Bid file generation method and device based on large model, equipment and medium

Abstract

The invention discloses a bidding file generation method and device based on a large model, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a target bidding file, and extracting a target entity in the target bidding file; constructing a target knowledge graph corresponding to the target bid inviting file based on the target entity, and determining target structured data corresponding to the target knowledge graph; retrieving target matching information corresponding to the target structured data from a target knowledge base, and generating a target cue word based on the target matching information and the target structured data; the target knowledge base is a knowledge base constructed based on product information, enterprise information and historical bidding files; and generating an initial bidding file based on the target cue word and a preset generation rule by using the target large model, obtaining revision information corresponding to the initial bidding file, and determining a target bidding file corresponding to the initial bidding file based on a preset bidding file format and the revision information. The generation efficiency and the generation quality of the bidding file can be improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and medium for generating tender documents based on a large model. Background Technology

[0002] The current field of bid document preparation faces numerous challenges. Firstly, when submitting bids, the required product technical parameters, company qualification certificates, and historical case data are often scattered across multiple unstructured documents, such as PDF manuals, Word technical specifications, and scanned copies. These documents not only vary in format but also lack a unified retrieval mechanism, making information extraction exceptionally cumbersome. Furthermore, manual retrieval is not only inefficient but also prone to missing crucial information, thus affecting bid quality. Secondly, with product iterations and technological updates, the technical parameters of the same product may contain conflicting descriptions in different versions of documents. Relying on manual verification to ensure information consistency has a high error rate, increasing workload and the likelihood of bid rejection. Thirdly, traditional preparation processes often prove inadequate when faced with complex bidding requirements, especially those requiring customized technical solutions.

[0003] In conclusion, improving the efficiency and quality of bid document generation is a pressing technical issue that needs to be addressed. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for generating bid documents based on a large model, which can improve the efficiency and quality of bid document generation. The specific solution is as follows:

[0005] Firstly, this application provides a method for generating tender documents based on a large model, including:

[0006] Obtain the target tender document and extract the target entity from the target tender document;

[0007] Based on the target entity, construct the target knowledge graph corresponding to the target tender document, and determine the target structured data corresponding to the target knowledge graph;

[0008] The system retrieves target matching information corresponding to the target structured data from the target knowledge base, and generates corresponding target prompts based on the target matching information and the target structured data; the target knowledge base is a knowledge base built based on product information, enterprise information, and historical tender documents.

[0009] The target large model is used to generate the corresponding initial bid document based on the target prompt words and preset generation rules, and the revision information corresponding to the initial bid document is obtained. The target bid document corresponding to the initial bid document is determined based on the preset bid document format and the revision information.

[0010] Optionally, extracting the target entity from the target tender document includes:

[0011] Extract the text content from the target tender document and parse the page layout structure of the target tender document;

[0012] Based on the page layout structure, the text area and the mixed text and image area in the target tender document are determined, and the table area in the mixed text and image area is identified based on the preset table line detection method, and the signature area in the mixed text and image area is identified using the target detection model.

[0013] Target table data is constructed based on the text content and the table area, target signature data is constructed based on the text content and the signature area, and target text data corresponding to the text area is determined based on the text content.

[0014] The target tender document is subjected to compliance checks based on the target signature data, and the target entity in the target tender document is extracted based on the target text data and the target table data.

[0015] Optionally, extracting the target entity from the target tender document based on the target text data and the target table data includes:

[0016] Based on a preset regular expression rule base, the target structural features in the target text data and the target table data are determined, and the target positions corresponding to the target structural features are determined.

[0017] Using a preset machine learning model, extract the target entity corresponding to the target position in the target text data and the target table data;

[0018] Accordingly, the step of constructing the target knowledge graph corresponding to the target tender document based on the target entity, and determining the target structured data corresponding to the target knowledge graph, includes:

[0019] The target entity is used as a node, and a connection is established between nodes corresponding to different target entities with related relationships based on a preset graph database to obtain the target knowledge graph corresponding to the target tender document.

[0020] The target structured data corresponding to the target knowledge graph is determined using a graph database query language.

[0021] Optionally, before retrieving the target matching information corresponding to the target structured data from the target knowledge base, the method further includes:

[0022] Based on the product information, the enterprise information, and the historical tender documents, the corresponding first structured data and first unstructured data are determined;

[0023] Store the first structured data in the target relational database;

[0024] The first vector corresponding to the first unstructured data is determined using a target embedding model, and the first vector is stored in the target vector database;

[0025] The target knowledge base is determined based on the target relational database and the target vector database, and the data in the target knowledge base is updated in real time.

[0026] Optionally, retrieving target matching information corresponding to the target structured data from the target knowledge base includes:

[0027] Based on the Boolean retrieval method and the target entity corresponding to the target structured data, the second structured data is determined from the target relational database corresponding to the target knowledge base, and the second vector is determined from the target vector database corresponding to the target knowledge base;

[0028] The third vector corresponding to the target structured data is determined using a target embedding model, and the similarity between the second vector and the third vector is determined. The similarity greater than a preset similarity threshold is determined as the target similarity.

[0029] Determine the target vector corresponding to the target similarity from the second vector, and determine the second unstructured data corresponding to the target vector;

[0030] The second structured data and the second unstructured data are sorted based on a preset dynamic weighting algorithm, and the target matching information corresponding to the target structured data is determined based on the sorting result.

[0031] Optionally, after retrieving the target matching information corresponding to the target structured data from the target knowledge base, the method further includes:

[0032] If the target matching information is abnormal, then a corresponding alarm message is generated;

[0033] Based on the alarm information and the target knowledge graph corresponding to the target structured data, locate the abnormal node corresponding to the abnormal information and determine the abnormal entity corresponding to the abnormal node;

[0034] Based on the abnormal entity and the abnormal information, a corresponding report to be reviewed is generated, and the processing log and processing result corresponding to the report to be reviewed are obtained;

[0035] The abnormal information, the processing log, and the processing result are uploaded to the blockchain so that the abnormal information can be traced based on the blockchain.

[0036] Optionally, after obtaining the revision information corresponding to the initial bid document, the method further includes:

[0037] Based on the revision information, a difference analysis is performed on the revised initial tender document and the original initial tender document, and corresponding difference analysis results are generated.

[0038] The target prompts are updated based on the difference analysis results, and the preset generation rules are optimized based on the updated target prompts, so as to generate the tender documents using the target large model based on the optimized preset generation rules.

[0039] Secondly, this application provides a tender document generation apparatus based on a large model, comprising:

[0040] The target entity extraction module is used to obtain the target tender document and extract the target entities from the target tender document;

[0041] The target structured data determination module is used to construct a target knowledge graph corresponding to the target tender document based on the target entity, and to determine the target structured data corresponding to the target knowledge graph;

[0042] The target matching information retrieval module is used to retrieve target matching information corresponding to the target structured data from the target knowledge base, and generate corresponding target prompt words based on the target matching information and the target structured data; the target knowledge base is a knowledge base built based on product information, enterprise information and historical tender documents;

[0043] The target bid document determination module is used to generate a corresponding initial bid document based on the target prompt words and preset generation rules using the target big model, obtain the revision information corresponding to the initial bid document, and determine the target bid document corresponding to the initial bid document based on the preset bid document format and the revision information.

[0044] Thirdly, this application provides an electronic device, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor is used to execute the computer program to implement the aforementioned method for generating tender documents based on a large model.

[0047] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for generating tender documents based on a large model.

[0048] In this application, the following steps are taken: First, the target tender document is obtained, and the target entities in the target tender document are extracted. Then, a target knowledge graph corresponding to the target tender document is constructed based on the target entities, and the target structured data corresponding to the target knowledge graph is determined. Subsequently, target matching information corresponding to the target structured data is retrieved from the target knowledge base, and corresponding target prompt words are generated based on the target matching information and the target structured data. The target knowledge base is a knowledge base constructed based on product information, enterprise information, and historical tender documents. Finally, the target big model is used to generate the corresponding initial tender document based on the target prompt words and preset generation rules, and the revision information corresponding to the initial tender document is obtained. The target tender document corresponding to the initial tender document is determined based on the preset tender document format and the revision information. As can be seen from the above, this application first extracts target entity information from the target tender document; then, it constructs a target knowledge graph based on the target entity information and determines the target structured data corresponding to the target knowledge graph; next, it retrieves target matching information corresponding to the target structured data from the target knowledge base built based on product information, enterprise information, and historical tender documents, and generates target prompt words based on the target matching information and target structured data; finally, it uses a target big model to generate an initial tender document based on the generated target prompt words and preset generation rules. It then obtains revision information for the initial tender document and combines it with the preset tender document format and revision information to determine the final target tender document. Therefore, this application can automatically generate complete tender documents that are formatted correctly, accurate in content, and meet the tender requirements using a target big model, improving the efficiency and quality of tender document generation. In this way, this application can solve the key technical problems of low efficiency, error-proneness, poor content consistency, chaotic version management, and slow response speed inherent in traditional manual compilation methods, thereby improving the quality and competitiveness of tenders. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a method for generating tender documents based on a large model, as provided in this application;

[0051] Figure 2 This application provides a specific flowchart for retrieving matching information;

[0052] Figure 3 This application provides a specific flowchart for optimizing the rules for generating tender documents;

[0053] Figure 4 A flowchart illustrating a specific method for generating tender documents based on a large model, as provided in this application;

[0054] Figure 5 A schematic diagram of a tender document generation device based on a large model provided for this application;

[0055] Figure 6 This application provides a structural diagram of an electronic device. Detailed Implementation

[0056] 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.

