprise information consistency results;

[0172] The project risk assessment module 13 is configured to perform unsupervised learning on the project data, calculate the risk value of the project's bid-rigging and bid-rigging according to the project risk assessment information, and determine whether the project needs further text analysis based on the risk value of the project's bid-rigging and bid-rigging;

[0173] A text analysis module 14, which is configured to analyze the similarity of the technical part of the bidding documents and the similarity of the price part of the bidding documents for the projects requiring text analysis;

[0174] The bid rigging and collusion identification module 15 is configured to determine whether there is bid rigging and collusion for the high risk value projects obtained according to the project bid rigging and collusion risk values, in combination with the similarity of the technical part and the similarity of the price part of the bidding documents, and output an analysis report.

[0175] Furthermore, the matrix establishing module 11 is specifically configured as follows:

[0176] Based on the existing bidding data, one-to-one matching is performed between projects and winning bidders, and a two-dimensional matrix with bidding projects and winning bidders as axes is established;

[0177] The data of multiple participants in the same bidding project in the bidding data are combined with a two-dimensional matrix to be converted into a multi-dimensional matrix, including: taking at least one participant as the main body, merging the names of multiple participants containing the main body into a unique identifier, and taking the identifiers of the multiple participant information as the axis, upgrading the two-dimensional matrix to a multi-dimensional matrix to achieve one-to-many matching of bidding activities, successful bidders, and multiple participants.

[0178] Furthermore, the processing module 12 is specifically configured as follows:

[0179] Initial deletion includes: deleting data with the same project identifier "result" and the number of participating suppliers "count" ≥ the preset threshold, deleting data with incomplete project identifiers and corresponding supplier information, and deleting data with empty or all zero bid amounts;

[0180] Screening and merging, including: extracting projects with similar last digits of project identifier "result", identical bidding suppliers and bidding amounts, treating them as multiple bidding records, and retaining only the project corresponding to the latest "result";

[0181] Data restoration includes: extracting all items with bidding amounts between 0 and 100, and restoring all bidding prices according to the winning bid price and bidding document price score conversion rules.

[0182] Furthermore, the processing module 12 is further configured to:

[0183] According to the information of the winning bidder and the information of multiple participants in the multidimensional matrix, the number of times each supplier participated in the bidding is obtained, and the participation rate is determined based on a certain period of time, where the participation rate = the number of current participations / the total number of participations;

[0184] According to the multidimensional matrix, taking the bidding project as the benchmark, the number of winning bids of each winning bidder is traversed to determine the winning rate of each winning bidder, wherein the winning rate = the number of winning bids in the current period / the number of participations in the current period;

[0185] Based on the bidding projects, extract the winning bid amount of each project and the bid amount of each bidder in turn, calculate the price difference between each bidder amount and the winning bid amount, and get the quotation gap;

[0186] In the enterprise information database, extract all supplier enterprise information, conduct consistency comparison on all participating enterprise information of a single project, and obtain the enterprise information consistency result.

[0187] Furthermore, the project risk assessment module 13 is specifically configured as follows:

[0188] Obtain risk dimension indicators of projects and suppliers based on project risk assessment information of projects and suppliers, wherein the risk dimension indicators include price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate;

[0189] Obtain data containing existing bid-rigging cases as a training set, use risk dimension indicators as input, use the XGboost algorithm to train the training set, and perform unsupervised learning on the data to obtain a risk prediction model;

[0190] The risk value of bid rigging for all projects is calculated through the risk prediction model.

[0191] Furthermore, the text analysis module 14 is specifically configured as follows:

[0192] By identifying the relevant information of the keywords "contact person" and "telephone", the contact information of the bidding document is extracted, wherein the contact information includes the bidding contact person and contact information;

[0193] The extracted contact information is compared one-to-one between suppliers. If one or more pieces of information are consistent, the text similarity is determined to be similar;

[0194] If it does not exist, the technical solution content in the bidding document is extracted by identifying the relevant information of the keywords "solution" and "technical solution";

[0195] Use the BBPE (Byte-level BPE) algorithm at the subword granularity to segment the technical solution content;

[0196] The dual-tower SBERT model is used to calculate the final risk value of the similarity of the technical part after word segmentation.

