An artificial intelligence-based tenderer subject responsibility whole-process compliance management method and system

Abstract

The application relates to the technical field of artificial intelligence, knowledge graph, blockchain and electronic bidding compliance control, and discloses a bidding subject responsibility whole-process compliance control method and system based on artificial intelligence, S1, three data sources of a legal regulation database, a historical bidding case database and enterprise internal compliance system documents are integrated, three types of core entities of responsibility subject entities, behavior entities and consequence entities are extracted; according to national and local bidding regulations and case cause-effect logic, the scheme realizes intelligent examination of the whole bidding process, solves the problems of low artificial examination efficiency and different standards, and can accurately identify various implicit irregular behaviors. The system monitors and classifies the intervention risk in real time, changes post-audit into pre-audit and mid-audit control, effectively reduces the risk of complaints and bid invalidation. Relying on the knowledge graph and the blockchain technology, responsibility is clarified, and tamper-proof evidence is retained, meanwhile, the model can be iteratively updated, regulations can be adapted, and long-term operation of compliance management can be ensured.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, knowledge graphs, blockchain, and electronic bidding compliance management technology, specifically to a method and system for full-process compliance management of the main responsibility of the bidding entity based on artificial intelligence. Background Technology

[0002] Tendering and procurement are important methods for public resource transactions and corporate procurement. As the organizer and responsible party of procurement activities, the tendering party bears legal responsibility for the compliance of the entire tendering process. However, in practice, the problem of inadequate implementation of the tendering party's responsibility has long existed: tender documents often contain exclusive clauses such as specifying brands or setting unreasonable thresholds; abnormal scoring and deviations in the behavior of evaluation experts occur during the bid opening and evaluation stages; and collusion may occur between different bidders. These problems lead to high rates of tendering complaints, high rates of rejected bids, and difficulties in auditing and holding those responsible accountable.

[0003] Currently, existing bidding compliance management methods mainly rely on manual review and post-event auditing. Manual review is inefficient, lacks standardized criteria, and struggles to cover multiple stages of the entire bidding process; post-event auditing can only identify problems and cannot intervene in real time when violations occur. Although some electronic bidding platforms integrate rule engines for keyword matching or process node recording, their detection capabilities are limited and cannot identify implicit exclusive clauses, complex relationships, and dynamic abnormal scoring patterns. Furthermore, existing systems lack accurate association of the bidding entity's main responsibility and an immutable evidence storage mechanism throughout the entire process, making it difficult to trace responsibility. Therefore, we propose an AI-based method and system for full-process compliance management of bidding entity responsibility to address the aforementioned problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence. It solves many problems of traditional tendering, such as low efficiency due to reliance on manual review, weak ability to identify hidden violations, only able to conduct post-event audits, difficulty in tracing responsibility, and inability of the system to adapt to new regulations.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence, comprising the following steps:

[0006] S1. Integrate three major data sources: legal and regulatory databases, historical bidding case databases, and internal corporate compliance system documents. Extract three core entities: responsible entities, behavioral entities, and consequence entities. Based on national and local bidding regulations and case causal logic, establish edge connections between multiple responsibility nodes, compliance rule nodes, and historical risk event nodes to form a basic graph structure. Further employ graph neural networks to learn the implicit compliance dependency paths between entities, generating a dynamic responsibility transmission chain. Simultaneously, assign weight coefficients and timeliness labels to each responsibility node to differentiate the severity of responsibility and the validity period of applicable regulations, completing the construction and dynamic updating of the knowledge graph of the bidding entity's main responsibility.

[0007] S2. Upon receiving the bidding initiation command, the system automatically connects to business ports such as the electronic bidding platform, contract management system, and on-site bid evaluation terminal to collect multi-source heterogeneous data throughout the entire bidding process. The entire bidding process covers seven core stages: project initiation, document preparation, announcement publication, bid receipt, bid opening and evaluation, contract awarding and signing, and contract performance. The collected multi-source data includes structured data (basic project information, bidder information, expert information, process node data, etc.), semi-structured data (forms, lists, ledgers, etc.), and unstructured text data (bidding documents, bid documents, bid evaluation reports, announcement texts, etc.).