[0057] The current field of bid document preparation faces numerous challenges. Firstly, when submitting bids, the required product technical parameters, company qualification certificates, and historical case data are often scattered across multiple unstructured documents, such as PDF manuals, Word technical specifications, and scanned copies. These documents not only vary in format but also lack a unified retrieval mechanism, making information extraction exceptionally cumbersome. Furthermore, manual retrieval is not only inefficient but also prone to missing crucial information, thus affecting bid quality. Secondly, with product iterations and technological updates, the technical parameters of the same product may contain conflicting descriptions in different versions of documents. Relying on manual verification to ensure information consistency has a high error rate, increasing workload and the likelihood of bid rejection. Thirdly, traditional preparation processes often fall short when faced with complex bidding requirements, especially those requiring customized technical solutions. Therefore, this application provides a bid document generation scheme based on a large model, which can improve the efficiency and quality of bid document generation.

[0058] See Figure 1 As shown in the figure, this invention discloses a method for generating tender documents based on a large model, which may include:

[0059] Step S11: Obtain the target tender document and extract the target entity from the target tender document.

[0060] In this embodiment, the first step is to obtain the target tender document and parse its key requirements. Extracting the target entities from the target tender document can include: first, extracting the text content and parsing its page layout structure; then, determining the text and mixed text / image areas based on the page layout structure, identifying table areas within the mixed text / image areas using a preset table line detection method, and identifying signature areas within the mixed text / image areas using a target detection model; then, constructing target table data based on the text content and table areas, constructing target signature data based on the text content and signature areas, and determining target text data corresponding to the text areas based on the text content; finally, performing a compliance check on the target tender document based on the target signature data, and extracting the target entities from the target tender document based on the target text data and target table data. In one implementation, PyMuPDF can be used to extract the PDF text layer content from the target tender document, combined with PDFMiner to parse the page layout structure of the target tender document, thereby achieving accurate separation of text and layout information. Then, the text and mixed text / image areas in the target tender document can be determined. For content containing mixed text and images, the OpenCV (Open Source Computer Vision Library) table line detection algorithm can be integrated to identify table regions, and the YOLOv5 (YOLO, You Only Look Once) model can be deployed to identify signature regions, such as the official seal of the tendering party. Then, the extracted text content can be combined to reconstruct the target table data and target signature data, and the target text data for the plain text regions in the target tender document can be determined. Finally, a compliance check can be performed on the target tender document based on the target signature data to ensure accurate labeling of legally valid areas, and target entities can be extracted from the target text data and target table data.

[0061] It should be noted that the above-mentioned extraction of the target entity from the target tender document based on the target text data and the target table data may include: firstly, determining the target structural features in the target text data and the target table data based on a preset regular rule library, and determining the target position corresponding to the target structural features; then, using a preset machine learning model, extracting the target entity corresponding to the target position in the target text data and the target table data. Specifically, this embodiment can construct a rule entity recognition framework to obtain a preset regular rule library. Based on the regular rule library, matching the structured features such as clause numbers in the tender text data and the target table data, such as / No.\d+\.\d+Article / , the target position corresponding to the target structural features can be located. Then, a fine-tuned machine learning model can be used to identify semantic entities such as technical parameters of the target position in the target text data and the target table data. In addition, this embodiment can establish a three-level clause importance classification system: key clauses containing words such as "must" are highlighted in red, general clauses containing words such as "suggest" are processed in a normal manner, and clauses containing words such as "optional" are output with reduced weight, thereby realizing intelligent priority ranking of demand entities.

[0062] Step S12: Construct a target knowledge graph corresponding to the target tender document based on the target entity, and determine the target structured data corresponding to the target knowledge graph.

[0063] In this embodiment, the above-mentioned construction of the target knowledge graph corresponding to the target tender document based on the target entity, and the determination of the target structured data corresponding to the target knowledge graph, may include: first, using the target entity as a node, and establishing connections between nodes corresponding to different target entities with related relationships based on a preset graph database to obtain the target knowledge graph corresponding to the target tender document; then, using a graph database query language to determine the target structured data corresponding to the target knowledge graph. Specifically, this embodiment can use Neo4j to construct the target knowledge graph corresponding to the tender requirements, and use the parsed target entities, such as technical parameters and commercial terms, as nodes, dynamically connecting them through relationship edges to form a multi-dimensional relational network. Then, the Cypher query language is used to achieve automatic conversion from the target knowledge graph to the target structured data.

[0064] Step S13: Retrieve target matching information corresponding to the target structured data from the target knowledge base, and generate corresponding target prompt words based on the target matching information and the target structured data; the target knowledge base is a knowledge base built based on product information, enterprise information and historical tender documents.

[0065] It should be noted that this embodiment can design an enterprise bidding knowledge management model to integrate and manage structured and unstructured data, such as PDF technical documents and Word solution descriptions. Through natural language processing, image recognition, and metadata extraction technologies, product information, enterprise information, and information from historical bidding documents are automatically parsed, categorized, and structured into a database to form a target knowledge base. This provides a continuously evolving intelligent data foundation to support rapid response and accurate matching in subsequent bidding processes. Accordingly, before retrieving the target matching information corresponding to the target structured data from the target knowledge base, the process may further include: firstly, determining the corresponding first structured data and first unstructured data based on the product information, enterprise information, and historical bidding documents; then storing the first structured data in a target relational database; subsequently, using a target embedding model to determine the first vector corresponding to the first unstructured data and storing the first vector in a target vector database; finally, determining the target knowledge base based on the target relational database and the target vector database, and updating the data in the target knowledge base in real time. In one specific implementation, for structured data, an open-source MySQL database can be used to construct a corresponding relational data model for data storage. Structured data includes, but is not limited to, product technical parameter IDs, product models, parameter names, parameter values, and document versions. For unstructured data, an embedding model can be used to determine the corresponding vectors, which are then stored in a Milvus vector database. It should be noted that the knowledge base has a dynamic update mechanism. For structured data, MySQL triggers can be used to capture data changes in real time. For unstructured data, a file system monitoring service, such as Python Watchdog, can be used to monitor changes in unstructured data, and the Milvus vector database can be used to achieve real-time updates of the vector index through incremental index construction and NRT (Near Real-Time) retrieval mechanisms.

[0066] In this embodiment, an intelligent search engine can be built based on RAG (Retrieval-Augmented Generation) technology. This intelligent search engine can achieve high-precision information retrieval across documents, versions, and formats within the target knowledge base. It not only supports keyword search but also performs relevance matching based on contextual semantics, automatically recommending suitable technical parameters, case studies, and compliance documentation. When matching target information from the target knowledge base, a three-level nested JSON Schema can be determined based on the target structured data: project requirements. Technical Specifications Detailed parameters, such as the construction of Data Center A Server Configuration In one specific implementation, the retrieval strategy is as follows: First-level retrieval: keyword matching based on the technical terms of the tender document; Second-level retrieval: vector retrieval based on semantic similarity; Third-level retrieval: qualification certificate validity verification retrieval.

[0067] In one specific implementation, the above-mentioned retrieval of target matching information corresponding to the target structured data from the target knowledge base may include: firstly, based on a Boolean retrieval method and the target entity corresponding to the target structured data, determining second structured data from the target relational database corresponding to the target knowledge base, and determining a second vector from the target vector database corresponding to the target knowledge base; then, using a target embedding model to determine a third vector corresponding to the target structured data, determining the similarity between the second vector and the third vector, and determining the similarity greater than a preset similarity threshold as the target similarity; subsequently, determining the target vector corresponding to the target similarity from the second vector, and determining the second unstructured data corresponding to the target vector; finally, sorting the second structured data and the second unstructured data based on a preset dynamic weight algorithm, and determining the target matching information corresponding to the target structured data based on the sorting result. Specifically, see [link to documentation]. Figure 2 As shown, keyword matching content can be quickly filtered based on Elasticsearch's Boolean search. In one specific implementation, second structured data matching the target entity can be retrieved from the target relational database, thereby filtering out product models, parameter IDs, and corresponding document versions that meet the conditions. Simultaneously, corresponding second vectors can be retrieved from the target vector database, the Milvus vector library. Then, an embedding model can be used to map sentences or paragraphs in the target structured data to a vector space, obtaining a third vector. By calculating the similarity between the second and third vectors, a target similarity greater than a preset similarity threshold is determined, and second unstructured data is determined based on this target similarity. Finally, a dynamic weighting algorithm can be designed by combining semantic relevance, keyword matching degree, and data timeliness. Based on this dynamic weighting algorithm, the second structured data and the second unstructured data are sorted, forming a progressively precise retrieval pipeline that ensures the returned results are both comprehensive and accurate.

[0068] It should be noted that after retrieving the target matching information corresponding to the target structured data from the target knowledge base, the process may further include: if the target matching information is abnormal, generating corresponding alarm information; based on the alarm information and the target knowledge graph corresponding to the target structured data, locating the abnormal node corresponding to the abnormal information, and determining the abnormal entity corresponding to the abnormal node; generating a corresponding pending review report based on the abnormal entity and the abnormal information, and obtaining the processing log and processing result corresponding to the pending review report; and uploading the abnormal information, the processing log, and the processing result to the blockchain so that the abnormal information can be traced based on the blockchain. Specifically, when a significant difference is identified between the target matching information and the target structured data, such as a product parameter version deviation exceeding 5%, an alarm can be triggered and a knowledge graph-assisted multi-source evidence comparison can be initiated to locate the abnormal node corresponding to the abnormal information from the target knowledge graph and determine the abnormal entity corresponding to the abnormal node. After accurately locating the root cause of the conflict, a pending review report is generated and pushed to the human interface. Simultaneously, the processing logs and results corresponding to the reports to be reviewed can be determined. Finally, the complete anomaly information, processing logs, and processing results are uploaded to the blockchain so that the entire process of conflict tracking from discovery to resolution can be carried out based on the blockchain.

[0069] In this embodiment, corresponding target prompts, or Prompts, can be generated based on target matching information and target structured data. Specifically, prompts can be formulated according to the technical solution writing, response key point organization, advantage highlight extraction, and compliance statement, as shown below:

[0070] # Character Setting

[0071] As a senior bidding expert, you must strictly adhere to the "Tender Document No.: {{doc_id}}" and the "industry_standard}}" standards.