[0197] Furthermore, the text analysis module 14 is also specifically configured to:

[0198] Calculate the price difference of each bidder's itemized quotations, the price difference of the total price quotations, and whether there is any regularity in the quotations;

[0199] Calculate the price difference between the itemized bids of each bidder that participated but failed to win the bid and the winning bidder to see whether there is a similar price difference or a regular pattern in the overall bids;

[0200] If it exists, the final risk value of the price part similarity is determined as the highest value;

[0201] If it does not exist, the total price difference percentage and the itemized price difference percentage between all bidders shall be calculated based on the itemized price difference and the total price difference;

[0202] Output the final risk value of the price similarity based on the weighted average of the three dimensions of itemized quotation repetition rate, itemized quotation risk value, and total price quotation risk value;

[0203] Among them, the risk value of each bidder's sub-item quotation = 1-min{corresponding sub-item quotation price difference percentage}, the total price quotation risk = 1-min{corresponding total price quotation price difference percentage}, and the sub-item quotation repetition rate = the number of sub-items with consistent sub-item quotation price differences / the number of sub-items in the sub-item quotation table.

[0204] Furthermore, the bid rigging identification module 15 is specifically configured as follows:

[0205] If the text similarity is determined to be similar, it is directly determined that there is a string of bids;

[0206] If they are not determined to be similar, the risk value of each supplier will be output according to the project's bid-rigging risk value, the final risk value of the technical similarity, and the final risk value of the price similarity using the random forest method;

[0207] For each bidding project, the risk values ​​of all participating bidders are calculated and the risk level of the bidding project is output. The risk level is divided into: high risk, medium risk and low risk.

[0208] The bid rigging and collusion risk identification system of the disclosed embodiment is used to implement the bid rigging and collusion risk identification method in method embodiment 1 and embodiment 2, so the description is relatively simple. For details, please refer to the relevant description in the previous method embodiment, which will not be repeated here.

[0209] In addition, if Figure 4 As shown, the fourth embodiment of the present disclosure further provides an electronic device, including a memory 100 and a processor 200, wherein the memory 100 stores a computer program, and when the processor 200 runs the computer program stored in the memory 100, the processor 200 executes the above-mentioned various possible methods.

[0210] The memory 100 is connected to the processor 200. The memory 100 may be a flash memory, a read-only memory or other memory, and the processor 200 may be a central processing unit or a single-chip microcomputer.

[0211] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the above-mentioned various possible methods.

[0212] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), Digital Video Disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0213] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A method for identifying bid rigging and collusion risks, characterized in that: The method comprises: Establish a bidding relationship matrix for bidding projects based on bidding activities, winning bidders, and participating party information; Clean and sort out historical bidding data, and obtain project risk assessment information based on the bidding relationship matrix, including bid participation rate, bid winning rate, quotation gap and enterprise information consistency results; Perform unsupervised learning on project data, calculate the risk value of project bid rigging based on project risk assessment information, and use the risk value to determine whether the project needs further text analysis; For projects that require text analysis, we analyze the similarity of the technical part of the bidding documents and the similarity of the price part of the bidding documents; For high-risk projects obtained based on the project collusion risk value, the similarity of the technical part and the price part of the bidding documents are combined to determine whether there is collusion in bidding, and an analysis report is output.

2. The method according to claim 1, characterized in that: The bidding relationship matrix is ​​established for the bidding project based on the bidding activities, the winning bidder and the participating parties, including: Based on the existing bidding data, one-to-one matching is performed between projects and winning bidders, and a two-dimensional matrix with bidding projects and winning bidders as axes is established; The data of multiple participants in the same bidding project in the bidding data are combined with a two-dimensional matrix to be converted into a multi-dimensional matrix, including: taking at least one participant as the main body, merging the names of multiple participants containing the main body into a unique identifier, and taking the identifiers of the multiple participant information as the axis, upgrading the two-dimensional matrix to a multi-dimensional matrix to achieve one-to-many matching of bidding activities, successful bidders, and multiple participants.

3. The method according to claim 1, characterized in that The cleaning and combing of historical bidding data includes: Initial deletion includes: deleting data with the same project identifier "result" and the number of participating suppliers "count" ≥ the preset threshold, deleting data with incomplete project identifiers and corresponding supplier information, and deleting data with empty or all zero bid amounts; Screening and merging, including: extracting projects with similar last digits of project identifier "result", identical bidding suppliers and bidding amounts, treating them as multiple bidding records, and retaining only the project with the latest "result"; Data restoration includes: extracting all items with bidding amounts between 0 and 100, and restoring all bidding prices according to the winning bid price and bidding document price score conversion rules.

4. The method according to claim 2, characterized in that: The project risk assessment information is obtained based on the bidding relationship matrix, including: According to the information of the winning bidder and the information of multiple participants in the multidimensional matrix, the number of times each supplier participated in the bidding is obtained, and the participation rate is determined based on a certain period of time, where the participation rate = the number of current participations / the total number of participations; According to the multidimensional matrix, taking the bidding project as the benchmark, the number of winning bids of each winning bidder is traversed to determine the winning rate of each winning bidder, wherein the winning rate = the number of winning bids in the current period / the number of participations in the current period; Based on the bidding projects, extract the winning bid amount of each project and the bid amount of each bidder in turn, calculate the price difference between each bidder amount and the winning bid amount, and get the quotation gap; In the enterprise information database, extract all supplier enterprise information, conduct consistency comparison on all participating enterprise information of a single project, and obtain the enterprise information consistency result.

