Bid invitation file automatic generation method and system based on large model

Abstract

The invention discloses an automatic bid invitation file generation method and system based on a large model, and the method comprises the steps: obtaining and analyzing a bid invitation demand document, and determining a plurality of document logic units in the bid invitation demand document; analyzing each document logic unit to determine a key semantic entity, and analyzing the key semantic entity to determine a core demand of each document logic unit; retrieving a template knowledge framework matched with the core demand from a preset knowledge database based on the core demand, and constructing a retrieval data set based on the template knowledge framework and the core demand; the retrieval data set is input into a preset language large model for text output, and an initial bid invitation file is obtained; and verifying and optimizing the content of the generated initial bid invitation file to obtain a formal bid invitation file. According to the method, the big language model serves as an intelligent brain and is combined with a professional bid invitation knowledge base and a rule engine, the intention of a user can be accurately understood, high-quality and compliant texts are generated, and professionals are liberated from tedious and repeated document work.

Description

Technical Field

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

[0002] In today's digital and intelligent era, government procurement and corporate bidding activities are becoming increasingly frequent. As the guiding document for the entire procurement process, the quality of the bidding documents directly affects the success or failure of the project.

[0003] However, traditional bidding document preparation methods rely heavily on manual work by professionals, resulting in inherent pain points such as low efficiency, high cost, inconsistent quality, and difficulty in standardizing processes. Preparers need to spend a significant amount of time communicating repeatedly to clarify requirements, consulting vast amounts of laws and regulations to avoid omissions, and drafting and adjusting complex technical and commercial clauses based on their personal experience. This entire process is not only time-consuming but also highly susceptible to human error, omissions, or ambiguities, posing potential legal and performance risks. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for automatically generating tender documents based on a large model, comprising:

[0005] Obtain the tender requirements document and perform document analysis to identify multiple logical units within the tender requirements document;

[0006] Semantic parsing is performed on each document logical unit to identify key semantic entities, and the key semantic entities are analyzed to determine the core requirements of each document logical unit.

[0007] Based on the core requirements, a template knowledge framework matching the core requirements is retrieved from a pre-set knowledge database, and a retrieval dataset is constructed based on the template knowledge framework and the core requirements.

[0008] The retrieved datasets of each document logical unit are input into the preset language model, and the text is output based on the preset language model to obtain the initial tender document;

[0009] The content of the generated initial tender documents is verified and optimized to obtain the final, compliant formal tender documents.

[0010] Furthermore, the step of obtaining the bidding requirements document and performing document analysis on the bidding requirements document to determine multiple logical document units within the bidding requirements document includes:

[0011] Obtain the bidding requirements document and preprocess it;

[0012] Extract the basic textual structure information from the preprocessed tender requirements document, and divide the tender requirements document into multiple sentence segments based on the basic textual structure information;

[0013] Each sentence segment is represented by a vector to obtain a sentence segment vector. The cosine similarity between adjacent sentence segment vectors is calculated, and adjacent sentence segments with a cosine similarity lower than a preset first threshold are identified.

[0014] The boundary between adjacent sentences is defined as the boundary of the logical unit, and the tender requirement document is divided into multiple document logical units based on the boundary of the logical unit.

[0015] Furthermore, the semantic parsing of each document logical unit to determine key semantic entities includes:

[0016] Natural language processing techniques are used to extract semantic meaning information of each sentence segment from the logical units of a document, and semantic entities in each sentence segment are determined based on the semantic meaning information.

[0017] The frequency, position, and entity type of semantic entities in the sentence are determined, and the criticality of semantic entities is evaluated based on the frequency, position, and entity type to obtain the criticality evaluation value of each semantic entity.

[0018] Semantic entities whose criticality assessment value is greater than a preset first threshold are identified as key semantic entities in the document's logical unit.

[0019] Furthermore, the evaluation of the criticality of semantic entities based on frequency, location, and entity type yields a criticality evaluation value for each semantic entity, including:

[0020] Determine a pre-set baseline frequency, calculate the difference between the frequency of the semantic entity appearing in the sentence and the baseline frequency, evaluate the difference, and obtain the first evaluation value of the semantic entity.

[0021] The semantic entity is evaluated based on its position and entity type in the sentence, respectively, to obtain the second and third evaluation values ​​of the semantic entity.

[0022] The first, second, and third evaluation values ​​of the semantic entity are added together to obtain the semantic entity's criticality evaluation value.

[0023] Furthermore, the analysis of key semantic entities to determine the core requirements of each document logical unit includes:

[0024] Extract the entity relationships between key semantic entities in the sentence, and extract the requirement elements from the key semantic entities and entity relationships;

[0025] Fill in the demand elements according to the preset demand template to obtain the preliminary demand of each sentence, and extract the core semantic meaning of the preliminary demand of each sentence.

[0026] By combining the core semantic meanings of the initial requirements of each sentence, we can obtain the core requirements of the document's logical units.

[0027] Furthermore, the step of retrieving template knowledge frameworks matching the core requirements from a pre-set knowledge database, and constructing a retrieval dataset based on the template knowledge frameworks and the core requirements, includes:

[0028] Extract keywords from the core requirements and represent these keywords as vectors to obtain keyword vectors;

[0029] Calculate the relevance of each keyword to the core theme of the preset requirements, and normalize each relevance to obtain the weight of each keyword;

[0030] The core requirement vector is obtained by fusion calculation based on the weight of each keyword and the keyword vector, and the template clauses in the preset knowledge database are represented by vectors to obtain the template clause vector.

[0031] Calculate the cosine similarity between the core requirement vector and each template clause vector, and determine the preset template clause with the highest cosine similarity as the template knowledge framework that matches the core requirement;

[0032] The core requirements are mapped to the corresponding template knowledge frameworks, and a retrieval dataset is built based on the mapping relationship.

[0033] Furthermore, the formula for calculating the core requirement vector is as follows:

[0034]

[0035] Where E is the core demand vector, αi is the weight of the i-th keyword, P is the keyword vector of the i-th keyword, and n is the number of keywords.

[0036] Furthermore, the step of inputting the retrieval dataset of each document logical unit into a preset language model, and outputting text based on the preset language model to obtain the initial tender document, includes:

[0037] Design corresponding prompt words for the core requirements and corresponding template knowledge frameworks in the retrieval dataset of each document logical unit;

[0038] Based on the prompt words, the retrieval dataset of each document's logical unit is input into the preset language model, and the text is output through the preset language model to obtain the bidding chapter content of each document's logical unit;

[0039] The tender section content of each document's logical unit is assembled according to the structure of a standard tender document to form an initial tender document.

[0040] Furthermore, the process of verifying and optimizing the content of the generated initial tender document to obtain the final, compliant formal tender document includes:

[0041] Determine each tender clause in the generated initial tender document and verify the format, completeness, compliance, and consistency of each tender clause;

[0042] The selected tender terms that passed verification were filtered out, and the selected tender terms were optimized based on the reasons for failure to pass verification.

[0043] The optimized bidding terms are then matched to the corresponding content to obtain the final, compliant official bidding documents.

[0044] This invention also provides an automatic tender document generation system based on a large model, comprising:

[0045] The acquisition module is used to acquire the bidding requirements document and perform document analysis on the bidding requirements document to identify multiple document logical units in the bidding requirements document;

[0046] The analysis module is used to perform semantic parsing on each document logical unit, identify key semantic entities, and analyze the key semantic entities to determine the core requirements of each document logical unit.

[0047] The matching module is used to retrieve template knowledge frameworks that match the core requirements from a pre-set knowledge database, and to construct a retrieval dataset based on the template knowledge frameworks and the core requirements.

[0048] The generation module is used to input the retrieved datasets of each document logical unit into the preset language model, and output the text based on the preset language model to obtain the initial tender document;

[0049] The optimization module is used to verify and optimize the content of the generated initial tender documents to obtain the final, compliant formal tender documents.

[0050] Compared with existing technologies, the advantages of the automatic generation method and system for tender documents based on large models in this invention are as follows:

[0051] This invention completely changes the inefficient traditional manual bidding document writing mode, compressing the document preparation work that originally required several days or even weeks to be completed in hours. Through the intelligent document parsing engine, it can automatically identify and understand the deep logical structure of the bidding requirements document, accurately decompose complex long documents into multiple logical units, and avoid omissions and misjudgments that may occur in manual analysis.

[0052] This invention employs multi-level semantic parsing technology, which can not only identify key semantic entities on the surface, but also deeply understand the correlation between technical parameters, the logical relationship of constraints, and the implicit meaning of business requirements. Through intelligent analysis of key semantic entities, it can accurately capture the core demands of customers and transform vague business requirements into clear and executable technical specifications, effectively solving the information distortion problem commonly found in traditional requirements analysis.

[0053] This invention can accurately match the most suitable standard terms and expression frameworks for each core requirement, which not only conforms to industry best practices, but also meets the personalized needs of specific projects. Based on this, the large language model generates creative content, producing text content that is not only grammatically correct, but also has professional depth and logical rigor.

[0054] This invention establishes a multi-level automatic verification mechanism that conducts a comprehensive review from multiple dimensions. Through optimization, it can automatically correct problems and ensure that the final output of the tender documents meets the standards of professional tendering agencies.

[0055] This invention successfully solves the core contradiction in the preparation of bidding documents—both adhering to standards and specifications and reflecting the characteristics of the project. The generated bidding documents contain the necessary standardized clauses to ensure compliance, while also allowing for flexible adjustment of technical parameters and business conditions according to specific needs, thus achieving an organic unity between large-scale efficiency and personalized customization. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the process structure of the automatic generation method for tender documents based on a large model in an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the composition of the automatic tender document generation system based on a large model in an embodiment of the present invention. Detailed Implementation

[0058] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0059] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the platform or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0060] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0061] like Figure 1 As shown in the embodiments of this application, an automatic tender document generation method based on a large model is provided, including: S100: obtaining a tender requirement document and performing document analysis on the tender requirement document to determine multiple document logical units in the tender requirement document; S200: performing semantic parsing on each document logical unit to determine key semantic entities, and analyzing the key semantic entities to determine the core requirements of each document logical unit; S300: retrieving a template knowledge framework matching the core requirements from a preset knowledge database based on the core requirements, and constructing a retrieval dataset based on the template knowledge framework and the core requirements; S400: inputting the retrieval dataset of each document logical unit into a preset language large model, and outputting text based on the preset language large model to obtain an initial tender document; S500: verifying and optimizing the content of the generated initial tender document to obtain a final formal tender document that meets the requirements.

[0062] Furthermore, this invention completely changes the inefficient traditional manual drafting of tender documents, compressing the document preparation work that originally required days or even weeks to be completed in hours. Through an intelligent document parsing engine, it can automatically identify and understand the deep logical structure of tender requirements documents, accurately decomposing complex, long documents into multiple logical units, avoiding omissions and misjudgments that may occur during manual analysis. This invention employs multi-level semantic parsing technology, which can not only identify surface-level key semantic entities but also deeply understand the correlation between technical parameters, the logical relationships of constraints, and the implicit meaning of business requirements. Through intelligent analysis of key semantic entities, it can accurately capture the core demands of customers, transforming vague business requirements into clear and executable technical specifications, effectively solving the information distortion problem commonly found in traditional requirements analysis. This invention can... This invention can accurately match the most suitable standard clauses and expression frameworks for each core requirement, conforming to industry best practices while meeting the personalized needs of specific projects. Based on this, the large language model generates creative content, producing text content that is not only grammatically correct but also possesses professional depth and logical rigor. This invention establishes a multi-level automatic verification mechanism, conducting a comprehensive review from multiple dimensions. Through optimization, it can automatically correct problems, ensuring that the final output bidding documents meet the standards of professional bidding agencies. This invention successfully solves the core contradiction in bidding document preparation—both adhering to standards and specifications and reflecting project characteristics. The generated bidding documents include necessary standardized clauses to ensure compliance, while also allowing for flexible adjustment of technical parameters and business conditions according to specific needs, achieving an organic unity between scalability and personalized customization.

[0063] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The method involves obtaining a tender requirement document and performing document analysis on the tender requirement document to determine multiple document logical units within the tender requirement document. The method includes: obtaining the tender requirement document and preprocessing it; extracting textual basic structure information from the preprocessed tender requirement document and dividing the tender requirement document into multiple segments based on the textual basic structure information; representing each segment with a vector to obtain segment vectors; calculating the cosine similarity between adjacent segment vectors and determining adjacent segments with a cosine similarity lower than a preset first threshold; determining the boundary of the adjacent segments as the boundary of the logical unit, and dividing the tender requirement document into multiple document logical units based on the boundaries of the logical units.

[0064] Specifically, the original bidding requirements document undergoes preprocessing, including text cleaning, format standardization, and encoding unification, to provide clean and standardized text data for subsequent analysis. Based on this, the document's basic textual structure information, such as paragraph marks, heading levels, and layout features, is extracted to form a preliminary understanding of the document's organizational framework. Based on this structural information, the document is intelligently divided into multiple coherent sentence units, each representing a relatively complete semantic fragment. Each sentence unit is transformed into a high-dimensional vector representation using an advanced text embedding model. By calculating the cosine similarity between adjacent sentence unit vectors, the semantic association strength between them is accurately quantified. When the cosine similarity between adjacent sentence unit vectors is detected to be lower than a preset first threshold, this point is identified as a semantic topic transition point and determined as the boundary of a logical unit. This process adapts to different document structures and writing styles, accurately identifying different logical units such as technical specifications, business requirements, qualification standards, and terms of service, even if these units are not explicitly marked. In terms of efficiency, it achieves automated and detailed parsing of bidding documents, compressing the analysis work that would take hours manually into minutes. In terms of quality, it ensures the accuracy of logical unit division through deep semantic understanding, laying a reliable foundation for subsequent requirement extraction and document generation. In terms of adaptability, it can handle bidding requirement documents of various formats and styles, and has strong generalization ability and practical value.

[0065] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The step of performing semantic parsing on each document logical unit to determine key semantic entities includes: extracting semantic meaning information of each sentence from the document logical unit using natural language processing technology, and determining semantic entities in each sentence based on the semantic meaning information; determining the frequency, position, and entity type of the semantic entity in the sentence, and evaluating the keyness of the semantic entity based on the frequency, position, and entity type to obtain a keyness evaluation value for each semantic entity; and determining semantic entities with a keyness evaluation value greater than a preset first threshold as key semantic entities in the document logical unit.

[0066] Specifically, advanced natural language processing technology is used to perform deep semantic analysis on the divided logical units of the document. This not only identifies surface-level words and phrases but also delves into the complete semantic meaning and contextual relationships of each sentence. Based on this, semantic entities are accurately extracted from each sentence. These entities cover core conceptual elements in the bidding document, such as technical parameters, product specifications, service requirements, and qualification standards. A comprehensive analysis of the frequency, distribution, and type characteristics of each semantic entity within the sentence is conducted. Through these three dimensions, a precise criticality assessment value is generated for each semantic entity. A dynamic threshold mechanism is employed to classify semantic entities with assessment values ​​higher than a preset first threshold as critical semantic entities. In terms of accuracy, this step effectively avoids misjudgments that might result from a single indicator through multi-dimensional evaluation, ensuring that the extracted entities truly represent the core requirements of the document. In terms of efficiency, it achieves automatic and accurate location of key information points from massive amounts of text, significantly improving the efficiency and consistency of requirements analysis. In terms of practicality, the extracted key semantic entities provide a precise semantic foundation for subsequent template matching, content generation, and compliance checks, enabling the entire bidding document generation process to be built on a foundation of deep understanding.

[0067] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The method for evaluating the criticality of semantic entities based on frequency, position, and entity type to obtain a criticality evaluation value for each semantic entity includes: determining a pre-set baseline frequency, calculating the difference between the frequency of the semantic entity appearing in the sentence and the baseline frequency, and evaluating the difference to obtain a first evaluation value for the semantic entity; evaluating the position and entity type of the semantic entity appearing in the sentence based on position and entity type respectively to obtain a second evaluation value and a third evaluation value for the semantic entity; and adding the first evaluation value, the second evaluation value, and the third evaluation value of the semantic entity to obtain a criticality evaluation value for the semantic entity.

[0068] Specifically, a baseline frequency is pre-set through statistical analysis to provide a scientific reference standard for subsequent evaluation. The difference between the frequency of each semantic entity in the current sentence and the baseline frequency is calculated. This difference calculation not only reflects the absolute number of times the entity appears, but also its significance relative to the normal level. By evaluating the difference, an accurate quantitative first evaluation value is obtained. A refined position weight model is constructed, and hierarchical values ​​are assigned according to the differences in the importance of different positions in semantic expression. Entities in document titles and chapter titles are given the highest weight, entities in the first and last sentences of paragraphs are given a relatively high weight, and entities in the main text are given a basic weight. This hierarchical weighting mechanism ensures that entities occupying strategic positions receive due recognition of their importance, forming an accurate second evaluation value. Based on professional knowledge in the bidding field, a type importance spectrum is constructed, setting core entity types such as technical parameters and performance indicators as the highest level, important types such as business terms and time nodes as the medium level, and auxiliary descriptive entities as the basic level, thereby generating a targeted third evaluation value. The evaluation values ​​of the three dimensions are comprehensively calculated through a weighted summation model to generate a quantitative indicator that comprehensively reflects the criticality of semantic entities. In terms of accuracy assessment, this step effectively overcomes the one-sidedness and bias of single-indicator assessment through cross-validation of three orthogonal dimensions: frequency anomaly, positional importance, and type relevance. In terms of computational scientificity, relative assessment based on benchmark frequency avoids interference from document length and content density on absolute frequency assessment, making assessment results comparable between different documents. In terms of practical application, the quantified criticality assessment value provides accurate decision-making basis for subsequent template matching, requirement prioritization, and content generation, enabling the entire intelligent tender document processing workflow to be built on an objective and quantifiable analytical foundation.

[0069] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The step of analyzing key semantic entities to determine the core requirements of each document logical unit includes: extracting entity relationships between key semantic entities in the sentence and extracting requirement elements from the key semantic entities and entity relationships; filling the requirement elements according to a preset requirement template to obtain the preliminary requirements of each sentence and extracting the core semantic meaning of the preliminary requirements of each sentence; and aggregating the core semantic meanings of the preliminary requirements of each sentence to obtain the core requirements of the document logical unit.

[0070] Specifically, a relation extraction model is used to deeply analyze the semantic relationships between key semantic entities, identifying various entity relationship types, including dependencies between technical parameters, logical connections between business terms, and hierarchical structures between qualification requirements. Isolated key semantic entities and complex entity relationship networks are then integrated and analyzed to extract complete requirement elements. These elements include multiple dimensions such as the requirement subject, requirement action, requirement object, constraints, and modifiers, collectively forming a complete semantic framework for the requirement. These requirement elements are intelligently filled into a pre-defined structured requirement template to generate a preliminary requirement description for each sentence. This process ensures the standardization and completeness of the requirement expression. Through semantic compression and abstraction techniques, redundant modifiers and repetitive information are removed, retaining the most essential semantic core of the requirement description to form a precise core semantic representation. A semantic fusion algorithm is used to organically integrate the core semantic meanings of each sentence within the same logical unit, eliminating repetition and contradictions, supplementing missing relationships, and constructing a comprehensive, consistent, and hierarchically distinct set of core requirements for document logical units. In terms of depth of understanding, this step achieves a leap from isolated concepts to complete semantic networks through entity relationship mining, enabling the capture of complex relationships between demand elements; in terms of expression quality, template-based structured filling ensures the standardization and professionalism of demand descriptions, avoiding the ambiguity and inconsistency of natural language expressions; in terms of information integration, the extraction and fusion of core semantics achieves the sublimation from fragmented information to systematic demands, providing high-quality input for subsequent intelligent retrieval and content generation.

[0071] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The method involves retrieving template knowledge frameworks matching the core requirements from a preset knowledge database, and constructing a retrieval dataset based on the template knowledge frameworks and the core requirements. This includes: extracting keywords from the core requirements and representing them as vectors to obtain keyword vectors; calculating the relevance of each keyword to the core theme of the preset requirements, and normalizing each relevance to obtain the weight of each keyword; performing a fusion calculation based on the weights of each keyword and the keyword vectors to obtain a core requirement vector, and representing each preset template clause in the preset knowledge database as a vector to obtain a template clause vector; calculating the cosine similarity between the core requirement vector and each template clause vector, and determining the preset template clause with the highest cosine similarity as the template knowledge framework matching the core requirements; mapping the core requirements to the corresponding template knowledge frameworks, and constructing a retrieval dataset based on the mapping relationship.

[0072] Specifically, representative keywords are extracted from core needs and transformed into high-dimensional vector representations using advanced word embedding technology, forming a set of keyword vectors capable of capturing deep semantic features. A pre-defined core theme of the need is introduced as a semantic anchor. By calculating the semantic relevance of each keyword to the core theme and normalizing the relevance, the precise weight of each keyword is obtained, reflecting the relative importance of different keywords in expressing the core intent of the need. Multiple keyword vectors are then merged into a single core need vector through weighted averaging. This vector integrates all key semantic information and preserves the complete semantic outline of the need. Template terms from a pre-defined knowledge database are also transformed into vector representations, constructing a template term vector space. By calculating the cosine similarity between the core need vector and each template term vector, precise similarity matching is performed in the semantic space to identify the template knowledge framework that best matches the semantics of the core need. A fine-grained mapping relationship is established between the core need and the matching templates, and a high-quality retrieval dataset is constructed based on this semantic alignment. In terms of matching accuracy, this step overcomes the limitations of traditional keyword matching by calculating deep similarity in the semantic vector space, achieving intelligent semantic alignment between requirements and templates. In terms of efficiency optimization, vectorized representation and similarity calculation enable the system to quickly process massive template libraries and achieve real-time accurate matching. In terms of quality assurance, weight-based vector fusion ensures that the core semantics of the requirements are fully expressed, while cosine similarity measurement guarantees the semantic consistency of the matching results.

[0073] In an embodiment of this application, a method for automatically generating tender documents based on a large model is provided, wherein the calculation formula for the core requirement vector is:

[0074]

[0075] Where E is the core demand vector, αi is the weight of the i-th keyword, P is the keyword vector of the i-th keyword, and n is the number of keywords.

[0076] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The method involves inputting the retrieval datasets of each document logical unit into a preset language large model, and outputting text based on the preset language large model to obtain an initial tender document. This includes: designing corresponding prompt words for the core requirements and corresponding template knowledge frameworks in the retrieval datasets of each document logical unit; inputting the retrieval datasets of each document logical unit into the preset language large model based on the prompt words, and outputting text through the preset language large model to obtain the tender chapter content of each document logical unit; and assembling the tender chapter content of each document logical unit according to the structure of a standard tender document to form the initial tender document.

[0077] Specifically, professional prompts are tailored to the retrieval dataset of each document logical unit. These prompts are meticulously designed, encompassing both the semantic essence of the core requirements and the structured features of the matching template knowledge framework. The prompts employ a layered architecture: the foundation layer clarifies the role and professional requirements of the large model; the middle layer provides the key elements and constraints of the core requirements; and the application layer injects the professional expression style and structured features of the template framework, forming a professional instruction set that accurately guides the generation of the large model. The retrieval datasets of each document logical unit, combined with the corresponding customized prompts, are input into a pre-defined language large model. Based on a deep understanding of the core requirements semantics and the template knowledge framework, the large model generates bidding chapter content that conforms to professional standards. This creative content generation, based on an understanding of the essence of the requirements and the template logic, ensures that the output chapter content accurately reflects the original requirements while conforming to the professional expression standards of bidding documents. Based on a predefined standard bidding document structure template, the chapter content generated by each logical unit is intelligently sorted and organically integrated. The assembly process not only considers the logical order of the chapters but also ensures the coherence and consistency between them, eliminating content duplication or contradictions, and supplementing necessary transitions and connections, ultimately forming a structurally complete and logically clear initial bidding document. In terms of content quality, this step ensures the professionalism and accuracy of the content generated by the large model through refined prompt word design, greatly improving the standardization of the tender documents; in terms of generation efficiency, it processes the content generation of each logical unit in parallel, compressing the traditional manual writing process that takes several days to the minute level; in terms of consistency assurance, intelligent assembly based on standard structure ensures the overall coordination and unity of the document, avoiding style inconsistencies and logical gaps that may occur with manual assembly.

[0078] In the embodiments of this application, a method for automatically generating tender documents based on a large model is provided. The method involves verifying and optimizing the content of the generated initial tender document to obtain a final, compliant formal tender document. This includes: determining each tender clause in the generated initial tender document and verifying the format, completeness, compliance, and consistency of each tender clause; selecting the tender clauses that pass the verification and optimizing the selected tender clauses based on the reasons for failing the verification; and applying corresponding content coverage to the optimized tender clauses to obtain a final, compliant formal tender document.

[0079] Specifically, the generated initial tender documents undergo in-depth analysis, intelligently identifying and extracting various tender clauses, including key content such as technical specifications, commercial requirements, contract terms, and evaluation criteria. Based on this, a multi-level automated verification process is initiated. At the format level, the structural standardization, numbering continuity, and layout consistency of the clauses are checked. At the content completeness level, the completeness of each clause's elements, the sufficiency of its description, and the clarity of its parameters are verified. At the compliance level, the legality and standardization of the clause content are checked against relevant laws, regulations, and industry standards. At the consistency level, the logical connections between different clauses are analyzed to ensure the matching and coordination of technical parameters, commercial conditions, acceptance criteria, and other content. Clauses that pass the verification are selected, and the specific reasons for their failure are analyzed in depth. Based on these reasons, the selected clauses are optimized and improved, strengthening the rigor of key expressions, improving the completeness of detailed descriptions, and enhancing the logical connection between clauses to ensure a comprehensive improvement in document quality. The optimized clauses are then accurately applied to the corresponding positions in the original document. Through intelligent replacement and fusion processing, the overall document's coherence and consistency are ensured, forming the final tender document that meets the requirements for official release. In terms of quality assurance, this step significantly reduces the risk of human error through automated multi-dimensional verification, ensuring the professionalism and compliance of the tender documents. In terms of optimization, the optimization strategy based on positive features makes high-quality clauses more complete, and the overall document quality is systematically improved. In terms of efficiency improvement, the work that traditionally requires multiple people and multiple rounds of review is compressed into an automated process, which greatly shortens the document finalization cycle.

[0080] like Figure 2 As shown in the embodiments of this application, an automatic tender document generation system based on a large model is provided, comprising: an acquisition module for acquiring a tender requirement document and performing document analysis on the tender requirement document to determine multiple document logical units in the tender requirement document; an analysis module for performing semantic parsing on each document logical unit to determine key semantic entities and analyzing the key semantic entities to determine the core requirements of each document logical unit; a matching module for retrieving template knowledge frameworks that match the core requirements from a preset knowledge database and constructing a retrieval dataset based on the template knowledge frameworks and the core requirements; a generation module for inputting the retrieval datasets of each document logical unit into a preset language large model and outputting text based on the preset language large model to obtain an initial tender document; and an optimization module for verifying and optimizing the content of the generated initial tender document to obtain a final formal tender document that meets the requirements.

[0081] In summary, this invention provides a method and system for automatically generating tender documents based on a large-scale model. The method includes: acquiring and analyzing tender requirement documents to identify multiple document logical units; parsing each document logical unit to identify key semantic entities and analyzing these entities to determine the core requirements of each unit; retrieving template knowledge frameworks matching the core requirements from a pre-set knowledge database and constructing a retrieval dataset based on these frameworks and the core requirements; inputting the retrieval dataset into a pre-set language large-scale model for text output to obtain an initial tender document; and verifying and optimizing the content of the generated initial tender document to obtain a formal tender document. This invention uses a large-scale language model as an "intelligent brain," combined with a professional tender knowledge base and rule engine, to accurately understand user intent and generate high-quality, compliant text, freeing professionals from tedious and repetitive paperwork.

[0082] Finally, it should be noted that those skilled in the art can obviously make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims and their equivalents, this invention also intends to include these modifications and variations.

[0083] The above description is merely one embodiment of the present invention, and should not be construed as limiting the scope of the invention. Any structural changes made based on the present invention, as long as they do not depart from the essence of the invention, should be considered as falling within the protection scope of the present invention and subject to its restrictions. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the platform described above can be referred to the corresponding processes in the foregoing platform embodiments, and will not be repeated here.

[0084] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, platform, article, or device / platform that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to those processes, platforms, articles, or devices / platforms.

[0085] The technical solutions of the present invention have been described in conjunction with the accompanying drawings and further embodiments. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to closely related technical features, and the technical solutions resulting from such changes or substitutions will all fall within the scope of protection of the present invention.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for automatically generating tender documents based on a large model, characterized in that, include: Obtain the tender requirements document and perform document analysis to identify multiple logical units within the tender requirements document; Semantic parsing is performed on each document logical unit to identify key semantic entities, and the key semantic entities are analyzed to determine the core requirements of each document logical unit. Based on the core requirements, a template knowledge framework matching the core requirements is retrieved from a pre-set knowledge database, and a retrieval dataset is constructed based on the template knowledge framework and the core requirements. The retrieved datasets of each document logical unit are input into the preset language model, and the text is output based on the preset language model to obtain the initial tender document; The content of the generated initial tender documents is verified and optimized to obtain the final, compliant formal tender documents.

2. The method for automatically generating tender documents based on a large model according to claim 1, characterized in that, The process of obtaining the bidding requirements document and performing document analysis to identify multiple logical units within the bidding requirements document includes: Obtain the bidding requirements document and preprocess it; Extract the basic textual structure information from the preprocessed tender requirements document, and divide the tender requirements document into multiple sentence segments based on the basic textual structure information; Each sentence segment is represented by a vector to obtain a sentence segment vector. The cosine similarity between adjacent sentence segment vectors is calculated, and adjacent sentence segments with a cosine similarity lower than a preset first threshold are identified. The boundary between adjacent sentences is defined as the boundary of the logical unit, and the tender requirement document is divided into multiple document logical units based on the boundary of the logical unit.

3. The method for automatically generating tender documents based on a large model according to claim 2, characterized in that, The semantic parsing of each document's logical unit to determine key semantic entities includes: Natural language processing techniques are used to extract semantic meaning information of each sentence segment from the logical units of a document, and semantic entities in each sentence segment are determined based on the semantic meaning information. The frequency, position, and entity type of semantic entities in the sentence are determined, and the criticality of semantic entities is evaluated based on the frequency, position, and entity type to obtain the criticality evaluation value of each semantic entity. Semantic entities whose criticality assessment value is greater than a preset first threshold are identified as key semantic entities in the document's logical unit.

4. The method for automatically generating tender documents based on a large model according to claim 3, characterized in that, The assessment of the criticality of semantic entities based on frequency, location, and entity type yields a criticality assessment value for each semantic entity, including: Determine a pre-set baseline frequency, calculate the difference between the frequency of the semantic entity appearing in the sentence and the baseline frequency, evaluate the difference, and obtain the first evaluation value of the semantic entity. The semantic entity is evaluated based on its position and entity type in the sentence, respectively, to obtain the second and third evaluation values ​​of the semantic entity. The first, second, and third evaluation values ​​of the semantic entity are added together to obtain the semantic entity's criticality evaluation value.

5. The method for automatically generating tender documents based on a large model according to claim 3, characterized in that, The analysis of key semantic entities to determine the core requirements of each document's logical unit includes: Extract the entity relationships between key semantic entities in the sentence, and extract the requirement elements from the key semantic entities and entity relationships; Fill in the demand elements according to the preset demand template to obtain the preliminary demand of each sentence, and extract the core semantic meaning of the preliminary demand of each sentence. By combining the core semantic meanings of the initial requirements of each sentence, we can obtain the core requirements of the document's logical units.

6. The method for automatically generating tender documents based on a large model according to claim 5, characterized in that, The step of retrieving template knowledge frameworks matching the core requirements from a pre-set knowledge database and constructing a retrieval dataset based on the template knowledge frameworks and the core requirements includes: Extract keywords from the core requirements and represent these keywords as vectors to obtain keyword vectors; Calculate the relevance of each keyword to the core theme of the preset requirements, and normalize each relevance to obtain the weight of each keyword; The core requirement vector is obtained by fusion calculation based on the weight of each keyword and the keyword vector, and the template clauses in the preset knowledge database are represented by vectors to obtain the template clause vector. Calculate the cosine similarity between the core requirement vector and each template clause vector, and determine the preset template clause with the highest cosine similarity as the template knowledge framework that matches the core requirement; The core requirements are mapped to the corresponding template knowledge frameworks, and a retrieval dataset is built based on the mapping relationship.

7. The method for automatically generating tender documents based on a large model according to claim 6, characterized in that, The formula for calculating the core requirement vector is as follows: Where E is the core demand vector, αi is the weight of the i-th keyword, P is the keyword vector of the i-th keyword, and n is the number of keywords.

8. The method for automatically generating tender documents based on a large model according to claim 6, characterized in that, The process of inputting the retrieval datasets of each document logical unit into a preset language model and outputting text based on the preset language model to obtain the initial tender document includes: Design corresponding prompt words for the core requirements and corresponding template knowledge frameworks in the retrieval dataset of each document logical unit; Based on the prompt words, the retrieval dataset of each document's logical unit is input into the preset language model, and the text is output through the preset language model to obtain the bidding chapter content of each document's logical unit; The tender section content of each document's logical unit is assembled according to the structure of a standard tender document to form an initial tender document.

9. The method for automatically generating tender documents based on a large model according to claim 8, characterized in that, The process of verifying and optimizing the content of the generated initial tender document to obtain the final, compliant formal tender document includes: Determine each tender clause in the generated initial tender document and verify the format, completeness, compliance, and consistency of each tender clause; The selected tender terms that passed verification were filtered out, and the selected tender terms were optimized based on the reasons for failure to pass verification. The optimized bidding terms are then matched to the corresponding content to obtain the final, compliant official bidding documents.

10. A system for automatically generating tender documents based on a large model, characterized in that, include: The acquisition module is used to acquire the bidding requirements document and perform document analysis on the bidding requirements document to identify multiple document logical units in the bidding requirements document; The analysis module is used to perform semantic parsing on each document logical unit, identify key semantic entities, and analyze the key semantic entities to determine the core requirements of each document logical unit. The matching module is used to retrieve template knowledge frameworks that match the core requirements from a pre-set knowledge database, and to construct a retrieval dataset based on the template knowledge frameworks and the core requirements. The generation module is used to input the retrieved datasets of each document logical unit into the preset language model, and output the text based on the preset language model to obtain the initial tender document; The optimization module is used to verify and optimize the content of the generated initial tender documents to obtain the final formal tender documents that meet the requirements.

Images