[0072] # Search Context

[0073] 1. Key requirements for the tender: {{requirement_summary}}

[0074] 2. Match product parameters: {{product_specs}} (Source: Knowledge Base ID#{{kb_id}})

[0075] 3. Implementation Case Reference: {{similar_case}} ({{project_size}} scale project)

[0076] # Generation Requirements

[0077] **Structured Output**:

[0078] 1. Adopt a three-stage structure: "Pain Point Analysis → Solution → Technical Architecture"

[0079] 2. Compare key technical parameters with the bidding requirements using a table (columns: bidding item / our response / deviation rate).

[0080] 3. Embedding topology graphs / sequence diagrams (Mermaid syntax)

[0081] # Task Command

[0082] Transform the {{n}} core requirements in {{requirement_list}} into a response point tree, and sort them in descending order of weight.

[0083] # Organizational Rules

[0084] **Priority Strategy**:

[0085] - Needs that directly impact the score (accounting for ≥15% of the total score) are placed in Tier 1.

[0086] - Requirements containing words like "must" or "mandatory" are prioritized.

[0087] - Business terms are sorted by "Qualifications → Performance → Services"

[0088] **Presentation Format**:

[0089] ```markdown

[0090] 1. Requirement {{req_id}}: {{original_text}}

[0091] Response strategy: {{strategy}}

[0092] Supporting evidence: {{cert_id}} (valid before {{expire_date}})

[0093] Explanation of differences: {{deviation}} (<{{threshold}}%)

[0094] ### **Highlights and Strengths Summary Prompt**

[0095] ```prompt

[0096] # Refining the Framework

[0097] **Differentiated Positioning Matrix**:

[0098] | Dimensions | Common weaknesses of competitors | Our advantages |

[0099] |------------|-------------------|-------------------|

[0100] | Technical Performance | Response latency ≥ 5ms | **Self-developed protocol ≤ 1ms** |

[0101] | Cost Control | Maintenance Fee Rate 12% | **AI-Powered Maintenance Reduced to 8%** |

[0102] **Highlighting Requirements**:

[0103] 1. Data Support: Attached test report ({{test_report_no}})

[0104] 2. Patent Citation: Highlighting the core technology of {{patent_id}}

[0105] 3. Quantitative comparison: Performance indicators improved by {{x}}% (e.g., energy efficiency ratio increased by 30%)

[0106] # Compliance Engine

[0107] json

[0108] {

[0109] "Mandatory Clause": [

[0110] "Responding to {{clause}} clauses in the tender documents",

[0111] "Complies with Chapter {{standard}}"

[0112] ],

[0113] "Risk avoidance": [

[0114] "Avoid price discrimination statements",

[0115] "Disallowing Exclusivity Clauses" ]

[0117] }

[0118] Step S14: Utilize the target big model to generate the corresponding initial bid document based on the target prompt words and preset generation rules, obtain the revision information corresponding to the initial bid document, and determine the target bid document corresponding to the initial bid document based on the preset bid document format and the revision information.

[0119] In this embodiment, a target-oriented large model combined with target prompts can be used to automatically generate tender documents. This mainly includes writing technical solutions, organizing key response points, extracting advantages and highlights, and making compliance statements. It supports generating complete chapters from scratch or optimizing and polishing existing drafts. By introducing a template engine and logical verification mechanisms, it ensures that the generated content not only meets the bidding requirements but also reflects the company's core competitiveness, improving both efficiency and consistency in quality.

[0120] It should be noted that after obtaining the revision information corresponding to the initial bid document, the process may further include: firstly, based on the revision information, performing a difference analysis on the revised initial bid document and the original initial bid document, and generating corresponding difference analysis results; then, updating the target prompt words based on the difference analysis results, and optimizing the preset generation rules based on the updated target prompt words, so as to utilize the target large model to generate the bid document based on the optimized preset generation rules. For details, see [link to documentation]. Figure 3 As shown, the initially generated bid documents can be manually modified and improved to obtain revision information. Then, a difference analysis can be performed, comparing the versions before and after modification to identify areas for improvement and generate corresponding difference analysis results. Next, the target prompts are updated based on the difference analysis results to guide the subsequent intelligent generation process. Finally, the generation rules can be optimized based on the updated target prompts to ensure that the large model can more accurately understand and apply the prompts. In this way, this step can significantly improve the quality of subsequent bid document generation, forming a virtuous cycle of continuous iteration and improvement.

[0121] It should be noted that the revised initial tender documents can be automatically assembled and standardized in format using an intelligent formatting engine. Specifically, by integrating core content such as technical solutions, business responses, and advantage statements, standardized enterprise templates can be applied for automatic formatting, and dynamically generated charts, indexes, and other elements can be embedded to ultimately output a tender document that meets the bidding format requirements. That is, for the revised initial tender documents, the Jinja2 template engine can be used to inject the revised initial tender documents, business responses, and other content into the target tender document and automatically populate parameter tables. Then, based on the CSS3 (Cascading Style Sheets) style inheritance system, precise format control can be achieved, automatically generating multi-level heading numbering, such as 1.1.1, and cross-references for charts, such as "Figure X System Architecture," ensuring that fonts, line spacing, page numbers, etc., meet the bidding document format requirements, thereby establishing a complete formatting specification from chapter titles and table styles to cross-references, resulting in the target tender document.

[0122] In one specific implementation, see Figure 4As shown, the specific process of the bid document generation method based on the large model can be as follows: upload the bidding documents and parse the key requirements in the bidding documents, then use the RAG engine to retrieve the matching content corresponding to the key requirements in the product technical parameters, company qualification certificates and historical bidding cases in the knowledge base, then use the large model to generate the initial bid draft, and obtain the standard bid document by standardizing the format of the initial bid draft, and output the standard bid document.

[0123] As can be seen from the above, in this embodiment, the target tender document is first obtained, and the target entities in the target tender document are extracted; then, a target knowledge graph corresponding to the target tender document is constructed based on the target entities, and the target structured data corresponding to the target knowledge graph is determined; subsequently, target matching information corresponding to the target structured data is retrieved from the target knowledge base, and corresponding target prompt words are generated based on the target matching information and the target structured data; the target knowledge base is a knowledge base constructed based on product information, enterprise information, and historical tender documents; finally, the target big model is used to generate the corresponding initial tender document based on the target prompt words and preset generation rules, and the revision information corresponding to the initial tender document is obtained, and the target tender document corresponding to the initial tender document is determined based on the preset tender document format and the revision information. As can be seen from the above, this embodiment first extracts target entity information from the target tender document; then, it constructs a target knowledge graph based on the target entity information and determines the target structured data corresponding to the target knowledge graph; next, it retrieves target matching information corresponding to the target structured data from the target knowledge base constructed based on product information, enterprise information, and historical tender documents, and generates target prompt words based on the target matching information and target structured data; finally, it uses a target big model to generate an initial tender document based on the generated target prompt words and preset generation rules. It obtains revision information for the initial tender document and then combines it with the preset tender document format and revision information to determine the final target tender document. Therefore, this embodiment can automatically generate complete tender documents that are formatted correctly, accurate in content, and meet the tender requirements using a target big model, improving the efficiency and quality of tender document generation. In this way, this embodiment can solve the key technical problems of low efficiency, error-proneness, poor content consistency, chaotic version management, and slow response speed inherent in traditional manual compilation methods, thereby improving the quality and competitiveness of tenders.

[0124] Accordingly, see Figure 5 As shown in the embodiments of this application, a tender document generation device based on a large model is also provided, which may include:

[0125] The target entity extraction module 11 is used to obtain the target tender document and extract the target entities from the target tender document;

[0126] The target structured data determination module 12 is used to construct a target knowledge graph corresponding to the target tender document based on the target entity, and determine the target structured data corresponding to the target knowledge graph;

[0127] The target matching information retrieval module 13 is used to retrieve target matching information corresponding to the target structured data from the target knowledge base, and generate corresponding target prompt words based on the target matching information and the target structured data; the target knowledge base is a knowledge base built based on product information, enterprise information and historical tender documents;

[0128] The target bid document determination module 14 is used to generate a corresponding initial bid document based on the target prompt words and preset generation rules using the target big model, obtain the revision information corresponding to the initial bid document, and determine the target bid document corresponding to the initial bid document based on the preset bid document format and the revision information.

[0129] As can be seen from the above, this application first obtains the target tender document and extracts the target entities from the target tender document; then, it constructs a target knowledge graph corresponding to the target tender document based on the target entities and determines the target structured data corresponding to the target knowledge graph; subsequently, it retrieves target matching information corresponding to the target structured data from the target knowledge base and generates corresponding target prompt words based on the target matching information and the target structured data; the target knowledge base is a knowledge base constructed based on product information, enterprise information, and historical tender documents; finally, it uses the target big model to generate the corresponding initial tender document based on the target prompt words and preset generation rules, obtains the revision information corresponding to the initial tender document, and determines the target tender document corresponding to the initial tender document based on the preset tender document format and the revision information. As can be seen from the above, this application first extracts target entity information from the target tender document; then, it constructs a target knowledge graph based on the target entity information and determines the target structured data corresponding to the target knowledge graph; next, it retrieves target matching information corresponding to the target structured data from the target knowledge base built based on product information, enterprise information, and historical tender documents, and generates target prompt words based on the target matching information and target structured data; finally, it uses a target big model to generate an initial tender document based on the generated target prompt words and preset generation rules. It then obtains revision information for the initial tender document and combines it with the preset tender document format and revision information to determine the final target tender document. Therefore, this application can automatically generate complete tender documents that are formatted correctly, accurate in content, and meet the tender requirements using a target big model, improving the efficiency and quality of tender document generation. In this way, this application can solve the key technical problems of low efficiency, error-proneness, poor content consistency, chaotic version management, and slow response speed inherent in traditional manual compilation methods, thereby improving the quality and competitiveness of tenders.

[0130] In some specific embodiments, the target entity extraction module 11 may include:

[0131] The page layout structure parsing submodule is used to extract the text content in the target tender document and parse the page layout structure of the target tender document;

[0132] The region identification submodule is used to determine the text region and the mixed text and image region in the target tender document based on the page layout structure, and to identify the table region in the mixed text and image region based on a preset table line detection method, and to identify the signature region in the mixed text and image region using a target detection model;

[0133] The target text data determination submodule is used to construct target table data based on the text content and the table area, construct target signature data based on the text content and the signature area, and determine the target text data corresponding to the text area based on the text content.

[0134] The target entity extraction submodule is used to perform compliance checks on the target tender document based on the target signature data, and to extract the target entity from the target tender document based on the target text data and the target table data.

[0135] In some specific implementations, the target entity extraction submodule may include:

[0136] The target location determination unit is used to determine the target structural features in the target text data and the target table data based on a preset regular expression rule library, and to determine the target location corresponding to the target structural features;

[0137] The target entity extraction unit is used to extract the target entity corresponding to the target position in the target text data and the target table data using a preset machine learning model;

[0138] Accordingly, the target structured data determination module 12 may include:

[0139] The target knowledge graph determination unit is used to take the target entity as a node and establish connections between nodes corresponding to different target entities with related relationships based on a preset graph database to obtain the target knowledge graph corresponding to the target tender document.

[0140] The target structured data determination unit is used to determine the target structured data corresponding to the target knowledge graph using a graph database query language.

[0141] In some specific embodiments, the large-model-based tender document generation device may further include:

[0142] The first structured data determination module is used to determine the corresponding first structured data and first unstructured data based on the product information, the enterprise information, and the historical tender documents;

[0143] The first structured data storage module is used to store the first structured data into the target relational database;

[0144] The first vector determination module is used to determine the first vector corresponding to the first unstructured data using a target embedding model, and store the first vector in the target vector database;

[0145] The target knowledge base determination module is used to determine the target knowledge base based on the target relational database and the target vector database, and to update the data in the target knowledge base in real time.

[0146] In some specific embodiments, the target matching information retrieval module 13 may include:

[0147] The second vector determination unit is used to determine the second structured data from the target relational database corresponding to the target knowledge base based on the Boolean retrieval method and the target entity corresponding to the target structured data, and to determine the second vector from the target vector database corresponding to the target knowledge base;

[0148] The target similarity determination unit is used to determine the third vector corresponding to the target structured data using a target embedding model, and to determine the similarity between the second vector and the third vector, and to determine the similarity greater than a preset similarity threshold as the target similarity.

[0149] A target vector determination unit is used to determine the target vector corresponding to the target similarity from the second vector, and to determine the second unstructured data corresponding to the target vector;

[0150] The target matching information retrieval unit is used to sort the second structured data and the second unstructured data based on a preset dynamic weight algorithm, and to determine the target matching information corresponding to the target structured data based on the sorting result.

[0151] In some specific embodiments, the large-model-based tender document generation device may further include:

[0152] An alarm information generation module is used to generate corresponding alarm information if the target matching information is abnormal.

[0153] An abnormal entity determination module is used to locate the abnormal node corresponding to the abnormal information and determine the abnormal entity corresponding to the abnormal node based on the alarm information and the target knowledge graph corresponding to the target structured data.

[0154] The pending review report generation module is used to generate a corresponding pending review report based on the abnormal entity and the abnormal information, and to obtain the processing log and processing result corresponding to the pending review report;

[0155] An anomaly information uploading module is used to upload the anomaly information, the processing log, and the processing result to the blockchain so that the anomaly information can be traced based on the blockchain.

[0156] In some specific embodiments, the large-model-based tender document generation device may further include:

[0157] The difference analysis result generation module is used to perform a difference analysis on the revised initial bid document and the original initial bid document based on the revision information, and generate the corresponding difference analysis results;

[0158] The target prompt word update module is used to update the target prompt words based on the difference analysis results, and optimize the preset generation rules based on the updated target prompt words, so as to generate the tender document using the target big model based on the optimized preset generation rules.

[0159] Furthermore, embodiments of this application also disclose an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the large-model-based tender document generation method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0160] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0161] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0162] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the large-model-based tender document generation method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0163] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for generating tender documents based on a large model. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0164] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0165] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0167] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0168] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for generating tender documents based on a large model, characterized in that, include: Obtain the target tender document and extract the target entity from the target tender document; Based on the target entity, construct the target knowledge graph corresponding to the target tender document, and determine the target structured data corresponding to the target knowledge graph; Retrieve target matching information corresponding to the target structured data from the target knowledge base, and generate corresponding target prompt words based on the target matching information and the target structured data; The target knowledge base is a knowledge base built based on product information, enterprise information, and historical tender documents; The target large model is used to generate the corresponding initial bid document based on the target prompt words and preset generation rules, and the revision information corresponding to the initial bid document is obtained. The target bid document corresponding to the initial bid document is determined based on the preset bid document format and the revision information.

2. The method for generating tender documents based on a large model according to claim 1, characterized in that, The extraction of the target entity from the target tender document includes: Extract the text content from the target tender document and parse the page layout structure of the target tender document; Based on the page layout structure, the text area and the mixed text and image area in the target tender document are determined, and the table area in the mixed text and image area is identified based on the preset table line detection method, and the signature area in the mixed text and image area is identified using the target detection model. Target table data is constructed based on the text content and the table area, target signature data is constructed based on the text content and the signature area, and target text data corresponding to the text area is determined based on the text content. The target tender document is subjected to compliance checks based on the target signature data, and the target entity in the target tender document is extracted based on the target text data and the target table data.

3. The method for generating tender documents based on a large model according to claim 2, characterized in that, The step of extracting the target entity from the target tender document based on the target text data and the target table data includes: Based on a preset regular expression rule base, the target structural features in the target text data and the target table data are determined, and the target positions corresponding to the target structural features are determined. Using a preset machine learning model, extract the target entity corresponding to the target position in the target text data and the target table data; Accordingly, the step of constructing the target knowledge graph corresponding to the target tender document based on the target entity, and determining the target structured data corresponding to the target knowledge graph, includes: The target entity is used as a node, and a connection is established between nodes corresponding to different target entities with related relationships based on a preset graph database to obtain the target knowledge graph corresponding to the target tender document. The target structured data corresponding to the target knowledge graph is determined using a graph database query language.

4. The method for generating tender documents based on a large model according to claim 1, characterized in that... Before retrieving the target matching information corresponding to the target structured data from the target knowledge base, the method further includes: Based on the product information, the enterprise information, and the historical tender documents, the corresponding first structured data and first unstructured data are determined; Store the first structured data in the target relational database; The first vector corresponding to the first unstructured data is determined using a target embedding model, and the first vector is stored in the target vector database; The target knowledge base is determined based on the target relational database and the target vector database, and the data in the target knowledge base is updated in real time.

5. The method for generating tender documents based on a large model according to claim 4, characterized in that, The step of retrieving target matching information corresponding to the target structured data from the target knowledge base includes: Based on the Boolean retrieval method and the target entity corresponding to the target structured data, the second structured data is determined from the target relational database corresponding to the target knowledge base, and the second vector is determined from the target vector database corresponding to the target knowledge base; The third vector corresponding to the target structured data is determined using a target embedding model, and the similarity between the second vector and the third vector is determined. The similarity greater than a preset similarity threshold is determined as the target similarity. Determine the target vector corresponding to the target similarity from the second vector, and determine the second unstructured data corresponding to the target vector; The second structured data and the second unstructured data are sorted based on a preset dynamic weighting algorithm, and the target matching information corresponding to the target structured data is determined based on the sorting result.

6. The method for generating tender documents based on a large model according to claim 1, characterized in that, After retrieving the target matching information corresponding to the target structured data from the target knowledge base, the method further includes: If the target matching information is abnormal, then a corresponding alarm message is generated; Based on the alarm information and the target knowledge graph corresponding to the target structured data, locate the abnormal node corresponding to the abnormal information and determine the abnormal entity corresponding to the abnormal node; Based on the abnormal entity and the abnormal information, a corresponding report to be reviewed is generated, and the processing log and processing result corresponding to the report to be reviewed are obtained; The abnormal information, the processing log, and the processing result are uploaded to the blockchain so that the abnormal information can be traced based on the blockchain.

7. The method for generating tender documents based on a large model according to any one of claims 1 to 6, characterized in that, After obtaining the revision information corresponding to the initial bid document, the process further includes: Based on the revision information, a difference analysis is performed on the revised initial tender document and the original initial tender document, and corresponding difference analysis results are generated. The target prompts are updated based on the difference analysis results, and the preset generation rules are optimized based on the updated target prompts, so as to generate the tender documents using the target large model based on the optimized preset generation rules.

8. A tender document generation device based on a large model, characterized in that, include: The target entity extraction module is used to obtain the target tender document and extract the target entities from the target tender document; The target structured data determination module is used to construct a target knowledge graph corresponding to the target tender document based on the target entity, and to determine the target structured data corresponding to the target knowledge graph; The target matching information retrieval module is used to retrieve target matching information corresponding to the target structured data from the target knowledge base, and generate corresponding target prompt words based on the target matching information and the target structured data; The target knowledge base is a knowledge base built based on product information, enterprise information, and historical tender documents; The target bid document determination module is used to generate a corresponding initial bid document based on the target prompt words and preset generation rules using the target big model, obtain the revision information corresponding to the initial bid document, and determine the target bid document corresponding to the initial bid document based on the preset bid document format and the revision information.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the large model-based tender document generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the bid document generation method based on a large model as described in any one of claims 1 to 7.

Images