Artificial intelligence-based bid enclosing and bid stringing behavior monitoring management platform

Abstract

The invention belongs to the technical field of bid and bid behavior monitoring, and particularly discloses a bid and bid behavior monitoring management platform based on artificial intelligence, and the bid and bid behavior monitoring management platform adopts a chart normalization preprocessing technology, carries out the multi-dimensional similarity analysis of text content, chart features and image-text matching degree, and combines the order of images and texts with a recombination comparison mode. Omnibearing detection of the bidding document is realized, and the recognition rate of chart tampering and content recombination type bidding behaviors is effectively improved; based on a dynamic weight distribution mechanism of a project type, accurate risk assessment of different importance contents is realized through intelligent segmentation and differential weight setting of bidding document unit bodies; the multi-dimensional risk fusion analysis relation is constructed, the risk index is generated through comprehensive evaluation, and graded early warning is triggered, so that targeted monitoring on different structural weight markers is ensured, the detection breadth and depth are remarkably improved, and the problem of stiffness of a multi-modal analysis evaluation standard in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of monitoring bid rigging behavior, and relates to a monitoring and management platform for bid rigging behavior based on artificial intelligence. Background Art

[0002] With the increasing frequency of bidding activities in fields such as engineering construction, illegal and irregular behaviors such as bid rigging show a trend of being concealed and intelligent. Traditional manual verification methods rely on empirical judgment and limited data analysis, and it is difficult to cope with changing complex illegal behaviors, resulting in low supervision efficiency, and thus bringing major hidden dangers to market fair competition and asset security.

[0003] In the prior art, there are also some solutions related to the monitoring of bid rigging behavior. For example, the patent with the Chinese patent publication number CN112561670B discloses an intelligent identification system for bid rigging, which integrates the functions of online bidding and bid rigging identification. Under the bid rigging identification subsystem, it identifies the joint bid-rigging behavior, price joint behavior, and control of bid evaluation price behavior of bidders. Although it can identify bid rigging behaviors in bidding, it is mostly limited to simple comparison of the order content of bid documents, lacking the detection ability for concealed tampering means such as position replacement of chapter content and paragraph recombination. Especially, there are technical shortcomings in the semantic consistency analysis of non-continuous texts, resulting in poor identification effects for structural bid rigging behaviors.

[0004] Another patent with the Chinese patent publication number CN110992059B discloses a method for identifying and analyzing bid rigging behavior based on big data. It constructs a community detection model based on complex networks and identifies bid rigging behavior by analyzing the correlation relationship between bidding entities, which reflects the internal root cause of bid rigging behavior to a certain extent, but there are still significant limitations: overly relying on structured data (such as enterprise relationships, quotations, etc.), lacking the multi-modal analysis ability for non-text elements such as charts in bid documents, and thus it is difficult to identify new types of bid rigging behaviors based on visual data tampering; the evaluation strategy is rigid, and the weight allocation is not dynamically optimized in combination with the characteristics of project types, resulting in inaccurate risk assessment of key chapters such as technical bids and commercial bids, affecting the monitoring accuracy. Summary of the Invention

[0005] In view of this, to solve the problems raised in the above background art, a monitoring and management platform for bid rigging behavior based on artificial intelligence is proposed.

[0006] The object of the present invention can be achieved by the following technical solutions: The present invention provides a monitoring and management platform for bid rigging behavior based on artificial intelligence, including: a data collection module, which collects bid document data including bid documents, enterprise historical bidding records, and enterprise qualification information.

[0007] The document splitting module splits the text chapters and paragraphs of the tender document and marks them as individual units, and determines the project type of the tender document and the construction weights of each individual unit under the corresponding project type.

[0008] The behavior analysis module conducts multi-dimensional bid rigging behavior analysis on the similar writing features, bid price jump features, and lack of authenticity of enterprise qualifications features of each individual unit of the tender document based on the tender document data.

[0009] The risk assessment module correlates and integrates the construction weights of each individual unit with the multi-dimensional bid rigging behavior, evaluates the risk index of the bid rigging behavior of the tender document, and triggers differentiated warning signals according to the risk index.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention adopts the chart normalization preprocessing technology. Through multi-dimensional similarity analysis of text content, chart features, and graphic-text matching degree, combined with the order and recombination comparison method of graphics and text, it realizes the comprehensive detection of the tender document, and effectively improves the recognition rate of bid rigging behaviors such as chart tampering and content recombination.

[0011] (2) The present invention is based on the dynamic weight allocation mechanism of the project type. Through the intelligent splitting and differentiated weight setting of the tender document units, it realizes the accurate risk assessment of content with different importance levels.

[0012] (3) The present invention constructs a multi-dimensional risk fusion analysis relationship. Through comprehensive evaluation, a risk index is generated and a hierarchical warning is triggered, ensuring the targeted monitoring of tender bodies with different structural weights, significantly improving the breadth and depth of detection, and solving the problem of rigidity of multi-modal analysis and evaluation criteria in the prior art. Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 It is a schematic diagram of the module connection of the present invention.

[0015] Figure 2 It is a schematic diagram of the sequential matching of the document content of each individual unit of the tender document of the present invention and the corresponding individual units of the competitive tender document.

[0016] Figure 3 It is a display diagram of the re-matching and typesetting of each individual unit of the tender document of the present invention.

[0017] Figure 4Schematic diagram of multiple recombinant control matrices of the present invention.

[0018] Reference numerals: 1, 2, and 3 respectively represent the positions of each unit of the bidding document, 11, 22, and 32 respectively represent the positions of each unit of the competitive bidding document, and ① to n represent the numbers of the recombinant control matrix. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Please refer to Figure 1 As shown, the present invention provides a monitoring and management platform for bid rigging and collusive bidding behaviors based on artificial intelligence, including: a data acquisition module, a document segmentation module, a behavior analysis module, and a risk assessment module, which are connected in sequence.

[0021] The data acquisition module collects bid document data including bidding documents, enterprise historical bidding records, and enterprise qualification information through an interface to dock with the bidding and tendering system.

[0022] The enterprise historical bidding records include the corresponding historical quotation records of each project type, the bidding results of winning or not winning the bid, and the enterprise historical average quotation.

[0023] The document segmentation module divides the text chapters and paragraphs of the bidding document and marks them as each unit. Each chapter and paragraph corresponds to one unit, determines the project type of the bidding document and the construction weights of each unit under the corresponding project type. The project types of the bidding document include construction engineering, equipment technology, consulting, cleaning, etc.

[0024] In a preferred implementation manner, the determination of the project type of the bidding document and the construction weights of each unit under the corresponding project type includes: obtaining the project title of the bidding document, matching it with the corresponding keyword libraries of each project type, obtaining the quantity matching ratio of the project title phrases of the bidding document compared with the phrases in the corresponding keyword libraries of each project type, and defining the project type corresponding to the maximum quantity matching ratio as the project type of the bidding document.

[0025] The keyword library is used to store the corresponding title phrases of each project type. For example, the project title phrases for construction engineering include: municipal roads, renovation, engineering, construction, construction bidding; the project title phrases for equipment technology include: high-precision, numerical control machine tools, automation, production lines, equipment bidding.

[0026] Integrate the chapter titles and paragraph contents of each unit, perform text structure parsing on them through natural language processing technology, determine the content themes of each unit under the project type of the bidding document, and then set the construction weights of the corresponding content themes of each unit under this project type according to the requirements of the tender, where the content themes include background technology summary, project technology introduction, financial status report, etc.

[0027] The key requirements of tenders for different project types of bidding documents may focus on dimensions such as the feasibility, innovation, and financial status of the technical solutions, which in turn makes the construction weights of content themes such as project technology introduction and financial status report different in the bidding documents, specifically set by the tender administrator. For example, set the constraint value of the corresponding comprehensive construction weight of all units to 10. In the construction project type, when its tender requirement focuses on the innovation of the technical solution, set the construction weight of the corresponding abstract unit in the project technology introduction part to 5, and set the construction weight of the corresponding content unit in the financial status report part to 3; in the cleaning project type, when its tender requirement focuses on the financial status of the technical solution, set the construction weight of the corresponding abstract unit in the project technology introduction part to 2, and set the construction weight of the corresponding content unit in the financial status report part to 5.

[0028] The behavior analysis module conducts multi-dimensional analysis of bid rigging behaviors on the similar writing characteristics, price jump characteristics, and lack of authenticity of enterprise qualifications of each unit of the bidding document based on the bidding document data.

[0029] In a preferred implementation manner, the similar writing characteristics of each unit include: obtaining the data of each competing bid document during the same period from the bidding and tendering system, extracting the competing content themes of each competing bid document and its each unit, and thereby determining the similar writing characteristics of the text content of each unit of the bidding document and each competing bid document. , is the number of each unit of the bidding document, , is the number of each competing bid document, .

[0030] By preprocessing the bidding document and each bidding document with normalized chart format conversion, the table element data set and curve graph coordinate data set of each unit body of the bidding document and each bidding document are obtained. Specifically, the table-like graphics contained in each unit body of the document are converted into a matrix format, and the row elements and column elements in the matrix are marked as the table element data set; the curve graph-like graphics are converted into a coordinate system format, and the coordinate axes, block labels and other data attributes in the coordinate system are marked as curve graph coordinate data sets, where the coordinate axes include horizontal axis units (time units, weight units, currency units, etc.), vertical axis units (sales, output, etc.), and block labels include block area, block position, block marking symbols, etc., and the center points of different blocks in the curve graph are used as position point coordinates, so as to determine the writing features of the chart contents of each unit body of the bidding document and each bidding document that are similar. .

[0031] By searching for matching contents of graphics and texts in the unit, the matching features of graphics and texts in each unit of the bidding document are determined. .

[0032] Integrate the above multiple features to determine the similar writing features of each unit ,in It indicates the maximum value of the comprehensive similarity features of the text content and graphic content of each unit of the bidding document and each bidding document. Specifically, the similarity writing features of the text content and the similarity writing features of the graphic content reflect the similarity between the content of the bidding document itself and the content of the bidding document, and then the feature value with the highest similarity with all bidding documents is selected as the final similarity judgment standard of the bidding document; the graphic content matching features of each unit of the bidding document are relatively high, indicating that the writing content of the bidding document itself is relatively standardized, that is, the probability of each unit of the bidding document itself participating in bid rigging and collusion is relatively low, and then the graphic content matching features of each unit of the bidding document can be used as the quality correction factor for comparing the content of the bidding document itself with the content of the bidding document.

[0033] See also Figure 2 , Figure 3 As shown, in a further preferred embodiment, the corresponding determination method of the writing features of the text content similarity between each unit of the bidding document and each bidding document is as follows: S3.1, based on the content attributes of the corresponding documents such as content theme, text structure, text repetition rate, and chart style, the sequential content similarity features of each corresponding unit of the bidding document and each corresponding unit of the bidding document are determined according to the assignment operation of unit matching content, the text structure such as the number of words in the main title or subtitle, font format, etc.; the text repetition rate such as the text repetition rate of ordinary paragraphs, the overlap rate of error marks, etc.; the chart style such as table style, curve graph style, structure diagram style, etc.

[0034] Specifically, the document content of each unit of the tender document is sequentially matched with the corresponding units of the competitive bidding document, and assignment operations are performed on the matched content according to preset rules.

[0035] The preset rules are to set different assignments for different content attributes. For example, if the content theme of a certain unit of the tender document is the same as that of a certain unit of the competitive bidding document, then the content theme in the corresponding unit of the competitive bidding document is assigned a value of 2; if the subtitle font format of a certain unit of the tender document is the same as that of a certain unit of the competitive bidding document, then the subtitle font format in the corresponding unit of the competitive bidding document is assigned a value of 1; if the chart style of this unit of the tender document is also the same as that of this unit of the competitive bidding document, then the chart style in the corresponding unit of the competitive bidding document is assigned a value of 1. In this way, the total assignments of each unit of the tender document and each unit of the competitive bidding document are obtained, mapped into matching values, and then, according to the layout order of each unit of the competitive bidding document, the sequential content matching values of the corresponding units of the tender document and the corresponding units of the competitive bidding document are identified, which is the sequential content similarity feature.

[0036] S3.2. Identify the similar units of each unit of the tender document in the competitive bidding document according to the sequential content of the units, and re-match and typeset each unit of the tender document according to the layout order of the similar units in the competitive bidding document. Then, determine the re-adjusted content similarity feature of each unit of the tender document and the corresponding similar units of the competitive bidding document according to the assignment of the matched content.

[0037] Specifically, screen out the units corresponding to the highest matching value among the matching values of each unit of the tender document and each unit of the competitive bidding document as the similar units of each unit of the tender document in the competitive bidding document, and extract its matching value as the sequential content similarity feature of each unit of the tender document and the corresponding similar units of the competitive bidding document.

[0038] S3.3. Integrate the sequential content similarity feature and the re-adjusted content similarity feature into the similar writing feature of the text content. For example, accumulate the sequential content similarity feature and the re-adjusted content similarity feature. In this way, match the text content of the tender document with each competitive bidding document in turn to obtain the similar writing feature of the text content of each unit of the tender document and each competitive bidding document.

[0039] Please refer to Figure 4As shown, in a further preferred embodiment, the writing features of the chart content of each unit of the tender document and each competitive bidding document include: extracting row and column elements from the corresponding table element datasets of each unit of the tender document, and converting the row and column elements into coding symbols in the matrix arrangement format. For example, "printer" in "equipment type" is coded as D01 (D represents printing equipment, and 01 is the specific equipment serial number in this category), "copier" is coded as F01 (F represents copying equipment), "scanner" is coded as S01 (S represents scanning equipment), and P11 - P23 are the corresponding performance parameter values or operating parameter values of the row and column elements, such as consumable data (toner / ink capacity), network functions (network support protocol, security authentication protocol), basic parameters (print speed, resolution), etc.

[0040] By re - arranging the positions of the row and column elements in the corresponding table element datasets of each unit of the tender document, multiple re - organized comparison matrices are generated, and the matching degree of the coding symbols in the conversion matrix of the corresponding similar unit of the competitive bidding document is analyzed. The highest coding symbol matching degree is defined as the table matching degree between the corresponding unit of the tender document and the corresponding similar unit of the competitive bidding document.

[0041] Specifically, the proportion of the number of identical row and column elements in the same position compared to the total number of elements is defined as the coding symbol matching degree. For example, if a re - organized comparison matrix and the conversion matrix in the corresponding similar unit of the competitive bidding document contain M identical elements, and the re - organized comparison matrix contains N total elements, then the coding symbol matching degree between the re - organized comparison matrix and the conversion matrix in the corresponding similar unit of the competitive bidding document is M / N.

[0042] By comparing the coincidence rate of the curve atlas coordinate datasets of each unit of the tender document and the corresponding similar units of the competitive bidding documents, the similarity of the curve atlases between the corresponding units of the tender document and the corresponding similar units of the competitive bidding documents is determined.

[0043] Specifically, by comparing the curve atlas coordinate datasets of each unit of the tender document and the corresponding similar units of the competitive bidding documents, the similarity of the corresponding axis attributes of the curve atlas coordinate datasets, as well as the similarity of block labels such as block area, block position, and block marking symbols, can be obtained. Then, the coincidence rate is obtained by accumulating the multi - dimensional similarities. Among them, the similarity of the axis attributes can be determined by calculating the proportion of the number of matching coordinate units of the axis compared to the total number of words, the block area can be determined by the area ratio, the block position can be determined by the Euclidean distance formula, and the block marking symbols can be determined by calculating the proportion of the number of matching marking symbols compared to the total number of marking symbols.

[0044] Statistical table matching degrees and curve atlas similarity degrees of each unit of the tender document and the corresponding similar units of each competitive bidding document are constructed as the writing features of the chart content similarity between each unit of the tender document and each competitive bidding document.

[0045] In a further preferred embodiment, the method for determining the graphic and text content matching features of each unit of the tender document includes: retrieving elements or data attributes in the table element data set and the curve atlas coordinate data set from the text content of each unit of the tender document, and identifying the matching degree between the retrieved elements or data attributes and the corresponding source data in the text content, so as to determine the graphic and text content matching features of each unit of the tender document.

[0046] Specifically, it is judged whether there is source data by retrieving whether the elements or data attributes included in the table element data set or the curve atlas coordinate data set match the text content of the corresponding unit of the tender document. If they match, it means there is source data; otherwise, there is no source data. For example, it is identified whether the value range of the element values in the table element data set matches the corresponding element values in the text content of the corresponding unit of the tender document, and whether the coordinate axes and block labels in the curve atlas coordinate data set have matching texts in the text content of the corresponding unit. If they match, it means there is source data; otherwise, there is no source data. Furthermore, the proportion of the comprehensive data with source data in the elements or data attributes included in the table element data set or the curve atlas coordinate data set is counted, which is the graphic and text content matching feature of the corresponding unit.

[0047] In a further preferred embodiment, the bid price jump feature is obtained by accumulating multi-dimensional bid price change difference features.

[0048] The multi-dimensional bid price change difference features include: based on the enterprise's historical tender records, examining the difference rate of the commercial bid price under the current tender document project type compared with the enterprise's historical average bid price and the change fitting rate compared with the market response. .

[0049] Specifically, the difference rate of the commercial bid price of the tender document compared with the enterprise's historical bid price is determined by identifying the ratio of the difference value between the commercial bid price of the tender document and the enterprise's historical average bid price to the conventional difference value.

[0050] By the change fitting rate of the commercial bid price of the tender document compared with the market response is determined. Among them, the current bid price change rate refers to the percentage change of the commercial bid price of the current tender document compared with the enterprise's previous bid price for the same type of project, and the market price change rate is the percentage change of the price of the same type of project calculated based on market data. The closer the change fitting rate is to 100%, the higher the fitting degree of the enterprise's commercial bid price and the market response.

[0051] Obtain the set of price differences between the tender documents and each competitive bid document for arithmetic pattern analysis to determine the commercial price deviation rate of the tender documents. Among them, the arithmetic patterns include equal differences where each difference element in the set of price differences can form an arithmetic sequence, fluctuating differences where each difference element in the set of price differences is within a certain difference value range, and irregular differences other than the above two cases. Furthermore, take the ratio of the corresponding equal difference value of the arithmetic sequence or the average fluctuating difference value or the corresponding preset fixed value of the irregular difference to the preset reference difference as the commercial price deviation rate of the tender document.

[0052] For example, in units of ten thousand yuan, when the set of price differences between the tender document and each competitive bid document is {15, 20, 35, 40}, then take the ratio of the equal difference value 5 to the preset reference difference value 50, which is 0.1, as the commercial price deviation rate of the tender document; when the set of price differences between the tender document and each competitive bid document is {15, 22, 31, 39}, where each difference element is 7, 9, 8 and all are within the range of 10, then take the ratio of the average fluctuating difference value to the preset reference difference value 50, which is 0.16, as the commercial price deviation rate of the tender document; when the set of price differences between the tender document and each competitive bid document is {10, 22, 32, 45}, which belongs to irregular differences other than the above two cases, then take the ratio of the corresponding preset fixed value 2 of the irregular difference to the preset reference difference value 50, which is 0.04, as the commercial price deviation rate of the tender document.

[0053] In a further preferred implementation manner, the characteristics of the lack of authenticity of the enterprise qualification include: verifying the authenticity characteristics of the enterprise unit information by docking with the industrial and commercial registration system , specifically, obtain the number of corresponding historical successful tender records of the enterprise that can be retrieved in the industrial and commercial registration system, take the number of historical successful tender records as the authenticity characteristics of the corresponding enterprise unit information, and set the authenticity characteristics of the enterprise unit information that cannot be retrieved in the industrial and commercial registration system to -1. For example, if the number of corresponding historical successful tender records of a certain enterprise retrieved in the industrial and commercial registration system is 2, then take 2 as the final authenticity characteristics of the enterprise unit information; if a certain enterprise cannot be retrieved in the industrial and commercial registration system, then take -1 as the authenticity characteristics of the enterprise unit information.

[0054] Capture the tender IP addresses of the tender documents and each competitive bid document, define that the tender IP address of the tender document does not match the enterprise address, or different enterprise tender documents are uploaded from the same IP, or other abnormal IP address behaviors as address anomalies, and analyze the lack of authenticity characteristics of the tender address by identifying the address anomalies between the tender document and at least one competitive bid document Specifically, the number of specific phenomena involved in the address anomalies of the bidding documents and the number of bidding documents related thereto are counted, and the corresponding proportion is obtained by comparing the above cumulative number with the total number of all anomalies and all bidding documents, and the proportion result is used as the feature of the lack of authenticity of the bidding address.

[0055] For example, if there is an address anomaly phenomenon in which different corporate bids are uploaded on the same IP, the number of specific phenomena involved in the bid document in terms of address anomaly and time anomaly is recorded as 1; at the same time, the number of different corporate bids uploaded on the same IP is obtained, which is the number of related bidding documents, such as 2; the above-mentioned cumulative number 3 is obtained by statistics, and it is compared with the total number of all anomalies and all bidding documents (such as 10), and the corresponding proportion is 3 / 10, which is the feature of missing authenticity of the bidding address.

[0056] Integrate the authenticity characteristics of enterprise unit information and the authenticity loss characteristics of bidding address into the authenticity loss characteristics of enterprise qualifications in bidding documents ,in They represent the preset enterprise unit information authenticity feature and bidding address authenticity missing feature corresponding assignment weights, which are used to map the influence of different feature parameters on the enterprise qualification authenticity missing feature, and can be set through experience, such as .

[0057] The present invention adopts chart normalization preprocessing technology, through multi-dimensional similarity analysis of text content, chart features and chart-text matching, combined with the order and reorganization comparison method of chart and text, to achieve all-round detection of bidding documents and effectively improve the recognition rate of collusion in bidding such as chart tampering and content reorganization.

[0058] The risk assessment module associates and integrates the construction weights of each unit with multi-dimensional bid-rigging and collusion behaviors, evaluates the bid-rigging and collusion risk index of the bidding documents, and triggers differentiated early warning signals based on the risk index.

[0059] In a preferred embodiment, the bid-rigging and bid-collusion risk index of the bidding document includes: accumulating the weight of each unit and the similar writing features of each unit of the bidding document, and taking the price jump feature and the authenticity loss feature of the bidding document as the risk compensation factor. Mapped in the range of 0-1, the fusion calculation obtains the bid-rigging and bid-collusion risk index of the bidding document. ,in represents the construction weight of the i-th unit, Indicates the price jump characteristics of the bidding document. The specific examples are as follows:

[0060] Table 1: Example of bid-rigging and collusion risk index data

[0061]

[0062] The dynamic weight allocation mechanism of the present invention based on project types realizes accurate risk assessment of content with different importance levels through intelligent segmentation of tender document units and differential weight settings.

[0063] In a further preferred embodiment, triggering differential warning signals according to the risk index includes: triggering a warning signal when the risk index of bid rigging behavior in the tender document exceeds a preset allowable value, and integrating the text content similarity writing features and chart content similarity writing features of each unit of the tender document and each competing tender document.

[0064] Define that at least one similarity writing feature in the text content and chart content exceeding a similarity threshold is used as a screening condition.