5. The method according to claim 1, characterized in that The unsupervised learning of the project data and the calculation of the risk value of the project bid rigging based on the project risk assessment information include: Obtain risk dimension indicators of projects and suppliers based on project risk assessment information of projects and suppliers, wherein the risk dimension indicators include price, enterprise information, bid winning rate, bid participation rate, and supplier co-occurrence rate; Obtain data containing existing bid-rigging cases as a training set, use risk dimension indicators as input, use the extreme gradient boosting XGboost algorithm to train the training set, and perform unsupervised learning on the data to obtain a risk prediction model; The risk value of bid rigging for all projects is calculated through the risk prediction model.

6. The method according to claim 1, characterized in that The analysis of the similarity of the technical part of the bidding documents includes: By identifying the relevant information of the keywords "contact person" and "telephone number", the contact information of the bidding document is extracted, wherein the contact information includes the bidding contact person and contact information; The extracted contact information is compared one-to-one between suppliers. If one or more pieces of information are consistent, the text similarity is determined to be similar; If it does not exist, the technical solution content in the bidding document is extracted by identifying the relevant information of the keywords "solution" and "technical solution"; The technical solution content is segmented using the byte-level byte pair encoding (BBPE) algorithm at the subword granularity; The twin-tower twin network SBERT model is used to calculate the final risk value of the similarity of the technical part after word segmentation.

7. The method according to claim 6, characterized in that The analysis of the similarity of the price part of the bidding documents includes: Calculate the price difference of each bidder's itemized quotations, the price difference of the total price quotations, and whether there is any regularity in the quotations; Calculate the price difference between the bids of each bidder that participated but failed to win the bid and the bidder that won the bid, and check whether there is a situation where the price difference is the same or there is a regular pattern in the overall quotations; If it exists, the final risk value of the price part similarity is determined as the highest value; If it does not exist, the total price difference percentage and the itemized price difference percentage between all bidders shall be calculated based on the itemized price difference and the total price difference; Output the final risk value of the price similarity based on the weighted average of the three dimensions of itemized quotation repetition rate, itemized quotation risk value, and total price quotation risk value; Among them, the risk value of each bidder's sub-item quotation = 1-min{corresponding sub-item quotation price difference percentage}, the total price quotation risk = 1-min{corresponding total price quotation price difference percentage}, and the sub-item quotation repetition rate = the number of sub-items with consistent sub-item quotation price differences / the number of sub-items in the sub-item quotation table.

8. The method according to claim 7, characterized in that The high-risk projects obtained according to the risk value of bid rigging are judged whether there is bid rigging or bid collusion by combining the similarity of the technical part and the similarity of the price part of the bidding documents, including: If the text similarity is determined to be similar, it is directly determined that there is a string of tags; If they are not determined to be similar, the risk value of each supplier will be output according to the project's bid-rigging risk value, the final risk value of the technical similarity, and the final risk value of the price similarity using the random forest method; For each bidding project, the risk values ​​of all participating bidders are calculated and the risk level of the bidding project is output. The risk level is divided into: high risk, medium risk and low risk.

9. A bid rigging and collusion risk identification system, characterized in that: The system comprises: A matrix establishment module is configured to establish a bidding relationship matrix for a bidding project based on information on bidding activities, successful bidders, and participating parties; A processing module is configured to clean and sort out historical bidding data, and obtain project risk assessment information based on a bidding relationship matrix, wherein the project risk assessment information includes a bid participation rate, a bid winning rate, a bid gap, and a result of consistency of enterprise information; The project risk assessment module is configured to perform unsupervised learning on project data, calculate the risk value of bid rigging based on the project risk assessment information, and determine whether the project needs further text analysis based on the risk value of bid rigging; A text analysis module, which is configured to analyze the similarity of the technical part of the bidding documents and the similarity of the price part of the bidding documents for the projects requiring text analysis; The bid rigging and collusion identification module is configured to determine whether there is bid rigging and collusion for high-risk projects obtained based on the project bid rigging and collusion risk values, combined with the similarity of the technical part and the similarity of the price part of the bidding documents, and output an analysis report.

10. An electronic device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for identifying the risk of bid rigging and collusion as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying the risk of bid rigging and collusion according to any one of claims 1 to 8 is implemented.

Images