[0008] S3. Utilize natural language processing models to perform entity extraction and intent recognition on the collected unstructured and semi-structured text data, analyze the operational behaviors and business actions of each stage of the bidding process, and generate standardized process behavior sequences. Map each generated process behavior sequence to the knowledge graph of the bidding entity's main responsibility constructed in step S1. Combine the rule relationships within the graph to match and determine the responsible entity and compliance verification point corresponding to each process behavior, thus defining the detection scope and standards for subsequent AI compliance detection.

[0009] S4. Real-time AI Compliance Analysis Through Multi-Model Collaboration: Based on the identified compliance verification points, the corresponding AI compliance detection model group is invoked to perform real-time compliance analysis on the current process behavior, outputting a compliance score and risk label. The AI ​​compliance detection model comprises four core models, with each model having the following functions:

[0010] Employing a fine-grained classifier based on a pre-trained language model, this system specifically identifies exclusive violation clauses in tender documents that specify a particular brand, region, company size, or ownership type, and simultaneously outputs the exclusive risk level and corresponding text modification suggestions.

[0011] Extract multi-dimensional information such as bidder device fingerprints, network addresses, file metadata, business registration and equity, and related personnel to build a bidder similarity network and calculate the comprehensive correlation score between each pair of bidders. When the correlation score exceeds the preset collusion suspicion threshold, a collusion risk label is automatically generated and the bidder involved is marked.

[0012] Based on a temporal attention network, three types of abnormal scoring patterns are identified by comparing the historical scoring sequences of evaluation experts with the current project scoring sequence:

[0013] Deviation from the center pattern, meaning that the score of a single expert deviates from the average score of the entire expert group by a multiple of the preset standard deviation.

[0014] Fixed preference pattern, which means that experts consistently give abnormally high or low scores to the same type of bidders in multiple projects;

[0015] The time-related pattern refers to an unreasonable strong correlation between the scoring order and the order of the bidders' presentations.

[0016] Verify the time nodes and process sequence of each stage of the bidding process, and identify logical violations such as process reversal, node timeout, and time sequence disorder.

[0017] S5. A compliance scoring threshold and a high-risk label list are pre-set. When the compliance score output by AI detection falls below the preset threshold, or when a behavior matches a preset high-risk label, the system automatically triggers tiered intervention actions. Tiered interventions are divided into four categories from low to high: alert warning, forced modification, process blocking, and responsibility log solidification. The specific process for solidifying the responsibility log is as follows: record a snapshot of the process context at the time of intervention triggering, the basis for AI detection decisions, and the identity information of the on-site operators; synchronize the complete log to the tenderer's main responsibility traceability platform, and solidify electronic responsibility certificates based on digital signatures and timestamps, forming an undeniable record of responsibility.

[0018] S6. Core data such as full-process behavioral data, compliance scoring results, risk tags, tiered intervention records, and fixed responsibility logs are written into the compliance audit blockchain using hash encryption to ensure data immutability and full traceability. Simultaneously, a unique compliance traceability code is generated for each responsible entity and each bidding project, allowing for rapid retrieval of full-process audit data.

[0019] This invention provides an artificial intelligence-based full-process compliance management system for the main responsibility of tendering entities, used to execute the method described, characterized in that it includes:

[0020] The data acquisition module establishes data interfaces with the tenderer's electronic tendering platform, contract management system, bid evaluation site terminal, and internal office system to achieve automated and real-time acquisition of multi-source heterogeneous data in the seven stages of tendering. It supports unified access and preliminary cleaning of structured, semi-structured, and unstructured data, providing raw data support for subsequent analysis.

[0021] The knowledge graph module is deployed based on a graph database and is responsible for storing and updating the knowledge graph of the main responsibilities of the tendering party. The module has a built-in responsibility node retrieval interface and risk path reasoning engine, which can quickly complete process behavior matching, responsibility entity positioning, risk transmission path deduction, and support dynamic iteration of graph nodes and relationships based on new regulations and new cases.

[0022] The AI ​​detection engine module is the core intelligent detection unit of this system, comprising four independent sub-modules, each corresponding to one of the four AI compliance detection models in the methodology. Each sub-module uses a deep learning model or ensemble learning model to independently compute and collaboratively output results. This module also incorporates an active learning feedback loop: it collects data on the adoption / rejection of risk warnings and modification suggestions by staff, marking adopted and rectified samples as positive samples, and rejecting samples that lead to complaints, bid rejections, or audit issues as negative samples. It periodically uses positive and negative samples to incrementally train and fine-tune the AI ​​model, ensuring the model continuously adapts to the latest laws and regulations, local bidding rules, and new types of violations.

[0023] The tiered intervention module has built-in three-level scoring thresholds (first threshold, second threshold, and third threshold, with the first threshold > second threshold > third threshold) and a full list of risk label tiers. It executes tiered intervention strategies based on the compliance score and risk labels output by the AI ​​detection engine.

[0024] Level 1 intervention: If the compliance score is between the second and first thresholds, or if a general risk label is hit, the system will pop up a warning and push modification suggestions without disrupting existing business processes.

[0025] Level 2 intervention: If the compliance score is between the third and second thresholds, or if a moderate violation label is hit, the current operation interface will be locked. The operator must complete the rectification and risk downgrade before the process can continue.

[0026] Level 3 intervention: If the compliance score is lower than or equal to the third threshold, or if serious violations such as bid rigging or fraudulent bidding are detected, the system will automatically terminate the current bidding process, send a process blocking report to the legal representative and compliance management personnel of the bidding entity, and automatically solidify all evidence in the entire chain.

[0027] The blockchain evidence storage module connects to industry consortium blockchain nodes to perform hash-based on-chain evidence storage of key bidding process data, AI detection results, risk warning records, intervention operation logs, and responsibility certificates. Leveraging the decentralized, tamper-proof, and traceable characteristics of blockchain, it ensures the authenticity and validity of all compliant data, while forming a chain-like cross-verification with the digital signatures and timestamps in the responsibility logs.

[0028] The compliance traceability interface module provides a visual human-computer interaction interface for tenderers, regulators, and auditors. It supports multi-dimensional searches by responsible entity, project number, and time interval, intuitively displaying the entire project compliance audit path and electronic responsibility certificates. The module includes a sub-module for tracking changes in responsible entities: when the same responsible entity triggers a preset number of Level 3 intervention events in different projects, it automatically extracts quantitative scores for each responsibility dimension and generates a compliance risk profile for the entity through a radar chart rendering engine. The risk profile covers four responsibility dimensions: document preparation, bid evaluation fairness, information confidentiality, and contract performance, intuitively displaying the risk distribution and supporting one-click export of standardized reports, which can be directly submitted to internal supervisory bodies or industry administrative supervision departments.

[0029] Beneficial effects

[0030] This invention provides a method and system for full-process compliance management of the main responsibilities of tendering entities based on artificial intelligence. Compared with existing technologies, it has the following advantages:

[0031] This AI-based method and system for full-process compliance management of tendering entities' responsibilities utilizes artificial intelligence technology to automate the entire tendering process, effectively addressing the problems of low efficiency, inconsistent standards, and incomplete regulatory coverage associated with traditional manual review. The system employs multiple AI detection models to accurately identify various violations, such as hidden exclusive clauses, bidder collusion, abnormal evaluation scores, and process sequence conflicts, thus overcoming the shortcomings of traditional rule engines and significantly improving the comprehensiveness and accuracy of violation identification.

[0032] This has changed the previous post-audit control model, and instead conducts real-time monitoring during business operations. It also implements graded intervention measures according to risk levels, and promptly stops violations through early warning reminders, mandatory rectification, and process blocking. This reduces bid rejections, complaints, and economic losses from the source, and shifts compliance control from post-event accountability to pre-event prevention and in-event control.

[0033] By leveraging knowledge graphs to accurately link process behaviors with responsible personnel, the system clarifies accountability. Combining blockchain, digital signatures, and timestamps to solidify data, and relying on compliance traceability codes to form a complete and verifiable chain of evidence, the system solves the problems of difficult accountability and easily tampered data. Furthermore, the system possesses self-learning and iteration capabilities, adapting to updates in regulations and local rules, ensuring long-term stability and effective implementation of bidding compliance management. Attached Figure Description

[0034] Figure 1 This is a flowchart of a method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence, according to the present invention.

[0035] Figure 2 This is a block diagram of a full-process compliance management system for the main responsibility of the tendering party based on artificial intelligence, according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] like Figure 1-2 As shown:

[0038] Example

[0039] A method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence, specifically including the following steps:

[0040] S1. First, build a basic data source pool and connect to the bidding and tendering laws and regulations database (including the National Bidding and Tendering Law, Government Procurement Law and supporting regulations), the historical bidding case database (industry violation cases, administrative penalty cases, and complaint cases in the past 3-5 years), and the enterprise internal compliance system database (internal bidding management methods, job responsibilities, and accountability systems of the bidding entity).

[0041] An entity extraction algorithm is used to extract responsible entities (tenderers, project managers, document drafters, bid evaluation organizers, contract managers, etc.), behavioral entities (document drafting, announcement publication, scoring, signing, etc.), and consequence entities (bid rejection, complaints, administrative penalties, internal accountability, etc.) from three types of documents.

[0042] Compliance rule nodes are built based on legal provisions, and risk event nodes are built based on case causal relationships. The three types of entity nodes, rule nodes, and risk nodes are connected by directed edges to form a basic graph.

[0043] A Graph Neural Network (GNN) is introduced to perform deep learning on the implicit compliance dependencies and responsibility transmission relationships between entities, generating a dynamic responsibility transmission chain. Each responsibility node is assigned a weight coefficient ranging from 0 to 1 based on the importance of regulations and the responsibilities of different positions. Timeliness tags are also added, indicating the effective date and repeal date of the regulations, enabling dynamic updates to the graph. The graph is ultimately stored in a graph database, connecting to the retrieval interface and inference engine of the knowledge graph module.

[0044] S2. Upon receiving the bidding start instruction initiated by the project establishment, the data acquisition module immediately starts the data acquisition task and connects to various business terminals in stages.

[0045] Data collection includes project initiation, document preparation, collection of project approval documents, tender document texts, and information on the personnel responsible for preparation;

[0046] The process includes: announcement publication, bid acceptance, collection of bidding announcements, registration information, bid documents, bidder basic information, network access logs, and file upload metadata;

[0047] Bidding and evaluation stage: Collecting bidding records, evaluation expert information, scoring record sheets, on-site operation logs, equipment network data, etc.

[0048] The data collection process includes contract signing, contract performance stages, collection of winning bid announcements, winning bid notices, contract texts, and performance progress ledgers. During the data collection process, the system automatically distinguishes data types; structured data is directly stored in the database, while semi-structured and unstructured text data are uniformly transferred to the natural language processing parsing unit for initial data classification and cleaning.

[0049] S3, the natural language processing model, uses a pre-trained large language model as its foundation to complete two core tasks:

[0050] Entity extraction: Accurately extract key entities such as project name, personnel, company, behavior, time, and location from massive amounts of text;

[0051] Intent recognition: Determines whether the current operation belongs to a specific business action such as document preparation, announcement publication, or bid evaluation and scoring. Based on the analysis results, a standardized process behavior sequence is generated, tracing the entire project operation trajectory in chronological order. Each process behavior sequence is input into the knowledge graph module, and compliance rule nodes within the graph are matched through a retrieval interface. This automatically binds the corresponding responsible personnel, positions, and compliance verification points, clarifying the detection content and judgment criteria for subsequent AI detection.

[0052] S4, the AI ​​detection engine module schedules the corresponding sub-models to perform parallel detection based on the matched compliance verification points:

[0053] The clause exclusivity detection submodule employs a fine-grained text classification model to perform semantic analysis on each paragraph and clause of the tender document. It not only identifies explicit keywords such as "designated brand," but also uses semantic understanding to identify implicit exclusivity clauses that disguise regional or corporate qualification barriers. The model outputs three levels of exclusivity risk: high, medium, and low, and generates directly modifiable text suggestions based on the relevant legal provisions.