[0065] Count the enterprises corresponding to the competing tender documents that meet the screening conditions, and count the increase value of the corresponding similarity writing features of the competing tender documents of the corresponding enterprises exceeding the similarity threshold.

[0066] Based on the increase value, perform interval division of different risk levels, allocate warning signals for different risk levels, obtain the risk level of the enterprises that meet the screening conditions, and trigger a warning according to the allocated warning signal. During the process of triggering a warning by the warning signal, the larger the increase value, the higher the risk of bid rigging between the enterprise corresponding to the competing tender document and the enterprise corresponding to the tender document, and thus the higher the level of the matching warning signal; conversely, the smaller the increase value, the lower the level of the matching warning signal. For example, set the increase value as [J1, J2], (J2, J3], (J3, J4], then J1 - J2 corresponds to a high risk level, J2 - J3 corresponds to a medium risk level, J3 - J4 corresponds to a low risk level, and J1, J2, J3, J4 respectively correspond to set constants, where J1 > J2 > J3 > J4. Set different warning signals for different intervals. For example, set a warning pop-up window for enterprises with a high risk level, a risk assessment pop-up window for medium risk levels, and a warm reminder pop-up window for low risk levels.

[0067] The present invention constructs a multi-dimensional risk fusion analysis relationship, generates a risk index through comprehensive evaluation and triggers hierarchical warnings, ensuring targeted monitoring of bid bodies with different structural weights, significantly improving the breadth and depth of detection, and solving the problem of rigid multi-modal analysis and evaluation criteria in the prior art.

[0068] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

Claims

1. An artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform, characterized by: include: Data collection module, which collects bidding document data including bidding documents, historical bidding records of enterprises, and enterprise qualification information; The document segmentation module segments the text of the bidding document into chapters and paragraphs and marks them into units, determines the project type of the bidding document and the construction weight of each unit under the corresponding project type; The behavior analysis module conducts multi-dimensional bid-rigging and collusion analysis based on the bidding document data, including similar writing features of each unit of the bidding document, price jump features, and lack of authenticity of enterprise qualifications; The risk assessment module correlates and integrates the construction weights of each unit with multi-dimensional bid-rigging and collusion behaviors, evaluates the bid-rigging and collusion risk index of the bidding documents, and triggers differentiated early warning signals based on the risk index.

2. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 1 is characterized in that: The determination of the bidding document project type and the construction weight of each unit under the corresponding project type includes: obtaining the project title of the bidding document, matching it with the associated word library corresponding to each project type, and obtaining the bidding document project type; Integrate the chapter titles and paragraph contents of each unit, perform text structured analysis through natural language processing technology, determine the content theme of each unit under the bidding document project type, and then focus on setting the construction weight of the corresponding content theme of each unit under the project type according to the bidding requirements.

3. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 1 is characterized in that: The similar writing features of each unit body include: Obtain the bidding document data of the same period, extract the bidding content themes of each bidding document and its units, and determine the similar writing features of the text content of each unit of the bidding document and each bidding document; By preprocessing the bidding document and each bidding document by normalizing the chart format, the table element data set and the curve graph coordinate data set of each unit body of the bidding document and each bidding document are obtained, and then the writing features of the chart content similarity between each unit body of the bidding document and each bidding document are determined; By searching for matching contents of graphics and texts in the units, the matching features of graphics and texts in each unit of the bidding document are determined; The above multiple features are integrated to determine similar writing features of each unit.

4. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 3 is characterized in that: The corresponding determination method of the writing features of the units of the bidding document and the text contents of the bidding documents are similar is as follows: Based on the corresponding document content attributes of content theme, text structure, word repetition rate, and chart style, the order content similarity characteristics of each unit body corresponding to the bidding document and each unit body corresponding to the tender document are determined by the assignment operation of unit body matching content; Identify similar units of each unit of the bidding document in the tender document according to the unit sequence content, and re-match and layout each unit of the bidding document according to the layout order of the similar units in the tender document, and then determine the similarity characteristics of the readjusted content of each unit of the bidding document and the corresponding similar units of the tender document according to the assignment of matching content; The similarity features of sequential content and the similarity features of re-adjusted content are integrated into the similarity writing features of text content, and the similarity writing features of the text content of each unit of the bidding document and each bidding document are obtained.

5. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 4 is characterized in that: The units of the bidding document have similar writing features to the charts and tables of the bidding documents, including: Extract row and column elements from the corresponding table element data set of each unit of the bidding document, and convert the row and column elements into coding symbols according to the matrix arrangement format; By reorganizing the positions of row and column elements of the table element data set corresponding to each unit of the bidding document, multiple reorganized comparison matrices are generated, and the coding symbol matching degree between the reorganized comparison matrices and the conversion matrix in the corresponding similar unit of the bidding document is analyzed, and the highest coding symbol matching degree is defined as the table matching degree between the corresponding unit of the bidding document and the corresponding similar unit of the bidding document; By comparing the overlap rate of the curve graph coordinate data sets in each unit body of the bidding document and the corresponding similar unit body of the bidding document, the similarity of the curve graph of the corresponding unit body of the bidding document and the corresponding similar unit body of the bidding document is determined; The table matching degree and curve graph similarity of each unit of the bidding document and the corresponding similar unit of each bidding document are statistically analyzed to construct the writing features of the similarity between the chart contents of each unit of the bidding document and each bidding document.

6. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 3 is characterized in that: The method of determining the matching features of the graphic and text content of each unit of the bidding document includes: searching for elements or data attributes in a table element data set and a curve graph coordinate data set in the text content of each unit of the bidding document, and identifying the degree of matching between the retrieved elements or data attributes and the corresponding source data in the text content, so as to determine the matching features of the graphic and text content of each unit of the bidding document.

7. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 1 is characterized in that: The quotation jump feature is obtained by accumulating multi-dimensional quotation change difference features, and the multi-dimensional quotation change difference features include: Based on the company's historical bidding records, check the difference rate of the commercial quotation under the current bidding document project type compared with the company's historical average quotation, and the change fit rate compared with the market response; Obtain a set of price differences between the bid document and each bidding document and conduct an arithmetic analysis to determine the commercial price deviation rate of the bid document.

8. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 1 is characterized in that: The authenticity of the enterprise qualifications is lacking, including: Verify the authenticity of enterprise information by connecting to the industrial and commercial registration system; Capturing the bidding IP addresses of the bidding document and each bidding document, and analyzing the authenticity loss feature of the bidding address of the bidding document by identifying the address anomaly between the bidding document and at least one bidding document; The authenticity features of enterprise unit information and the authenticity loss features of bidding address are integrated into the authenticity loss features of enterprise qualifications in bidding documents.

9. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 1 is characterized in that: The bid-rigging and collusion risk index for evaluating bid documents comprises: accumulating the construction weights of each unit and the similar writing features of each unit of the bid document, and taking the quotation jump features and the lack of authenticity of enterprise qualifications of the bid document as risk compensation factors, and fusion calculation to obtain the bid-rigging and collusion risk index of the bid document.

10. The artificial intelligence-based bid-rigging and bid-collusion monitoring and management platform according to claim 3 is characterized in that: The differentiated early warning signals triggered according to the risk index include: triggering an early warning signal when the bid-rigging and collusion risk index of the bidding document exceeds a preset allowable value, and integrating the similarity writing features of the text content of each unit of the bidding document and each bidding document, and the similarity writing features of the chart content; Define at least one similar writing feature in the text content and the graphic content exceeding a similarity threshold as a screening condition; Count the corresponding enterprises of the bidding documents that meet the screening conditions, and count the increase values ​​of the corresponding similar writing features of the bidding documents of the corresponding enterprises that exceed the similarity threshold; Different risk levels are divided based on the increase value, and early warning signals are assigned to different risk levels. The risk level of enterprises that meet the screening conditions is obtained, and early warnings are triggered according to the assigned early warning signals.

Images