Full-automatic digital scheme generation method and system based on large model algorithm development

Abstract

The invention belongs to the technical field of natural language processing and generation, and discloses a full-automatic digital scheme generation method based on large model algorithm development, which adopts an integrated architecture, analyzes a bidding document through one key, generates a bidding document through one key, and improves the efficiency of bidding. The problems that in a traditional method, the requirement analysis efficiency is low, scheme generation is complex, personalized customization is difficult, and normative verification is insufficient are solved. The system greatly improves the accuracy, the efficiency, the automation degree and the intelligent level of digital scheme compilation, supports multi-user cooperation and real-time adjustment, and provides an efficient, low-cost and intelligent solution for enterprises.

Description

Technical Field

[0001] The present invention belongs to the field of natural language processing and generation technology, and in particular relates to a fully automatic digital solution generation method and system based on large model algorithm development. Background Art

[0002] With the rapid development of artificial intelligence (AI), particularly the rise of generative artificial intelligence (AIGC) technology centered around large language models (LLMs), companies have significantly improved the efficiency and intelligence of bid document generation. Traditional bid document preparation is often time-consuming and labor-intensive, requiring a high level of professional expertise from the compilers. In large-scale projects, low efficiency and proneness to errors have long plagued companies. The introduction of AI technology offers new possibilities for resolving these issues, transforming bid document generation from a labor-intensive process to one that is intelligent, automated, and efficient.

[0003] However, the generation of bidding documents still faces a series of core pain points. This process involves multiple links, such as analysis of bidding documents, breakdown of project requirements, design of technical solutions, quotation calculation, and formatting and verification. Specific issues include: ① Complexity and high cost: The requirements of different projects are complex and diverse, and the bidding document preparation process is cumbersome, requiring a lot of manpower and material resources, and the cycle is long, making it difficult to meet the needs of efficient output. ② High error rate: Manual processing of large amounts of data and documents can easily lead to content errors, omissions of clauses, or irregular formats. These errors can seriously affect the competitiveness and effectiveness of the bid. ③ Difficulty in inheriting professional experience: Certain key points in the writing of bidding documents rely on the industry experience accumulated by bidders over a long period of time, and this implicit knowledge is often difficult to systematically transmit or share. ④ Difficulty in balancing personalization and standardization: Compilers need to strictly abide by industry standards and specifications while meeting the personalized needs of customers, which poses a dual challenge to the accuracy and flexibility of proposal writing.

[0004] In recent years, large-scale pre-trained model (LLM) technology has made rapid progress in the field of natural language processing (NLP). The emergence of LLM has greatly improved machine performance in tasks such as text generation, semantic understanding, and conversational interaction, even surpassing human performance in some tasks. Compared to traditional text processing methods, LLM-based technologies are built on the foundation of deep learning and large-scale data training, achieving precise processing of text tasks through high-dimensional feature representation and complex semantic modeling.

[0005] In LLM technology, the Transformer architecture, as a core technology, has become the mainstream approach to solving natural language understanding and generation tasks due to its innovative design based on the Attention Mechanism. Composed of an encoder and a decoder, the Transformer efficiently captures long-range dependencies and deeply models text semantics through multi-head attention and feedforward neural networks, making it widely used in fields such as machine translation, summarization, and question-answering systems.

[0006] Improved Transformer-based models (such as GPT, T5, and BERT) have further promoted the application and development of NLP technology. For example, the GPT (Generative Pre-trained Transformer) series of models uses a unidirectional autoregressive language modeling approach and excels in text generation tasks; while BERT (Bidirectional Encoder Representations from Transformers) uses a bidirectional encoding strategy to perform well in semantic understanding and text classification tasks. In addition, T5 (Text-to-Text Transfer Transformer) unifies the input and output formats of various text tasks, transforming complex tasks into simple text-to-text conversion problems, significantly improving multi-task processing capabilities.

[0007] Building on the aforementioned models, modern LLMs further enhance their adaptability to domain-specific tasks by introducing techniques such as fine-tuning and prompt tuning. Fine-tuning enhances the model's specialized capabilities through additional training on domain data, while prompt tuning guides the model's output by designing input formats, effectively reducing its reliance on large amounts of labeled data.

[0008] Traditional digital proposal development methods, including manual writing and simple template-based generation, suffer from significant shortcomings in efficiency and accuracy, making them unable to meet the demands of the big data era for efficiency, precision, and intelligence. These methods often rely on manual experience and repetitive labor, which is not only time-consuming and labor-intensive, but also prone to errors and non-standard formatting due to human oversight. They are no longer adaptable to the complex and volatile market environment and fierce industry competition.

[0009] Through the above analysis, the problems and defects of the existing technology are as follows:

[0010] Traditional digital proposal development methods, including manual writing and simple template-based generation, suffer from significant shortcomings in efficiency and accuracy, making them unable to meet the demands of the big data era for efficiency, precision, and intelligence. These methods often rely on manual experience and repetitive labor, which is not only time-consuming and labor-intensive, but also prone to errors and non-standard formatting due to human oversight. They are no longer adaptable to the complex and volatile market environment and fierce industry competition. Summary of the Invention

[0011] In view of the problems existing in the prior art, the present invention provides a fully automatic digital solution generation method based on large model algorithm development.

[0012] The present invention is implemented as follows: a fully automatic digital solution generation method based on large model algorithm development includes:

[0013] Step 1, raw data input:

[0014] The system receives raw data provided by the user in text format; the document format can be PDF or Word. The quality and integrity of the data are crucial at this stage.

[0015] Step 2, data preprocessing;

[0016] Step 3, data enhancement;

[0017] Step 4: Preview the tender document outline;

[0018] Step 5, secondary data enhancement:

[0019] The system will further enhance the data through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or select the system's public library data to enhance the versatility of the results.

[0020] Step 6, document generation algorithm;

[0021] Step 7, result output;

[0022] Step 8, monitor logs and records:

[0023] Throughout the entire process, the system monitors each stage and records relevant logs; monitoring and logging are used to track system performance, detect anomalies, and conduct security audits.

[0024] Furthermore, the data is preprocessed:

[0025] The raw data undergoes preprocessing, including text content extraction, data cleaning, and data enhancement;

[0026] 1) Text content extraction: Using document analysis algorithms, the text content of the bidding documents is extracted and processed, and then the extracted text content is cleaned;

[0027] 2) Data cleaning: Text analysis models are used to remove duplicate data, correct format inconsistencies, and fill in missing values. Text data cleaning involves removing irrelevant words and special characters.

[0028] 3) Data enhancement: After data cleaning, the data is further enhanced. The text data is extracted by extracting the information of each title node, and then reformatted through the text analysis model to obtain the content of each part.

[0029] Furthermore, the data is enhanced:

[0030] After feature extraction, the system will enhance the data; this is achieved through data expansion. Users view the content obtained through algorithm analysis and can choose to modify and fill in the content themselves; or if no changes are made, the algorithm will enhance itself. Finally, through the document analysis algorithm and text analysis model, the outline of the bidding document is obtained.

[0031] Further, the outline of the bidding document is previewed:

[0032] The system will provide the bidding document outline to the user for preview. After verification, the user can enter the bidding document generation module. If there is an error, the user can also choose to restore with one click, return to the bidding document extraction module, and make modifications.

[0033] Furthermore, the document generation algorithm:

[0034] The content of each module verified by users is assembled and filled through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies to ensure that the generated content complies with legal and ethical standards.

[0035] Furthermore, the result output is:

[0036] Before generating results, the system will perform post-processing, including content quality verification, logical consistency verification, and compliance review; for the proprietary library data uploaded by users, the system will perform private isolation and privacy processing. Users cannot obtain other companies' proprietary library data, nor can they download their own company's proprietary library data; this can prevent user data from being abused, tampered with, or leaked; the generated results will also be implicitly processed within the system and then output to the user in the form of text and a Word document.

[0037] Another object of the present invention is to provide a fully automatic digital solution generation system based on large model algorithm development, comprising:

[0038] The data input module is used for the system to receive raw data provided by users. The data format is text; the document format can be PDF or Word. The quality and integrity of the data are crucial at this stage.

[0039] Data preprocessing module, used for preprocessing of raw data, including text content extraction, data cleaning, and data enhancement;

[0040] The data enhancement module is used to enhance the data after feature extraction. This is achieved through data expansion. Users can view the content obtained through algorithm analysis and choose to modify and fill in the content themselves. Alternatively, the algorithm will automatically enhance the content without making any changes. Finally, the document analysis algorithm and text analysis model are used to obtain the tender document outline.

[0041] The preview module is used to provide the user with a preview of the tender document outline. After verifying that it is correct, the user can enter the tender document generation module. If there is an error, the user can also choose to restore it with one click, returning to the tender document extraction module to make further changes.

[0042] The secondary data enhancement module is used to improve the robustness of the model. The system will further enhance the data. This is achieved through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or they can choose the system's public library data to enhance the versatility of the results.

[0043] The document generation algorithm module is used to assemble and fill the content of each module verified by users through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies (content review and filtering) to ensure that the generated content complies with legal and ethical standards;

[0044] The result output module is used to perform post-processing before generating results, including content quality verification, logical consistency verification, and compliance review. The system will isolate and protect the privacy of user-uploaded private library data. Users cannot access other companies' private library data or download their own company's private library data. This prevents user data from being abused, tampered with, or leaked. The generated results will also be implicitly processed within the system and then output to the user in the form of text and Word documents.

[0045] The monitoring module is used to monitor each stage of the entire process and record relevant logs. Monitoring and logging are used to track system performance, detect anomalies, and conduct security audits to ensure the security and stability of the algorithm.

[0046] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the fully automatic digital solution generation method developed based on the large model algorithm.

[0047] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the fully automatic digital solution generation method developed based on the large model algorithm.

[0048] Another object of the present invention is to provide an information data processing terminal, which is used to implement the fully automatic digital solution generation system developed based on the large model algorithm.

[0049] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0050] First, large language models (LLMs) and multimodal generation technologies have significant advantages in semantic understanding, feature extraction, and large-scale data processing. Based on these technical advantages, the present invention constructs a fully automatic digital proposal generation system based on a combination of artificial intelligence-driven large language models and large model calculation algorithms. The system extracts and analyzes key information in bidding documents through a deep semantic understanding model, and uses multimodal generation technology to achieve diversified and structured expression of bidding proposal content.

[0051] This system utilizes an integrated architecture to analyze and generate bidding documents with a single click, addressing the challenges of traditional methods, such as inefficient demand analysis, complex proposal generation, difficulty in personalized customization, and insufficient compliance verification. The system significantly improves the accuracy, efficiency, automation, and intelligence of digital proposal development, while supporting multi-user collaboration and real-time adjustments, providing businesses with an efficient, low-cost, and intelligent solution.

[0052] (1) This invention applies artificial intelligence and deep learning technologies to the automatic parsing of bidding documents and the generation of tender documents. A large language model based on deep learning can extract deep semantic information from complex texts, significantly improving the accuracy and efficiency of text parsing compared to traditional rule-matching methods.

[0053] (2) This invention uses a fine-tuned large language model combined with a multi-task learning mechanism to optimize different task modules: ① Key information extraction: Through named entity recognition (NER) and relationship extraction technology, core information such as project name, bidding scope, and time nodes in the bidding documents are accurately identified. ② Demand analysis and intelligent matching: Using deep semantic analysis technology, the requirements in the bidding documents are matched with the company's solutions to generate customized response solutions, thereby improving the pertinence and effectiveness of bid document generation.

[0054] (3) This invention achieves integrated intelligent processing from tender document parsing to bid document generation, avoiding the inconsistency and inefficiency caused by manual operations: ① Automated generation: By generating a model, a draft tender document covering the technical solution, commercial quotation, and qualification certificate is quickly generated. ② Intelligent optimization: By combining historical data and industry standards, the generated document is intelligently adjusted to optimize the bidding strategy, improve document quality and the success rate of the bid.

[0055] (4) The present invention significantly reduces the time and cost required for manual intervention through intelligent analysis and automatic generation technology, ensures efficient and accurate document processing, and provides comprehensive technical support for enterprises to participate in bidding activities.

[0056] Second, current bidding document generation technology is still in its early stages globally, relying on template filling and keyword retrieval. This technology is unable to effectively address the diverse and heterogeneous nature of regulations, complex industry terminology, and flexible bid evaluation rules. This leads to fragmented content, broken logic, and incomplete responses. This is particularly true for semantic parsing of localized Chinese regulations and technical standards. Common models like GPT or BERT lack deep understanding and suffer from fundamental flaws such as "misinterpretation of clauses" and "parameter mismatches." Furthermore, most domestic solutions lack the ability to construct a four-dimensional knowledge network of regulations, standards, ratings, and clauses that can be updated in real time, making it difficult to meet the multi-dimensional requirements of "compliant, complete, and differentiated" bidding documents in industries such as engineering, healthcare, and information technology.

[0057] This invention focuses on the four core capabilities of "intelligent compliance generation + implicit demand perception + strategic resource planning + format fidelity response" and innovatively constructs an intelligent bidding document generation system driven by multimodal data. By constructing a cross-industry layered dynamic knowledge graph, it achieves full semantic association and call of legal provisions, technical parameters, scoring indicators and commercial terms. It also introduces a deep attention mechanism to identify "industry unwritten rules" and implicit requirements not explicitly marked in bidding documents, such as the equipment operation and maintenance cycle in the power industry and the clinical certification path in the medical industry. It intelligently generates a regulatory compliance matrix and a technical priority map, providing rigid guidance for the generation of technical solutions.

[0058] Breaking through the technical bottleneck of traditional bidding strategy formulation based on manual experience, this invention proposes for the first time a closed-loop decision-making engine based on "data input - model calculation - strategy output," embedded with three intelligent subsystems: technical solutions, resource allocation, and cost planning. Based on project requirements and the company's historical strengths, it automatically matches registered personnel, builds a personnel list for proposed equipment, and generates documents such as quality and schedule plans. This enables an integrated process from qualification matching, certificate collection, resume extraction, to the automatic insertion of social security certificates, significantly improving the rationality of staffing and the completeness of response documents. Verified in over 5,000 projects, it has reduced personnel costs by an average of 23%, increased critical path accuracy by 40%, and boosted the winning bid rate by 39%.

[0059] To address the complex, ever-changing, and demanding formats of tender document responses, this paper has developed a full-format intelligent response system, overcoming the inability of traditional text generation tools to handle complex tables and layouts. Through a three-tiered mechanism of format semantic parsing, field rule modeling, and pixel-level format restoration, it automatically fills in and maintains the format fidelity of all specified format documents, including "Project Manager Resume Form," "Equipment Details Form," and "Commercial Terms Response Form." It also automatically completes layout details such as chart numbering, directory generation, footer control, and field verification. With a response terms coverage rate of up to 99.8%, it completely eliminates the risk of bid rejection due to "format deviations," filling a global gap in automated generation technology for the entire bidding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of a fully automatic digital solution generation method based on large model algorithm development provided by an embodiment of the present invention.

[0061] Figure 2 This is a structural block diagram of a fully automatic digital solution generation system developed based on a large model algorithm provided by an embodiment of the present invention.

[0062] Figure 3 This is an overall flow chart of the algorithm provided by an embodiment of the present invention.

[0063] Figure 4 The present invention provides an automatic bidding document generation rendering.

[0064] Figure 5 This is a complete response diagram for analyzing the bidding document format provided by an embodiment of the present invention.

[0065] Figure 6 This is an automatic extraction diagram for responding to bidding requirements one by one provided by an embodiment of the present invention.

[0066] Figure 7 This is a diagram for automatically arranging vehicle information provided by an embodiment of the present invention.

[0067] Figure 8This is a personnel information automatic configuration diagram provided by an embodiment of the present invention.

[0068] Figure 9 This is a diagram of generated content with both pictures and texts provided by an embodiment of the present invention.

[0069] Figure 10 This is a reference format automatic filling diagram provided by an embodiment of the present invention.

[0070] Figure 11 This is an automatic typesetting diagram for automatically generating a directory provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0072] like Figure 1 As shown, an embodiment of the present invention provides a fully automatic digital solution generation method based on large model algorithm development, comprising the following steps:

[0073] S101, original data input:

[0074] The system receives raw data provided by the user in text format; the document format can be PDF or Word. The quality and integrity of the data are crucial at this stage.

[0075] S102, data preprocessing;

[0076] S103, data enhancement;

[0077] S104, preview of the bidding document outline;

[0078] S105, secondary data enhancement:

[0079] The system will further enhance the data through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or select the system's public library data to enhance the versatility of the results.

[0080] S106, document generation algorithm;

[0081] S107, result output;

[0082] S108, monitoring logs and records:

[0083] Throughout the entire process, the system monitors each stage and records relevant logs; monitoring and logging are used to track system performance, detect anomalies, and conduct security audits.

[0084] The core of the fully automatic digital solution generation method based on the large model algorithm provided by the embodiment of the present invention is to use a pre-trained language model combined with a structured enhancement strategy to complete the closed-loop processing of bidding documents from raw input to multi-dimensional output. In the first stage, the system receives raw text data (such as PDF, Word, etc.) and automatically identifies key fields, industry terms, technical indicators and response elements. This link plays a decisive role in the accuracy of subsequent processing, so the system has an embedded data integrity detection module to ensure that the input materials meet basic structural and semantic requirements.

[0085] Then it enters the pre-processing and data enhancement stage, where the system structures the original document content through segmentation parsing, named entity recognition, keyword extraction and other algorithmic means; and supplements missing fields by introducing a semantic completion model. The data after the first round of enhancement is used to generate an outline preview of the bidding document, showing the main chapter framework, technical response items, business modules and qualification terms. Users can make targeted adjustments based on industry characteristics or project requirements. The system supports secondary enhancement processing, and users can choose to introduce enterprise-specific databases (such as proprietary technical solution libraries, historical bidding documents, patent documents, etc.), or enable the industry public database provided by the system to enhance the pertinence and generalization capabilities of the generated results.

[0086] The document generation phase leverages a deep language modeling engine to perform nested reasoning, context modeling, and logical reorganization on the enhanced structured corpus, automatically generating core bidding documents such as technical proposals, commercial terms response forms, and qualification certifications. This process combines automated formatting with precise content matching to ensure that generated documents meet both content integrity and format compliance requirements. The system ultimately outputs editable Word or PDF files and also generates a data mapping table for easy traceability and secondary revisions.

[0087] The entire document generation process is supported by a monitoring and logging system that records every processing step, input and output changes, and model invocation, enabling performance tracking, anomaly detection, and compliance auditing. The system supports archiving logs by project, enterprise, industry, and other dimensions, providing a reliable basis for later model fine-tuning, output traceability, and behavioral auditing, creating a secure, transparent, and intelligent closed-loop mechanism for the entire document generation process.

[0088] Data preprocessing provided by the embodiment of the present invention:

[0089] The raw data undergoes preprocessing, including text content extraction, data cleaning, and data enhancement;

[0090] 1) Text content extraction: Using document analysis algorithms, the text content of the bidding documents is extracted and processed, and then the extracted text content is cleaned;

[0091] 2) Data cleaning: Text analysis models are used to remove duplicate data, correct format inconsistencies, and fill in missing values. Text data cleaning involves removing irrelevant words and special characters.

[0092] 3) Data enhancement: After data cleaning, the data is further enhanced. The text data is extracted by extracting the information of each title node, and then reformatted through the text analysis model to obtain the content of each part.

[0093] The data enhancement provided by the embodiment of the present invention is as follows:

[0094] After feature extraction, the system will enhance the data; this is achieved through data expansion. Users view the content obtained through algorithm analysis and can choose to modify and fill in the content themselves; or if no changes are made, the algorithm will enhance itself. Finally, through the document analysis algorithm and text analysis model, the outline of the bidding document is obtained.

[0095] Preview of the bidding document outline provided by the embodiment of the present invention:

[0096] The system will provide the bidding document outline to the user for preview. After verification, the user can enter the bidding document generation module. If there is an error, the user can also choose to restore with one click, return to the bidding document extraction module, and make modifications.

[0097] The document generation algorithm provided by the embodiment of the present invention:

[0098] The content of each module verified by users is assembled and filled through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies to ensure that the generated content complies with legal and ethical standards.

[0099] The result output provided by the embodiment of the present invention is:

[0100] Before generating results, the system will perform post-processing, including content quality verification, logical consistency verification, and compliance review; for the proprietary library data uploaded by users, the system will perform private isolation and privacy processing. Users cannot obtain other companies' proprietary library data, nor can they download their own company's proprietary library data; this can prevent user data from being abused, tampered with, or leaked; the generated results will also be implicitly processed within the system and then output to the user in the form of text and a Word document.

[0101] like Figure 2 As shown, an embodiment of the present invention provides a fully automatic digital solution generation system based on large model algorithm development, including:

[0102] The data input module is used for the system to receive raw data provided by users. The data format is text; the document format can be PDF or Word. The quality and integrity of the data are crucial at this stage.

[0103] Data preprocessing module, used for preprocessing of raw data, including text content extraction, data cleaning, and data enhancement;

[0104] The data enhancement module is used to enhance the data after feature extraction. This is achieved through data expansion. Users can view the content obtained through algorithm analysis and choose to modify and fill in the content themselves. Alternatively, the algorithm will automatically enhance the content without making any changes. Finally, the document analysis algorithm and text analysis model are used to obtain the tender document outline.

[0105] The preview module is used to provide the user with a preview of the tender document outline. After verifying that it is correct, the user can enter the tender document generation module. If there is an error, the user can also choose to restore it with one click, returning to the tender document extraction module to make further changes.

[0106] The secondary data enhancement module is used to improve the robustness of the model. The system will further enhance the data. This is achieved through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or they can choose the system's public library data to enhance the versatility of the results.

[0107] The document generation algorithm module is used to assemble and fill the content of each module verified by users through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies (content review and filtering) to ensure that the generated content complies with legal and ethical standards;

[0108] The result output module is used to perform post-processing before generating results, including content quality verification, logical consistency verification, and compliance review. The system will isolate and protect the privacy of user-uploaded private library data. Users cannot access other companies' private library data or download their own company's private library data. This prevents user data from being abused, tampered with, or leaked. The generated results will also be implicitly processed within the system and then output to the user in the form of text and Word documents.

[0109] The monitoring module is used to monitor each stage of the entire process and record relevant logs. Monitoring and logging are used to track system performance, detect anomalies, and conduct security audits to ensure the security and stability of the algorithm.

[0110] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the fully automatic digital solution generation method developed based on the large model algorithm.

[0111] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the fully automatic digital solution generation method developed based on the large model algorithm.

[0112] Another object of the present invention is to provide an information data processing terminal, which is used to implement the fully automatic digital solution generation system developed based on the large model algorithm.

[0113] The present invention is specifically implemented:

[0114] The present invention is divided into three steps: semantic understanding of bidding documents, key information extraction and demand analysis, and full-automatic generation and optimization of bidding documents.

[0115] The overall process of the algorithm is as follows:

[0116] (1) Algorithm flow, such as Figure 3 ;

[0117] The flowchart describes a complete service chain for generating synthesis algorithms, from the input of raw data to the output of the final result. The specific steps are as follows:

[0118] 1. Original data input (user uploads bidding documents):

[0119] The system receives raw data (tender documents) provided by the user in text format. Document formats can be PDF, Word, etc. The quality and integrity of the data are crucial at this stage.

[0120] 2. Data preprocessing:

[0121] The raw data undergoes preprocessing, including text content extraction, data cleaning, and data enhancement.

[0122] 1) Text content extraction: Using document analysis algorithms, the text content of the bidding documents is extracted and processed, and then the extracted text content is cleaned.

[0123] 2) Data Cleaning: Text analysis models remove duplicate data, correct formatting inconsistencies, and fill in missing values. Text data cleaning involves removing irrelevant words and special characters.

[0124] 2) Data augmentation: After data cleaning, we further enhance the data. Text data is extracted by extracting information from each title node and then reformatted using a text analysis model to obtain the content of each section. Data augmentation aims to increase data diversity and enhance the model's generalization capabilities.

[0125] 3. Data Augmentation:

[0126] After feature extraction, the system enhances the data. This is achieved through data augmentation. Users review the content analyzed by the algorithm and can choose to modify and fill in the content themselves. Alternatively, the algorithm will automatically enhance the data without making any changes. Finally, through the document analysis algorithm and text analysis model, the bid document outline is obtained.

[0127] 4. Preview of the bidding document outline:

[0128] The system will provide the bidding document outline to the user for preview. After verification, the user can enter the bidding document generation module. If there is an error, the user can also choose to restore with one click, return to the bidding document extraction module, and make modifications.

[0129] 5. Secondary data enhancement:

[0130] To improve the robustness of the model, the system further enhances the data. This is achieved through data augmentation. Users can choose to upload their own library data to enhance the independence of the generated results, or select the system's public library data to enhance the versatility of the results.

[0131] 6. Document generation algorithm:

[0132] The content of each module verified by users is assembled and filled through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies (content review and filtering) to ensure that the generated content complies with legal and ethical standards.

[0133] 7. Result output:

[0134] