Bid document requirement intelligent checking method and system driven by large language model

Abstract

The invention belongs to the technical field of artificial intelligence, and particularly relates to a bidding document requirement intelligent checking method and system driven by a large language model. Comprising a bidding file requirement knowledge base, a bidding file analysis module, a bidding file analysis module, an intelligent checking module and an auxiliary modification module. The method has the advantages that based on the large language model technology, the problems that in the bidding document checking process, semantic understanding depth is insufficient, multi-modal data integration is missing, dynamic rule generation lags and the like can be effectively solved, and high-precision and full-automatic intelligent checking required by the bidding document is achieved. The method has the following advantages: the checking efficiency is improved: the automatic checking is 95% faster than the manual checking, the average time is shortened from 3 hours per copy to 9 minutes per copy, the checking efficiency is greatly improved, and the bidding document compilation cycle is shortened.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a method and system for intelligent verification of tender document requirements driven by a large language model. Background Technology

[0002] In the bidding field, the quality of bid documents directly affects a company's success in winning a bid. Bidding documents typically contain complex format requirements, technical specifications, and commercial terms. However, current bid document verification mainly relies on manual checks or automated tools based on rule engines and traditional natural language processing (NLP) technologies. Manual verification is inefficient, costly, and prone to errors, while rule engines and traditional NLP technologies lack semantic understanding capabilities, making them ill-suited for complex and ever-changing bid document requirements, unable to identify logical contradictions and potential rules, and inflexibly adaptable to the diverse needs of different projects. This patent proposes an intelligent verification technology based on a large language model. Leveraging the powerful semantic understanding and logical reasoning capabilities of large language models, it achieves automated and intelligent verification of different bid document requirements for different projects. Compared with existing technologies, this patent not only efficiently identifies format errors and missing content, but also handles complex logic and implicit requirements, and automatically generates detailed verification reports and optimization suggestions, significantly improving verification efficiency and accuracy, reducing labor costs, and increasing the bid success rate. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for intelligent verification of tender document requirements driven by a large language model. This method and system can effectively solve problems such as insufficient semantic understanding depth, lack of multimodal data integration, and lag in dynamic rule generation during the verification process of tender documents, and achieve high-precision, fully automatic intelligent verification of tender document requirements.

[0004] The technical solution of the present invention is as follows: a large language model-driven intelligent verification system for tender document requirements, including a tender document requirements knowledge base, a tender document parsing module, a tender document parsing module, an intelligent verification module, and an auxiliary modification module;

[0005] The tender document requires the knowledge base to collect historical tender documents, tender documents, and industry standard data to form a training set;

[0006] The tender document parsing module performs multimodal requirement parsing and dynamic rule extraction;

[0007] The bid document parsing module uses a large language model to perform file format conversion, text cleaning, segmentation and tagging of the bid document, parse the bid document content, extract key information, and associate it with the bid document requirements to generate verification prompts.

[0008] The intelligent verification module uses a trained large language model to verify the bid documents, identify non-compliance items in the bid document requirement library, visualize the non-matching content in the bid documents through an attention mechanism, and generate a verification report to achieve accurate semantic alignment and error tracing driven by dynamic rules.

[0009] Based on the verification results, the auxiliary modification module dynamically generates optimization suggestions by combining the search enhancement generation method with the historical winning bid case database, helping users quickly locate problems and make targeted modifications to improve the quality of bid documents.

[0010] The knowledge base for bid document requirements employs a fine-tuning method to perform lightweight adjustments on the large language model, enhancing its understanding of bidding terminology. Knowledge extraction and structuring are then performed to construct a knowledge base for bid document requirements that includes bid document format, review factors, maximum score, evaluation methods, and scoring criteria.

[0011] The multimodal requirement parsing in the tender document parsing module includes using the multimodal function of a large language model to uniformly transform the text, tables, and technical parameter charts in the tender document into a structured semantic representation.

[0012] The dynamic rule extraction in the tender document parsing module is based on chain reasoning of a large language model, which transforms the tender terms into executable verification logic. When special terms are added to the tender document, extended verification rules are dynamically generated to achieve targeted verification of the tender documents.

[0013] A method for intelligent verification of tender document requirements driven by a large language model includes the following steps:

[0014] Step 001: Collect historical bidding documents, tender documents, industry standards, laws and regulations, etc., and build a training dataset;

[0015] Step 002: The collected data is cleaned and formatted to remove redundant information, standardize document format, and store the text in segments and chapters;

[0016] Step 003: Use a large language model to perform semantic analysis on the preprocessed data and extract key elements, including the structural requirements, content requirements, scoring criteria, and mandatory requirements in laws, regulations, and industry standards of the tender documents;

[0017] Step 004: Store the extracted knowledge in a structured form to build a knowledge base for tender document requirements, supporting efficient querying and dynamic updates;

[0018] Step 005: Input the tender documents corresponding to the current bidding project into the system and use a large language model to perform text semantic parsing;

[0019] Step 006: Use Tabnet to identify the table area in the tender document and extract the table content; use ChartOCR to identify the chart area in the tender document and extract the key information of the chart; use the multimodal function of the large language model to uniformly transform the text, table and chart information into a structured semantic representation.

[0020] Step 007: Use a large language model to extract the review rules for this bidding project from the bidding documents and update them to the verification rule base;

[0021] Step 008: Input the tender documents to be checked into the system, and use a large language model to perform semantic parsing on the documents to identify the document structure and content;

[0022] Step 009: Based on the requirements in the knowledge base, use Stanford NLP to extract key information from the parsed file, including basic information of the bidder, project-related information, technical solutions, and commercial terms;

[0023] Step 010: Use the text similarity algorithm of the large language model to calculate similarity, associate the extracted information with the knowledge base of bidding requirements to form a mapping relationship, and provide a basis for subsequent verification;

[0024] Step 011: Based on the bidding requirements knowledge base, use the ReportLab library to check whether the layout, font, table of contents, etc. of the bid documents meet the specifications;

[0025] Step 012: Use the trained large language model to intelligently check the compliance of the bid documents based on the knowledge base of the bid document requirements;

[0026] Step 013: Categorize the errors found during the verification process, mark the parts that do not meet the requirements, and use the NLTK Natural Language Toolkit to generate a detailed verification report, including the error location, error type, and error details;

[0027] Step 014: Based on the verification results, generate optimization suggestions by enhancing the search function and combining them with the historical winning bid case database;

[0028] Step 015: Modify the tender documents online according to the optimization suggestions, and re-verify them to ensure that the modified documents meet the requirements.

[0029] The checks in step 012 include checking whether the tender documents contain typos and prohibited words; checking whether the tender documents involve any veto items; checking whether the technical solution covers all technical requirements; checking whether the commercial terms comply with the provisions of the tender documents; checking whether the qualification certificates are complete and valid; checking whether the price matches the technical solution; checking whether the construction period is reasonable; and checking whether there are any self-contradictory clauses.

[0030] In step 015, the parameters and prompt words of the large language model are optimized by combining user feedback and actual verification results, thereby improving the accuracy and reliability of the verification.

[0031] The beneficial effects of this invention are as follows: Based on large language model technology, this invention can effectively solve problems such as insufficient semantic understanding depth, lack of multimodal data integration, and lag in dynamic rule generation during the bid document verification process, achieving high-precision, fully automated intelligent verification as required by bid documents. It has the following advantages: Improved verification efficiency: Automated verification is 95% faster than manual verification, reducing the average time from 3 hours / document to 9 minutes / document, significantly improving verification efficiency and shortening the bid document preparation cycle; Enhanced verification accuracy: Based on semantic understanding and logical reasoning, it can identify errors that are difficult to detect using traditional methods, with an error identification accuracy rate of over 95%, improving verification accuracy; Dynamic rule updates: The response time for adding special requirements from the bidding party is shortened from 2 days to 10 minutes; Increased bid success rate: It ensures that bid documents meet the bidding requirements, increasing the bid success rate. Attached Figure Description

[0032] Figure 1 The flowchart of a method for intelligent verification of tender document requirements driven by a large language model provided by the present invention is shown. Detailed Implementation

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] A large language model-driven intelligent verification system for tender document requirements includes a tender document requirements knowledge base, a tender document parsing module, a tender document parsing module, an intelligent verification module, and an auxiliary modification module.

