Collaborative bidding and bidding risk assessment method and system based on mapping knowledge domain and large model

Abstract

The invention relates to the field of artificial intelligence and big data analysis, and discloses a method and a system for evaluating the risk of bidding and consignment based on a knowledge graph and a big model, so as to improve the accuracy and the efficiency of the risk evaluation of bidding and consignment. In the scheme of the invention, through automatic data acquisition, information of enterprise shareholders, high management, historical transactions and the like is obtained from an enterprise big data platform, and an association network between enterprises and a relation network between high management are constructed based on a graph construction technology, so that potential subjects and behavior patterns of bid enclosing and bid stringing can be visually presented. In addition, on the basis of the OCR technology, bidding and tendering files are identified and compared, similarity features in the files can be found, and therefore the bid surrounding and tendering possibility is further proved. And finally, comprehensive analysis and evaluation are carried out by using the large model in combination with the multi-source data, so that intelligent identification and risk evaluation of bid enclosing and bid stringing behaviors are realized.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and big data analysis, and specifically to a method and system for assessing the risk of bid rigging and collusion based on knowledge graphs and big models. Background Art

[0002] In recent years, with the widespread adoption of bidding mechanisms, as a means of market-based competition, bidding has played a vital role in various fields, including engineering construction, procurement, and services. However, bid rigging and collusion not only undermine fair competition in the market but also seriously impact project quality and economic returns. How to effectively identify and prevent bid rigging through technical means has become a pressing issue for regulators and businesses.

[0003] Traditional bid rigging identification relies primarily on manual review and empirical judgment, which has the following limitations: (1) The amount of data is huge and scattered: Information related to bidding and tendering is scattered in different systems and data sources, including company registration information, historical transaction data, bidding and tendering announcements, etc., which makes it difficult for manual analysis to fully cover them.

[0004] (2) Highly concealed: Bid rigging and collusion are usually hidden through complex networks of relationships, such as through secret connections between affiliated companies or different executives, and traditional means are difficult to fully reveal.

[0005] (3) Inefficiency: Manual review requires a lot of time and manpower, and efficiency is difficult to guarantee when faced with massive amounts of bidding data.

[0006] Therefore, traditional bid rigging and bid collusion risk assessment methods have the problems of low identification accuracy and low efficiency. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method and system for assessing the risk of bid rigging and collusion based on knowledge graphs and large models, so as to improve the accuracy and efficiency of bid rigging and collusion risk assessment.

[0008] The technical solution adopted by the present invention to solve the above technical problems is: In one aspect, the present invention provides a method for assessing bid rigging risk based on a knowledge graph and a large model, comprising the following steps: S1. Obtain relevant data on multiple companies, including but not limited to: basic company information, shareholder structure information, executive information, historical bidding records, historical negative records of the company, and historical negative records of the company's executives; S2. Based on the acquired enterprise-related data, use graph construction technology to build enterprise association graphs and executive association graphs; S3. Use OCR technology to identify multiple bidding documents and format and structure the recognition results; S4. Use data comparison technology to conduct a multi-dimensional comparison of the OCR recognition results of the multiple bidding documents, including but not limited to: format consistency, font consistency, quotation method and price similarity, and common defects; S5. Based on the comparison results of the multiple bidding documents and the enterprise and executive association maps, a comprehensive analysis is performed using a pre-trained large model to assess the risk of bid rigging and collusion, and a corresponding risk score is generated; S6. Automatically generate a bid rigging and collusion risk assessment report based on the generated risk score.

[0009] Furthermore, in step S1, multiple enterprise-related data are obtained from the enterprise information platform through the API interface or authorized crawler. The multiple enterprises refer to different enterprises participating in the same bidding project, as well as other enterprises that have an affiliated relationship with these enterprises; in step S3, the multiple bidding documents refer to the bidding documents submitted by different bidding enterprises in the same bidding project.

[0010] Furthermore, step S1 also includes: pre-processing the acquired raw data, including but not limited to: deduplication processing, missing data completion and format unification processing.

[0011] Furthermore, in step S2, based on the acquired enterprise-related data, the enterprise association map and the executive association map are established using the map construction technology, including: Automatically identify and extract related companies based on shareholder structure and historical corporate cooperation data, establish nodes and edges through relationships such as shareholders, cooperative projects, and joint investments, form a relationship network between companies, and obtain a corporate association map; By analyzing the historical job information of corporate executives and their roles in multiple companies, we build a network of relationships between executives and obtain an executive relationship map.

[0012] Furthermore, in step S6, the bid-rigging risk assessment report includes: enterprise association analysis, bidding document similarity analysis, risk assessment results and recommended measures, and is output in the form of a system interface or PDF document.

[0013] On the other hand, the present invention also provides a bid-rigging risk assessment system based on knowledge graph and large model, including: The data collection module is used to obtain data related to multiple companies, including but not limited to: basic company information, shareholder structure information, executive information, historical bidding records, historical bad records of the company and historical bad records of the company's executives; The knowledge graph construction module is used to build enterprise association graphs and executive association graphs based on the acquired enterprise-related data and using graph construction technology; OCR recognition module, used to recognize multiple bidding documents using OCR technology, and format and structure the recognition results; A document comparison module is used to perform a multi-dimensional comparison of the OCR recognition results of the plurality of bidding documents using data comparison technology, with the comparison contents including but not limited to: format consistency, font consistency, quotation method and price similarity, and common defects; The risk assessment module is used to conduct a comprehensive analysis based on the comparison results of the multiple bidding documents and the enterprise association map and executive association map using a pre-trained large model to assess the risk level of bid rigging and collusion, and generate a corresponding risk score. Based on the generated risk score, a bid rigging and collusion risk assessment report is automatically generated.

[0014] Furthermore, the system also includes: The data preprocessing module is used to preprocess the acquired raw data, including but not limited to: deduplication, missing data completion and format unification.

[0015] Furthermore, the system also includes: The optimization and feedback module is used to regularly retrain the large model based on new data and manual feedback.

[0016] Furthermore, the data collection module obtains multiple enterprise-related data from the enterprise information platform through an API interface or an authorized crawler.

[0017] Furthermore, the knowledge graph construction module uses graph construction technology based on the acquired enterprise-related data to establish an enterprise association graph and an executive association graph, including: Automatically identify and extract related companies based on shareholder structure and historical corporate cooperation data, establish nodes and edges through relationships such as shareholders, cooperative projects, and joint investments, form a relationship network between companies, and obtain a corporate association map; By analyzing the historical job information of corporate executives and their roles in multiple companies, we build a network of relationships between executives and obtain an executive relationship map.

[0018] The beneficial effects of the present invention are: (1) Improve the accuracy of identifying bid-rigging and collusion: By combining multi-dimensional data fusion with large-scale model analysis, this invention comprehensively considers multiple factors, including corporate connections, executive interactions, and similarities in the content and format of bidding documents, significantly improving the accuracy of identifying bid-rigging. Existing technologies often rely on single indicators (such as document similarity or corporate relationships) to assess risk. By integrating multi-dimensional data, this invention avoids the limitations of single analysis methods and ensures the accuracy of identification results.

[0019] (2) Improving the efficiency of risk assessment: This invention significantly reduces manual review workloads and improves the efficiency of bid-rigging risk assessments through automated data collection, OCR recognition, and document comparison. By automatically analyzing multidimensional data using a large model, the system can quickly process large volumes of bidding documents and enterprise data, significantly improving the responsiveness and real-time nature of risk assessments.

[0020] (3) Realize comprehensive multi-dimensional risk assessment: This method not only relies on bid document similarity but also incorporates multiple data dimensions, such as shareholder relationships, collaboration history, document formats, and executive connections, to achieve a comprehensive assessment of bid rigging and collusion. This multi-dimensional analysis approach makes the assessment more comprehensive and accurate, better identifying potential bid rigging and collusion, and reducing the risk of misjudgment and omission.

[0021] (4) Improving the transparency and fairness of the bidding process: Through an intelligent risk assessment mechanism, this invention provides greater transparency and fairness in the bidding process. Companies and regulatory authorities can access detailed risk reports in real time, ensuring a more open and fair bidding process. This helps prevent malicious collusion and unfair market competition, and promotes standardized development of the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the bid rigging and collusion risk assessment method based on knowledge graph and large model in the present invention.

[0023] Figure 2 This is a structural diagram of the bid-rigging and collusion risk assessment system based on knowledge graph and large model in the present invention. DETAILED DESCRIPTION

[0024] The present invention aims to provide a bid-rigging and collusion risk assessment method and system based on knowledge graphs and large models, so as to improve the accuracy and efficiency of bid-rigging and collusion risk assessment. The core idea is to obtain information such as corporate shareholders, executives, and historical transactions from the enterprise big data platform through automated data collection, and to construct a network of associations between enterprises and a network of relationships between executives based on graph construction technology, which can intuitively present the potential subjects and behavior patterns of bid-rigging and collusion. In addition, by identifying and comparing bidding documents based on OCR technology, similarity features in the documents can be discovered, thereby further supporting the possibility of bid-rigging and collusion. Finally, a large model is used in combination with the above-mentioned multi-source data for comprehensive analysis and evaluation, thereby realizing intelligent identification and risk assessment of bid-rigging and collusion behaviors.

[0025] In terms of specific implementation, the process of the bid-rigging risk assessment method based on knowledge graph and large model provided by the present invention can be found in Figure 1 , which includes the following implementation steps: 1. Obtain enterprise data: In this step, data such as basic company information, historical bidding records, and negative records of the company and its executives are collected from various data sources. In one exemplary embodiment, data such as company registration information, shareholder structure, historical transaction records, historical negative records, executive information, and bidding records can be obtained from enterprise information databases such as Qichacha and Tianyancha through API interfaces or authorized crawlers. The collected raw data is then deduplicated, missing values ​​are supplemented, and formatted uniformly to ensure data quality and facilitate subsequent analysis. Finally, the cleaned data is stored in a structured database to support the subsequent construction of the knowledge graph.

[0026] 2. Knowledge graph construction: In this step, a relationship map between enterprises and a relationship map of executives are constructed based on the collected data to analyze potential risks of bid rigging and collusion. In an exemplary embodiment, equity penetration analysis technology can be used to extract shareholder relationships, cooperation records, etc. between enterprises from the data to construct a relationship network between enterprises. At the same time, a relationship network between executives is constructed by analyzing the identity information and job changes of executives. Based on the relationship network between enterprises and the relationship network between executives, a multi-dimensional map is constructed to obtain an enterprise relationship map and an executive relationship map, which supports the display of information such as the degree of connection and cooperation history between different enterprises and different executives. These map data are stored and visualized through a graph database (such as Neo4j). In addition, the relationship map can also be used to explore potential risk points of bid rigging and collusion, such as affiliated companies frequently winning bids together, and the same executive managing multiple companies.

[0027] 3. Obtain bidding document data: In this step, the specific bidding document data is obtained from the bidding platform or relevant channels, including bidding documents, technical solutions, quotation lists, etc.

[0028] 4.OCR recognition: In this step, OCR recognition is performed on the bidding documents to extract the text content such as technical solutions and quotation lists in the documents, and the file format information is retained.

[0029] 5. File comparison: In this step, the bidding documents identified by OCR are compared to check for similarities in file format, font, quotation, errors, etc., to help identify duplicates. In an exemplary embodiment, format consistency analysis can be used to check whether the layout, tables, paragraphs and other structures of the documents are highly consistent; font consistency analysis can be used to detect whether the same font type, size, typesetting, etc. are used; by comparing the quotation lists in different documents, it can be determined whether there are similar or identical prices; by checking the documents for typos, repeated paragraphs or semantically similar parts, it can be determined whether there are common defects. Finally, the text similarity between the bidding documents will be output to identify potential plagiarism or template-based behavior.

[0030] 6. Large model risk assessment: In this step, a large model analyzes the associations between companies and documents based on the document comparison results and knowledge graph data, calculating a bid-rigging risk score. In one exemplary implementation, multidimensional data (such as company associations, document similarities, and historical behavior) extracted from the knowledge graph and OCR comparison is fed into the large model, which uses deep learning techniques to analyze this multidimensional data and identify potential bid-rigging risks. For example, the model can predict abnormal patterns in current behavior by analyzing the characteristics of historical bid-rigging behavior.

[0031] 7. Output report: In this step, a bid-rigging risk assessment report is generated, providing a risk score, risk analysis, and recommended measures. In one exemplary implementation, a detailed risk report is generated based on the large-scale model assessment results. The report includes links between companies, document similarity analysis, and potential risk points, and is presented graphically to facilitate understanding and decision-making by decision-makers.

[0032] The structure of the bid-rigging risk assessment system based on knowledge graph and large model provided by the present invention can be found in Figure 2 , which includes the following functional modules: (1) Data acquisition module: This module is used to obtain data such as basic enterprise information, historical bidding records, and bad records of enterprises and executives from different data sources. Among them, basic enterprise information includes the enterprise's registration information, industry classification, shareholder structure, personal information of enterprise executives, positions, historical change records, etc.; historical bidding records include participated projects, winning bid records, etc.; enterprise bad records include violations, complaints, penalties, etc.; related bad records of executives include participation in illegal projects and misconduct of executives.

[0033] (2) Knowledge graph construction module: This module is used to construct enterprise and executive relationship maps based on data collection results. The enterprise relationship map analyzes equity, partnership, and competitive relationships among enterprises using data such as shareholder structure and historical bidding records, thereby building a network of these enterprises. The executive relationship map establishes relationships between executives based on job changes and their inter-company tenure. These relationship maps provide data support for subsequent bid-rigging risk analysis.

[0034] (3) OCR recognition module: This module uses OCR technology to perform text recognition on bidding documents, extracting key information, including technical proposal descriptions, quotation lists, qualification certificates, and other important content. While extracting text content, it also preserves the document's formatting (such as tables and layout), providing raw data for subsequent comparison and analysis.

[0035] (4) File comparison module: This module compares multiple bidding documents to analyze whether there is potential bid rigging or collusion.

[0036] The dimensions of comparison include: Same format: Check whether the layout, tables, paragraphs and other structures of the document are highly consistent.

[0037] Font consistency: Check whether the same font type, size, layout, etc. are used.

[0038] Similar quotation methods and prices: Compare the quotation lists in different documents to determine whether the prices are close or the same.

[0039] Common defects: Check the documents for typos, repeated paragraphs, or semantically similar sections.

[0040] (5) Risk Assessment Module: This module uses a large model to conduct comprehensive analysis of multi-dimensional data: Correlation analysis: Based on the correlation data between companies and executives, determine whether there are affiliated companies or affiliated executives involved in the same project, which may pose a risk of bid rigging or collusion.

[0041] Document similarity analysis: By comparing the results, we can detect whether there are multiple bidding documents with similar content, format or quotation, so as to identify potential bid rigging and collusion.

[0042] Historical behavior analysis: Combined with the company's and executives' historical bad records and historical bidding records, analyze whether there are any abnormal patterns in their bidding behavior.

[0043] Based on the above analysis, the large-scale model calculates a bid-collusion risk score and outputs a bid-collusion risk assessment report. The report includes the following: Enterprise Association Analysis Results: assessing whether there are any unusual associations between enterprises; Bid Document Similarity Analysis Results: assessing the similarity between bidding documents; Risk Assessment Results: Based on the comprehensive assessment, a bid-collusion risk score is assigned and risk classification is performed based on the score.

[0044] Example: This embodiment will elaborate on the implementation process of the bid rigging and collusion risk assessment method based on knowledge graph and big model.

[0045] S1. Data collection and preprocessing: In this embodiment, the following data related to multiple enterprises is automatically obtained from enterprise information platforms such as Qichacha and Tianyancha through API interfaces or authorized crawlers: 1) Basic information of the enterprise: including enterprise name, registered capital, establishment time, business scope, etc.

[0046] 2) Shareholder structure information: including shareholder name, equity ratio, shareholder relationship, etc.

[0047] 3) Executive information: including executive name, position, past employment record, affiliated companies, etc.

[0048] 4) Historical information: including the company's historical bidding records, the company's historical bad records, and the executives' historical bad records, etc.

[0049] 5) Bidding records: including bidding project name, bidding documents, bidding documents, bidding results, project amount, etc.

[0050] The above data sources are synchronized through regular updates and real-time acquisition, and stored in the local database for subsequent processing.

[0051] Next, the acquired raw data undergoes the following preprocessing steps: 1) Deduplication: Remove duplicate records to ensure data uniqueness.

[0052] 2) Completing missing data: Supplement missing corporate information (such as executive information) and use public data sources (such as business registration information) for automatic filling.

[0053] 3) Format unification: Unify the formats of fields such as date and amount to ensure data consistency and availability.

[0054] The processed data will be stored in a structured database such as MySQL or PostgreSQL for subsequent analysis.

[0055] It is understandable that the “multiple enterprises” here refer to different enterprises participating in the same bidding project, as well as other enterprises that have affiliated relationships with these enterprises (such as shareholders, senior executives, cooperation history, etc.).

[0056] S2. Knowledge graph construction and relationship analysis: This embodiment uses a graph database (such as Neo4j) to build a knowledge graph.

[0057] Enterprise relationship map: Based on the shareholder structure and historical cooperation data of enterprises, related enterprises are automatically identified and extracted, and nodes and edges are established through relationships such as shareholders, cooperative projects, and joint investments to form a relationship network between enterprises.

[0058] Executive Relationship Map: Analyze the historical job information and roles of corporate executives in multiple companies, build a relationship network between executives, and identify multiple companies managed by the same executive.