[0054] The relationship graph detection submodule automatically captures the IP addresses, device MAC addresses, bid document codes, font formats, business registration information, and shareholder and senior management relationship information of all bidders to construct a bidder relationship network. A comprehensive relationship score is calculated based on multi-dimensional weighted indicators. If the score exceeds a preset collusion threshold, suspected colluding bidders are immediately marked, and a visualized relationship graph is generated.

[0055] The abnormal scoring pattern identification submodule uses a temporal attention network to load all historical scoring data from the evaluation experts. It then compares the current project score with historical data and scores from other experts in the same group over time: if a single expert's score deviates from the expert group's average by more than a preset standard deviation, it is identified as a deviation from the center pattern; if an expert consistently gives extreme scores to similar bidders in multiple consecutive projects, it is identified as a fixed preference pattern; and if the scoring order highly overlaps with the order of the bidder's on-site presentation, it is identified as a time-related pattern. Upon identifying anomalies, corresponding risk tags are simultaneously assigned.

[0056] The time logic conflict detection submodule verifies nodes such as announcement release time, bid deadline, bid opening time, bid evaluation duration, and public announcement duration according to the legal procedures and time limits for bidding, and identifies problems such as process reversal, timeout, and time sequence contradictions.

[0057] After all model tests are completed, the engine outputs a compliance score of 0 to 100 points and a corresponding risk label based on the combined results.

[0058] S5. The system presets three levels of scoring thresholds: the first threshold (80 points), the second threshold (60 points), and the third threshold (40 points), and also divides the risk into three categories: general, moderate, and severe.

[0059] Level 1 intervention: If the score is 60 < score ≤ 80, or if a general label is hit, a pop-up window will remind you of the risk issues and rectification suggestions, and the process will proceed normally;

[0060] Level 2 intervention: 40 < score ≤ 60, or a moderate violation label is hit, the current operation interface is locked, the operator must complete the rectification and re-test, and the process can only be unlocked after the risk meets the standard;

[0061] Level 3 Intervention: If the score is ≤40, or if serious tags such as bid rigging or fraudulent bidding are triggered, the system will immediately terminate the bidding process and send a blocking report to the legal representative and compliance officer.

[0062] Upon triggering any level of intervention, the system automatically executes the solidification of the responsibility log: capturing a full-process interface snapshot at the moment the intervention is triggered, the original data and decision basis detected by the AI ​​model, the operator's account, identity, operation time, and other information; digitally signing the log using national cryptographic algorithms and adding a standard timestamp to generate a legally valid electronic responsibility certificate; and simultaneously uploading the log to the tenderer's main responsibility traceability platform to complete the solidification of the certificate.

[0063] S6, the blockchain evidence storage module hashes all core data, including process behavior data, compliance scores, risk tags, intervention records, fixed logs, and electronic responsibility certificates, and writes the hash values ​​to the corresponding blocks on the consortium blockchain. The blockchain uses a chain structure, ensuring that data cannot be tampered with or deleted once it's on the chain. Simultaneously, the system generates unique compliance traceability codes for each responsible person and each bidding project, binding each traceability code to a blockchain block address and responsibility certificate. External personnel can quickly query the entire audit process data using the traceability codes.

[0064] In daily operations, the system collects records of operators' adoption and rejection of risk warnings and modification suggestions: samples where the project progresses normally after adoption and rectification are marked as positive samples; samples where the project results in complaints, bid cancellations, or audit issues after rejection are marked as negative samples. The system uses positive and negative samples to incrementally train and fine-tune the four major AI models on a monthly / quarterly basis, continuously improving the model recognition accuracy and adapting to new regulations and new types of violations.

[0065] The "Responsible Entity Change Tracking and Risk File Compliance Tracing Interface" module's "Change Tracking" submodule continuously counts the number of times each responsible person triggers Level 3 intervention. Once the cumulative number reaches a preset threshold, the system extracts historical risk data for that person across four dimensions: document preparation, bid evaluation fairness, information confidentiality, and contract performance. This data is then used to generate a visual compliance risk file via a radar chart rendering engine, intuitively displaying the risk level in each dimension. Simultaneously, it supports one-click generation of standard text reports, which can be directly used for internal supervision or reported to industry regulatory authorities.

[0066] This invention provides an artificial intelligence-based full-process compliance management system for the main responsibility of tendering entities, and a method for its execution, including:

[0067] The data acquisition module establishes data interfaces with the tenderer's electronic tendering platform, contract management system, bid evaluation site terminal, and internal office system to achieve automated and real-time acquisition of multi-source heterogeneous data in the seven stages of tendering. It supports unified access and preliminary cleaning of structured, semi-structured, and unstructured data, providing raw data support for subsequent analysis.

[0068] The knowledge graph module is deployed based on a graph database and is responsible for storing and updating the knowledge graph of the main responsibilities of the tendering party. The module has a built-in responsibility node retrieval interface and risk path reasoning engine, which can quickly complete process behavior matching, responsibility entity positioning, risk transmission path deduction, and support dynamic iteration of graph nodes and relationships based on new regulations and new cases.

[0069] The AI ​​detection engine module is the core intelligent detection unit of this system, comprising four independent sub-modules, each corresponding to one of the four AI compliance detection models in the methodology. Each sub-module uses a deep learning model or ensemble learning model to independently compute and collaboratively output results. This module also incorporates an active learning feedback loop: it collects data on the adoption / rejection of risk warnings and modification suggestions by staff, marking adopted and rectified samples as positive samples, and rejecting samples that lead to complaints, bid rejections, or audit issues as negative samples. It periodically uses positive and negative samples to incrementally train and fine-tune the AI ​​model, ensuring the model continuously adapts to the latest laws and regulations, local bidding rules, and new types of violations.

[0070] The tiered intervention module has built-in three-level scoring thresholds (first threshold, second threshold, and third threshold, with the first threshold > second threshold > third threshold) and a full list of risk label tiers. It executes tiered intervention strategies based on the compliance score and risk labels output by the AI ​​detection engine.

[0071] Level 1 intervention: If the compliance score is between the second and first thresholds, or if a general risk label is hit, the system will pop up a warning and push modification suggestions without disrupting existing business processes.

[0072] Level 2 intervention: If the compliance score is between the third and second thresholds, or if a moderate violation label is hit, the current operation interface will be locked. The operator must complete the rectification and risk downgrade before the process can continue.

[0073] Level 3 intervention: If the compliance score is lower than or equal to the third threshold, or if serious violations such as bid rigging or fraudulent bidding are detected, the system will automatically terminate the current bidding process, send a process blocking report to the legal representative and compliance management personnel of the bidding entity, and automatically solidify all evidence in the entire chain.

[0074] The blockchain evidence storage module connects to industry consortium blockchain nodes to perform hash-based on-chain evidence storage of key bidding process data, AI detection results, risk warning records, intervention operation logs, and responsibility certificates. Leveraging the decentralized, tamper-proof, and traceable characteristics of blockchain, it ensures the authenticity and validity of all compliant data, while forming a chain-like cross-verification with the digital signatures and timestamps in the responsibility logs.

[0075] The compliance traceability interface module provides a visual human-computer interaction interface for tenderers, regulators, and auditors. It supports multi-dimensional searches by responsible entity, project number, and time interval, intuitively displaying the entire project compliance audit path and electronic responsibility certificates. The module includes a sub-module for tracking changes in responsible entities: when the same responsible entity triggers a preset number of Level 3 intervention events in different projects, it automatically extracts quantitative scores for each responsibility dimension and generates a compliance risk profile for the entity through a radar chart rendering engine. The risk profile covers four responsibility dimensions: document preparation, bid evaluation fairness, information confidentiality, and contract performance, intuitively displaying the risk distribution and supporting one-click export of standardized reports, which can be directly submitted to internal supervisory bodies or industry administrative supervision departments.