[0035] The tender document requires a knowledge base for storing structured bidding knowledge. Its construction process is as follows: First, historical tender documents, winning bid announcements, tender document templates, and relevant industry standards and regulations are collected from procurement platforms and bidding agency websites using web crawling technology to form an initial dataset. Then, the collected documents in various formats (such as PDF, DOC, and XLSX) are preprocessed, including converting PDF documents into editable text using OCR (Optical Character Recognition) technology, and cleaning, denoising, and standardizing the text using natural language processing tools. Next, an efficient fine-tuning method based on parameters such as LoRA (Low-Rank Adaptation) is used to perform domain adaptation fine-tuning on the selected large language model (such as DeepSeek, LLaMA, etc.). During fine-tuning, specialized terminology and clause logic in the bidding field are constructed into an instruction fine-tuning dataset, focusing on improving the model's understanding and reasoning ability regarding key concepts such as "evaluation factors," "technical parameters," and "rejection clauses." Finally, this finely tuned large language model is used to extract knowledge from the preprocessed data, identifying and extracting key elements such as document format requirements, review factors, score allocation, evaluation methods, and mandatory clauses. These elements are then stored in a structured form (such as JSON or using the graph database Neo4j) to form a queryable and updatable knowledge base for bid document requirements.

[0036] The tender document parsing module is responsible for in-depth analysis of tender documents. This module achieves this through two core functions: multimodal information parsing and dynamic rule extraction. Regarding multimodal requirement parsing, the module first utilizes specialized tools to process the tender documents: for example, it uses the TabNet model to identify and extract table content from the document, uses tools such as ChartOCR to parse technical parameter charts, and combines the multimodal understanding capabilities of a large language model to uniformly transform the text descriptions, table data, and chart information in the document into a machine-understandable structured semantic representation. For example, for a text requirement "Submit audited financial statements for the past three years," the module can parse it into a structured data object containing fields such as {"Requirement Item": "Financial Statements", "Format": "PDF", "Content": ["Balance Sheet", "Profit and Loss Statement", "Cash Flow Statement"], "Years": "Past Three Years"}. Regarding dynamic rule extraction, this module utilizes the chain-of-thought (CoT) capability of the large language model to transform the clause-based requirements of the tender document into computer-executable verification logic. For example, when faced with the clause "Bidders must possess ISO 9001 quality management system certification," the module guides the large model through step-by-step reasoning, ultimately generating a rule usable by the program: "IF the bidder's qualification documents do not contain a valid ISO 9001 certification certificate THEN is marked as 'non-conformity'." When new or special clauses appear in the tender documents, this process can dynamically generate extended verification rules and update them to the system's verification rule library.

[0037] The tender document parsing module is responsible for processing the tender documents to be reviewed. Its workflow is as follows: First, it performs file format conversion and content extraction, using libraries such as pdfplumber and python-docx to convert the tender documents (whether in PDF or Word format) into plain text. Next, the text is cleaned and standardized, such as removing headers and footers, standardizing numbering formats, and using natural language processing technology for intelligent segmentation and chapter marking to accurately identify core chapters such as "Business Section," "Technical Solution," and "Quotation List." Then, the module uses a fine-tuned large language model to extract key information from these chapters, such as the bidder's name, qualification certificates, technical parameter responses, construction period, and quotation. Through semantic similarity calculation, it establishes a mapping between the extracted information and the corresponding requirements in the tender document requirements knowledge base, preparing for subsequent review.

[0038] The intelligent verification module is the core of the system, responsible for performing compliance checks. This module loads the outputs of the aforementioned modules, namely the structured bidding requirements, dynamically generated verification rules, and the parsed bid document content. During verification, the module drives a trained large language model to perform multi-dimensional, in-depth semantic-level verification of the bid document. Verification content includes, but is not limited to: checking format conformity based on rules, identifying typos or prohibited words; utilizing the model's logical reasoning ability to check whether the technical solution fully covers and positively responds to all technical requirements of the bidding document, whether commercial terms (such as payment methods and liability for breach of contract) are consistent with regulations, whether qualification documents are complete and valid, whether the price calculation is accurate and matches the technical solution, and whether there are any inconsistencies in the document content. For identified non-compliance items, the module uses the large language model's attention mechanism to locate the specific location in the original bid document and highlight it for visualization. Finally, the module automatically generates a structured verification report, detailing the location, type, specific description, and possible reasons for each non-compliance item.

[0039] The auxiliary modification module provides corrective support to users after the intelligent verification module identifies problems. This module is based on Retrieval Enhanced Generation (RAG) technology. The system maintains a historical database of winning bids. When a non-compliance is detected (e.g., "insufficient performance evidence for similar projects"), the module uses this non-compliance as a search condition to perform a vector similarity search in the database, finding the most relevant historical winning bids and their solutions. Subsequently, the description of the non-compliance, the context of the bidding requirements, and the retrieved relevant case information are input as prompts into the large language model, dynamically generating targeted and actionable modification suggestions. For example, it might suggest that the user supplement specific types of performance evidence and provide corresponding material templates or expression examples. Users can then directly modify their bid documents in the system's editing interface based on these suggestions. The system also supports re-verification after modification and can continuously optimize the prompts and parameters of the large language model based on user feedback and verification results, forming a self-optimizing closed-loop system.

[0040] like Figure 1 As shown, a method for intelligent verification of tender document requirements driven by a large language model includes the following steps:

[0041] Step 001: Automatously collect historical bidding announcements, bidding documents, winning bid announcements, bid templates, and relevant industry technical specifications and national laws and regulations from designated procurement platforms, bidding agency websites, and public databases using web crawling tools (such as the Scrapy framework in Python). Use these multi-source, multi-format documents as the original training dataset for constructing a knowledge base for the bidding and tendering field.

[0042] Step 002: Clean and format the raw data collected in Step 001. Specifically, this includes: using an OCR engine (such as Tesseract) to recognize and convert the image text in the scanned PDF file into editable text; using document processing libraries (such as python-docx and pdfplumber for Python) to parse the document structure and remove redundant information such as headers, footers, and watermarks; and using natural language processing tools (such as Spacy or NLTK) to intelligently segment and divide the text into chapters based on features such as "Chapter 1" and "Section 1," and finally storing the processed content in a structured JSON or XML format for subsequent processing.

[0043] Step 003: Input the preprocessed text data from Step 002 into a domain-fine-tuned large language model. By designing specific prompts, guide the model to perform sequence labeling and relation extraction, thereby identifying and extracting key elements related to the tender document from the text. These elements include, but are not limited to: structural requirements of the document (such as directory hierarchy, document format), content requirements (such as key technical points, qualification certificate list), scoring criteria (such as score allocation, calculation formula), and mandatory clauses in laws and regulations. This process can be combined with rule templates (such as regular expressions) to improve extraction accuracy.

[0044] Step 004: Transform the unstructured knowledge elements extracted in Step 003 into a structured data model. For example, abstract the "scoring criteria" into objects containing attributes such as "scoring items," "scores," and "scoring details." This structured data can be stored in a relational database (such as MySQL) or a graph database (such as Neo4j), and indexed to support efficient multi-condition queries. This knowledge base is designed with an update interface; when new specifications are released, an incremental learning process can be triggered to achieve dynamic knowledge updates.

[0045] Step 005: Input the tender documents (usually in PDF format) of the project to be processed into the system. First, obtain all the text information through OCR and text extraction technology. Then, use a large language model to perform macro-level semantic understanding of the entire document, identify the core parts of the document, such as "Instructions to Bidders", "Contract Terms", "Technical Specifications", etc., and gain a preliminary understanding of the key content of each part.

[0046] Step 006: For non-textual information in the tender documents, a specialized domain model is used for extraction: a table recognition model (such as TabNet based on deep learning) is used to detect and parse tables in the document to restore their row and column structure; a chart recognition tool (such as ChartOCR) is used to interpret curves and data points in technical parameter charts. Then, leveraging the multimodal capabilities of a large language model (or through text description), the extracted table content and chart information are fused with the text semantics obtained in Step 005 to generate a unified and complete structured semantic representation of the tender requirements, ensuring no information is omitted.

[0047] Step 007: Based on the structured semantic information obtained in steps 005 and 006, the natural language clauses in the tender documents are transformed into verification rules that can be automatically executed by a computer using the logical reasoning capabilities of the large language model (such as chain reasoning techniques). For example, the clause "The project manager must possess an Information Systems Project Manager qualification" is transformed into the logical rule: "If the project manager's qualification certificate in the tender document does not contain 'Information Systems Project Manager,' then this is marked as a 'major non-conformity.'" These newly generated rules are automatically added to the system's verification rule base.

[0048] Step 008: Parse the tender documents to be reviewed. Use a file parsing library (such as pdfplumberforPDF, python-docxforWord) to convert them into plain text, preserving as much of the original formatting information as possible. Utilize natural language processing technology to identify the chapter structure of the tender documents (such as the business section, technical section, and pricing section), and perform preliminary semantic block segmentation of the content of each chapter.