[0059] This map shows direct and indirect connections between companies and executives, allowing the system to identify potential risk points. For example, if two companies have common shareholders or executives, there may be a risk of bid rigging.

[0060] The constructed knowledge graph can be visualized using graphical tools (such as Gephi and Cytoscape). Users can use the visualized graph to examine the degree of connections and risk points between companies and executives. The system also supports dynamic analysis of the graph, such as querying information on all related companies and executives of a specific company.

[0061] S3.OCR technology and bidding document comparison: Bidding documents are usually in the form of PDFs or scanned images. Therefore, OCR technology is used to perform text recognition and format extraction on multiple bidding documents. This example uses open source OCR tools (such as Paddlepaddle) or commercial OCR tools (such as ABBYY FineReader) to process the files and extract the following content: 1) Technical solution description, quotation list, qualification certificate, etc. in the bidding documents.

[0062] 2) Structured data such as tables and graphics that appear in the document.

[0063] The recognized text will be formatted and structured to ensure the accuracy and usability of the text content during subsequent analysis.

[0064] Next, multiple bid documents are compared and their textual similarity is analyzed. Natural language processing (NLP) techniques, such as word embedding (Word2Vec, BERT) models, are used to calculate the semantic similarity between documents.

[0065] The system pays special attention to the following points for judgment: 1) Same format: Identify whether the layout structure, tables, paragraph layout and other formats in the file are highly consistent; 2) Font consistency: Compare the font type, size, and layout format of the files to determine if there are any similarities; 3) Similarity in quotation method and price: Compare the quotation list in the bidding documents to determine whether the quotation method is consistent and whether there are any abnormal situations with similar prices; 4) Identical defects: Check whether there are obvious identical defects in the bidding documents, such as identical typos, similar text paragraphs, etc., especially the frequently occurring “template” errors.

[0066] Based on the OCR recognition, the system conducts a comprehensive comparison of the bid documents' format, content, and other textual details. If the system detects duplicates or unusual similarities across multiple bid documents in these dimensions, it will flag them as potential bid-rigging risks, further supporting risk assessment.

[0067] It can be understood that the "multiple bidding documents" here refer to the bidding documents submitted by different bidding companies in the same bidding project.

[0068] S4. Comprehensive analysis and risk assessment of large models: The constructed knowledge graph data, OCR comparison results, and historical behavior data are input into the large model for comprehensive analysis. The data set includes: 1) The strength of linkages between enterprises and the risk paths.

[0069] 2) Similarity score of bidding documents.

[0070] 3) Historical behavioral characteristics of enterprises and executives.

[0071] This data will be converted into feature vectors and fed into a pre-trained large model.

[0072] This example uses large deep learning-based models (such as Transformer, BERT, GPT4, Tongyi Qianwen, and ChatGLM) to analyze multi-dimensional input data and generate a risk score for bid-rigging. The model learns behavioral patterns based on the characteristics of historical bid-rigging cases, enabling it to accurately identify potential risks.

[0073] 1) Risk score: Generate a bid-rigging risk score for each enterprise and bidding document. The higher the risk score, the more likely the behavior is bid-rigging or collusion.

[0074] 2) Abnormal behavior identification: The model can identify possible abnormal behavior patterns, such as two companies frequently winning bids in multiple projects or highly similar document contents.

[0075] S5. Risk Assessment Report: Based on the analysis results of the large model, the system automatically generates a detailed bid-rigging risk report, which includes: The connection paths and risk levels between enterprises.

[0076] Results of similarity analysis and format consistency analysis of bidding documents.

[0077] Comprehensive risk score and recommended actions (such as enhanced review, further investigation, etc.).

[0078] Reports can be provided to users through the system interface or automatically generated PDF documents.

[0079] S6. System optimization and feedback mechanism: 1) Data feedback and model update: Based on manual review or survey results, new data feedback is collected to update the knowledge graph and model. The system can continuously improve the database based on this feedback data and retrain the large model to improve its accuracy.

[0080] 2) Model optimization and iteration: At regular intervals, the system will retrain the model based on new data and feedback data to continuously improve the accuracy and efficiency of identifying bid rigging and collusion.

[0081] Although the embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, all without departing from the scope of protection of the present invention.