[0076] In this implementation plan: After system startup, the data acquisition module automatically captures multi-source data from the entire bidding process; after parsing via natural language processing, it maps the data to a knowledge graph to determine the responsible parties and compliance verification points; the AI ​​detection engine calls multiple models to complete real-time compliance detection, outputting scores and risk labels; the tiered intervention module executes corresponding warnings, rectifications, and process blocking operations based on the results, and solidifies the responsibility log; all core data is uniformly stored on the blockchain to generate compliance traceability codes; managers and auditors can complete the entire process responsibility query and traceability through the traceability interface and traceability codes; at the same time, the active learning loop continuously optimizes the AI ​​model to achieve closed-loop system operation.

[0077] This solution leverages artificial intelligence technology to automate the entire bidding process, effectively addressing the issues of low efficiency, inconsistent standards, and incomplete regulatory coverage inherent in traditional manual review. Utilizing multiple AI detection models, the system can accurately identify various violations such as hidden exclusive clauses, bidder collusion, abnormal evaluation scores, and process sequence conflicts, thus compensating for the shortcomings of traditional rule engines and significantly improving the comprehensiveness and accuracy of violation identification.

[0078] This has changed the previous post-audit control model, and instead conducts real-time monitoring during business operations. It also implements graded intervention measures according to risk levels, and promptly stops violations through early warning reminders, mandatory rectification, and process blocking. This reduces bid rejections, complaints, and economic losses from the source, and shifts compliance control from post-event accountability to pre-event prevention and in-event control.

[0079] By leveraging knowledge graphs to accurately link process behaviors with responsible personnel, the system clarifies accountability. Combining blockchain, digital signatures, and timestamps to solidify data, and relying on compliance traceability codes to form a complete and verifiable chain of evidence, the system solves the problems of difficult accountability and easily tampered data. Furthermore, the system possesses self-learning and iteration capabilities, adapting to updates in regulations and local rules, ensuring long-term stability and effective implementation of bidding compliance management.

[0080] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence, characterized in that: Includes the following steps: S1. Construct a knowledge graph of the main responsibilities of the tendering party. The knowledge graph includes responsibility nodes, compliance rule nodes, and historical risk event nodes. Edge connections are established between each node based on legal relationships and case causal relationships. S2. In response to the bidding initiation command, collect multi-source data throughout the entire bidding process, which includes at least the project initiation stage, document preparation stage, announcement release stage, bid receipt stage, bid opening and evaluation stage, bid awarding and contract signing stage, and contract performance stage. S3. Use a natural language processing model to extract entities and identify intents from the text data in the multi-source data, generate a process behavior sequence, and map the process behavior sequence to the knowledge graph of the main responsibility of the tenderer to determine the responsible entity and compliance verification point corresponding to each process behavior. S4. Based on the compliance verification point, call the corresponding AI compliance detection model to perform real-time compliance analysis on the current process behavior and generate a compliance score and risk label; the AI ​​compliance detection model includes a clause exclusivity detection model, a relational graph detection model, an abnormal scoring pattern recognition model, and a time logic conflict detection model. S5. When the compliance score is lower than the preset threshold or hits the preset high-risk label, a graded intervention action is triggered. The graded intervention action includes prompting and warning, forced modification, process blocking and solidification of responsibility log. S6. Store the process behavior, compliance score, risk label and intervention action in the compliance audit blockchain in an immutable manner, and generate a compliance traceability code associated with each responsible entity.

2. The method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence as described in claim 1, characterized in that: The specific method for constructing the knowledge graph of the main responsibility of the tendering party in S1 includes: extracting the responsible entity, behavioral entity, and consequence entity from the legal and regulatory database, historical tendering case database, and internal compliance system documents; using graph neural networks to learn the implicit compliance dependency paths between entities, generating a dynamic responsibility transmission chain, and setting weight coefficients and timeliness labels for each responsibility node.

3. The method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence as described in claim 1, characterized in that: The clause exclusivity detection model in S4 uses a fine-grained classifier based on a pre-trained language model to identify whether there are exclusivity clauses in the tender documents that specify a particular brand, region, scale, or ownership, and outputs the exclusivity risk level and modification suggestions.