[0049] Step 009: Based on the parsing and structuring in Step 008, use a Named Entity Recognition (NER) model (such as a BERT-based fine-tuned model or the Stanford CoreNLP tool) to accurately extract key information entities from the tender document text. These entities include: bidder name, registered address, relevant qualification certificate numbers, technical parameter response values, project duration, bid price, etc. The extracted information is then associated with the requirement entries in the knowledge base.

[0050] Step 010: Use a semantic vector model (such as Sentence-BERT) to convert the key information fragments in the tender document extracted in Step 009 and the tender requirement clauses in the knowledge base into high-dimensional vectors. By calculating the cosine similarity between the vectors, establish a semantic association mapping relationship between the content of the tender document and the tender requirements. For example, match the description of "project experience" in the tender document with "similar performance requirements" in the knowledge base to lay the foundation for subsequent accurate verification.

[0051] Step 011: Automatously check whether the format of the tender documents conforms to the specifications by calling the document parsing interface programmatically. For example, use Python's ReportLab library or similar tools to programmatically read the metadata of the PDF file and check whether its page numbers, font type and size, page margins, line spacing, and table of contents are automatically generated and linked accurately, and whether they strictly conform to the format requirements of the tender documents.

[0052] Step 012: The structured bidding requirements, verification rules, and parsed bid document content obtained in the previous steps are integrated using carefully designed prompts and input into a finely tuned large language model for deep semantic understanding and logical reasoning. The verification scope covers: substantive response (e.g., whether technical parameters meet standards), completeness (e.g., whether necessary supporting documents are missing), consistency (e.g., whether the price calculation is accurate), and compliance (e.g., whether there are any prohibitive commitments). The model will output detailed compliance judgments and reasons, including but not limited to: checking whether the bid document contains typos and prohibited words; checking whether the bid document involves any veto items; checking whether the technical solution covers all technical requirements; checking whether the commercial terms comply with the provisions of the bidding documents; checking whether the qualification certificates are complete and valid; checking whether the price matches the technical solution; checking whether the construction period is reasonable; and checking for any self-contradictory clauses.

[0053] Step 013: Classify and grade all non-conformities found in Step 012 according to their severity (e.g., "Rejected Items", "Significant Deviations", "Formatting Issues"). Using a report generation engine (e.g., Jinja2 template engine), automatically populate the error type, its specific location in the original text (page number, paragraph), detailed description, and suggested modifications into a preset template to generate a comprehensive and highly readable verification report (which can be exported in PDF, DOCX, etc. formats).

[0054] Step 014: For the identified issues, a solution is provided using Retrieval Enhanced Generation (RAG) technology. The system first retrieves successful case fragments from the historical winning bid case database (which has been vectorized) that are semantically closest to the current non-compliance item. Then, the non-compliance item description, the context of the bidding requirements, and the retrieved relevant cases are used as prompts and input into the large language model, which generates specific and actionable optimization suggestions (e.g., "Refer to the XX project case and supplement with a XX type of inspection report").

[0055] Step 015: The system displays the verification report and optimization suggestions to the user, and provides an online editing interface for the user to directly modify the tender document. After the user makes modifications, incremental verification of the modified part can be triggered. In addition, the system uses the user's modification behavior and the corresponding verification results as feedback data to continuously optimize the core large language model. By fine-tuning the model parameters or optimizing the prompt word strategy through reinforcement learning techniques (such as RLHF), a self-itergencing, increasingly accurate closed-loop system is formed.

Claims

1. A large language model-driven intelligent verification system for tender document requirements, characterized in that: It includes a knowledge base for tender document requirements, a tender document parsing module, a tender document parsing module, an intelligent verification module, and an auxiliary modification module; The tender document requires the knowledge base to collect historical tender documents, tender documents, and industry standard data to form a training set; The tender document parsing module performs multimodal requirement parsing and dynamic rule extraction; The bid document parsing module uses a large language model to perform file format conversion, text cleaning, segmentation and tagging of the bid document, parse the bid document content, extract key information, and associate it with the bid document requirements to generate verification prompts. The intelligent verification module uses a trained large language model to verify the bid documents, identify non-compliance items in the bid document requirement library, visualize the non-matching content in the bid documents through an attention mechanism, and generate a verification report to achieve accurate semantic alignment and error tracing driven by dynamic rules. Based on the verification results, the auxiliary modification module dynamically generates optimization suggestions by combining the search enhancement generation method with the historical winning bid case database, helping users quickly locate problems and make targeted modifications to improve the quality of bid documents.

2. The intelligent verification system for bid document requirements driven by a large language model as described in claim 1, characterized in that: The knowledge base for bid document requirements employs a fine-tuning method to perform lightweight adjustments on the large language model, enhancing its understanding of bidding terminology. Knowledge extraction and structuring are then performed to construct a knowledge base for bid document requirements that includes bid document format, review factors, maximum score, evaluation methods, and scoring criteria.

3. The intelligent verification system for bid document requirements driven by a large language model as described in claim 1, characterized in that: The multimodal requirement parsing in the tender document parsing module includes using the multimodal function of a large language model to uniformly transform the text, tables, and technical parameter charts in the tender document into a structured semantic representation.

4. The intelligent verification system for bid document requirements driven by a large language model as described in claim 1, characterized in that: The dynamic rule extraction in the tender document parsing module is based on chain reasoning of a large language model, which transforms the tender terms into executable verification logic. When special terms are added to the tender document, extended verification rules are dynamically generated to achieve targeted verification of the tender documents.

5. A method for intelligent verification of tender document requirements driven by a large language model, characterized in that, Includes the following steps: Step 001: Collect historical bidding documents, tender documents, industry standards, laws and regulations data to build a training dataset; Step 002: The collected data is cleaned and formatted to remove redundant information, standardize document format, and store the text in segments and chapters; Step 003: Use a large language model to perform semantic analysis on the preprocessed data and extract key elements, including the structural requirements, content requirements, scoring criteria, and mandatory requirements in laws, regulations, and industry standards of the tender documents; Step 004: Store the extracted knowledge in a structured form to build a knowledge base for tender document requirements, supporting efficient querying and dynamic updates; Step 005: Input the tender documents corresponding to the current bidding project into the system and use a large language model to perform text semantic parsing; Step 006: Use Tabnet to identify the table area in the tender document and extract the table content; use ChartOCR to identify the chart area in the tender document and extract the key information of the chart; use the multimodal function of the large language model to uniformly transform the text, table and chart information into a structured semantic representation. Step 007: Use a large language model to extract the review rules for this bidding project from the bidding documents and update them to the verification rule base; Step 008: Input the tender documents to be checked into the system, and use a large language model to perform semantic parsing on the documents to identify the document structure and content; Step 009: Based on the requirements in the knowledge base, use Stanford NLP to extract key information from the parsed file, including basic information of the bidder, project-related information, technical solutions, and commercial terms; Step 010: Use the text similarity algorithm of the large language model to calculate similarity, associate the extracted information with the knowledge base of bidding requirements to form a mapping relationship, and provide a basis for subsequent verification; Step 011: Based on the bidding requirements knowledge base, use the ReportLab library to check whether the layout, font, and table of contents of the bid documents meet the specifications; Step 012: Use the trained large language model to intelligently check the compliance of the bid documents based on the knowledge base of the bid document requirements; Step 013: Categorize the errors found during the verification process, mark the parts that do not meet the requirements, and use the NLTK Natural Language Toolkit to generate a detailed verification report, including the error location, error type, and error details; Step 014: Based on the verification results, generate optimization suggestions by enhancing the search function and combining them with the historical winning bid case database; Step 015: Modify the tender documents online according to the optimization suggestions, and re-verify them to ensure that the modified documents meet the requirements.

6. The intelligent verification method for bid document requirements driven by a large language model as described in claim 5, characterized in that, The checks in step 012 include checking whether the tender documents contain typos and prohibited words; checking whether the tender documents involve any veto items; checking whether the technical solution covers all technical requirements; checking whether the commercial terms comply with the provisions of the tender documents; checking whether the qualification certificates are complete and valid; checking whether the price matches the technical solution; checking whether the construction period is reasonable; and checking whether there are any self-contradictory clauses.

7. The intelligent verification method for bid document requirements driven by a large language model as described in claim 5, characterized in that: In step 015, the parameters and prompt words of the large language model are optimized by combining user feedback and actual verification results, thereby improving the accuracy and reliability of the verification.

Images