Claims

1. A bid-rigging risk assessment method based on knowledge graph and big model, characterized by: The following steps are involved: S1. Obtain relevant data on multiple companies, including basic company information, shareholder structure information, executive information, historical bidding records, historical bad records of the company, and historical bad records of the company's executives; S2. Based on the acquired enterprise-related data, use graph construction technology to build enterprise association graphs and executive association graphs; S3. Use OCR technology to identify multiple bidding documents and format and structure the recognition results; S4. Use data comparison technology to perform a multi-dimensional comparison of the OCR recognition results of the multiple bidding documents, including: format consistency, font consistency, quotation method and price similarity, and common defects; S5. Based on the comparison results of the multiple bidding documents and the enterprise and executive association maps, a comprehensive analysis is performed using a pre-trained large model to assess the risk of bid rigging and collusion, and a corresponding risk score is generated; S6. Automatically generate a bid rigging and collusion risk assessment report based on the generated risk score.

2. The bid-rigging risk assessment method based on knowledge graph and large model as claimed in claim 1 is characterized in that: In step S1, multiple enterprise-related data are obtained from the enterprise information platform through the API interface or authorized crawler. The multiple enterprises refer to different enterprises participating in the same bidding project, as well as other enterprises that have an affiliated relationship with these enterprises; in step S3, the multiple bidding documents refer to the bidding documents submitted by different bidding enterprises in the same bidding project.

3. The bid-rigging risk assessment method based on knowledge graph and large model as claimed in claim 1 is characterized in that: Step S1 also includes: pre-processing the acquired raw data, including: de-duplication processing, missing data completion and format unification processing.

4. The bid-rigging risk assessment method based on knowledge graph and large model as claimed in claim 1 is characterized in that: In step S2, based on the acquired enterprise-related data, the enterprise association map and the executive association map are established using the map construction technology, including: Automatically identify and extract related companies based on shareholder structure and historical corporate cooperation data, establish nodes and edges through relationships such as shareholders, cooperative projects, and joint investments, form a relationship network between companies, and obtain a corporate association map; By analyzing the historical job information of corporate executives and their roles in multiple companies, we build a network of relationships between executives and obtain an executive relationship map.

5. The bid-rigging risk assessment method based on knowledge graph and large model as claimed in claim 1 is characterized in that: In step S6, the bid-rigging risk assessment report includes: enterprise association analysis, bidding document similarity analysis, risk assessment results and recommended measures, and is output in the form of a system interface or PDF document.

6. The bid-rigging risk assessment system based on knowledge graph and big model is characterized by: include: The data collection module is used to obtain relevant data of multiple companies, including basic company information, shareholder structure information, executive information, historical bidding records, historical bad records of the company and the historical bad records of the company's executives; The knowledge graph construction module is used to build enterprise association graphs and executive association graphs based on the acquired enterprise-related data and using graph construction technology; OCR recognition module, used to recognize multiple bidding documents using OCR technology, and format and structure the recognition results; A document comparison module is used to use data comparison technology to perform multi-dimensional comparison of the OCR recognition results of the plurality of bidding documents, including format consistency, font consistency, quotation method and price similarity, and common defects; The risk assessment module is used to conduct a comprehensive analysis based on the comparison results of the multiple bidding documents and the enterprise association map and executive association map using a pre-trained large model to assess the risk level of bid rigging and collusion, and generate a corresponding risk score. Based on the generated risk score, a bid rigging and collusion risk assessment report is automatically generated.

7. The bid-rigging risk assessment system based on knowledge graph and large model as claimed in claim 6 is characterized in that: The system also includes: The data preprocessing module is used to preprocess the acquired raw data, including deduplication, missing data completion and format unification.

8. The bid-rigging risk assessment system based on knowledge graph and large model as claimed in claim 6 is characterized in that: The system also includes: The optimization and feedback module is used to regularly retrain the large model based on new data and manual feedback.

9. The bid-rigging risk assessment system based on knowledge graph and large model according to claim 6 is characterized in that: The data acquisition module obtains multiple enterprise-related data from the enterprise information platform through an API interface or an authorized crawler.

10. The bid-rigging risk assessment system based on knowledge graph and large model according to claim 6 is characterized in that: The knowledge graph construction module uses graph construction technology based on the acquired enterprise-related data to build enterprise association graphs and executive association graphs, including: Automatically identify and extract related companies based on shareholder structure and historical corporate cooperation data, establish nodes and edges through relationships such as shareholders, cooperative projects, and joint investments, form a relationship network between companies, and obtain a corporate association map; By analyzing the historical job information of corporate executives and their roles in multiple companies, we build a network of relationships between executives and obtain an executive relationship map.

Images