4. The method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence as described in claim 1, characterized in that: The association graph detection model in S4 is used to: extract device fingerprints, network addresses, file metadata and business association information between different bidders; construct a bidder similarity network and calculate the comprehensive association score between any two bidders; when the association score exceeds a preset collusion suspicion threshold, generate an association risk label and mark the corresponding bidder.

5. The method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence as described in claim 1, characterized in that: The abnormal scoring pattern recognition model in S4 employs a temporal attention network to compare and analyze the historical scoring sequences of each member of the evaluation committee with the current scoring sequences, identifying at least one of the following abnormal patterns: Deviation from the center pattern means that the current score deviates from the average score of all evaluation committee members by more than a preset standard deviation multiple; Fixed preference pattern, which means consistently giving abnormally high or low scores to the same type of bidders in multiple bid evaluations; The time-related pattern indicates that the scoring order and the order of bidder presentations show an unreasonable strong correlation.

6. The method for full-process compliance management of the main responsibility of the tendering party based on artificial intelligence as described in claim 1, characterized in that: The solidification of the responsibility log in S5 includes: recording a snapshot of the process context, decision basis, and operator identity information when a graded intervention action is triggered; synchronizing the responsibility log to the tenderer's main responsibility traceability platform to generate an undeniable electronic responsibility certificate; the responsibility certificate is solidified using digital signatures and timestamps, and forms a chain-verification association with the corresponding block in the compliance audit blockchain.

7. A compliance management system for the entire process of tendering entity responsibility based on artificial intelligence, used to execute the method of any one of claims 1 to 6, characterized in that, include: The data acquisition module is configured to interface with the tenderer's electronic tendering platform, contract management system, and bid evaluation site terminal to collect multi-source heterogeneous data throughout the entire tendering process. The knowledge graph module stores and dynamically updates the knowledge graph of the main responsibility of the tendering party, and is configured with a responsibility node retrieval interface and a risk path reasoning engine. The AI ​​detection engine module includes a clause exclusivity detection submodule, a relational graph detection submodule, an anomaly scoring pattern recognition submodule, and a time logic conflict detection submodule. The tiered intervention module is configured to execute multi-level intervention strategies, ranging from alerts and warnings to process blocking, based on the compliance score and risk label output by the AI ​​detection engine module. The blockchain evidence storage module is connected to at least one consortium blockchain node and is used to hash and store data on the blockchain for intervention events, the identity of responsible parties, and key process nodes. The compliance traceability interface module provides a visual traceability interface for the responsible parties of the tendering entity, supporting queries of compliance audit paths and responsibility vouchers by responsible party, project number, or time period.

8. The AI-based full-process compliance management system for the main responsibility of the tendering entity as described in claim 7, characterized in that, The AI ​​detection engine module also includes an active learning feedback loop, which is used to collect adversarial data for adoption or rejection, and to incrementally train or fine-tune the model.

9. The AI-based full-process compliance management system for the main responsibility of the tendering party as described in claim 7, characterized in that, The tiered intervention module performs tiered intervention according to the following strategy, wherein the preset thresholds include a first threshold, a second threshold, and a third threshold, and the first threshold is greater than the second threshold, and the second threshold is greater than the third threshold: Level 1 intervention: When the compliance score is higher than the second threshold and lower than or equal to the first threshold, or when a general risk warning label is hit, a pop-up prompt and modification suggestions are output, without blocking the process; Level 2 intervention: When the compliance score is higher than the third threshold and lower than or equal to the second threshold, or when a moderate violation risk label is hit, the current operation interface is locked, and the user is required to modify it until the risk level is reduced before continuing. Level 3 intervention: When the compliance score is lower than or equal to the third threshold, or when a serious risk label such as direct bid rigging or fraudulent bidding is triggered, the current bidding process is automatically terminated, a blocking report is sent to the legal representative and compliance officer of the bidding entity, and the entire chain of evidence is solidified.

10. The AI-based full-process compliance management system for the main responsibility of the tendering party as described in claim 7, characterized in that, The compliance traceability interface module is also equipped with a sub-module for tracking changes in responsible entities, which is used to accumulate the number of times three-level interventions are triggered, generate a compliance risk profile, and support one-click report submission.

Images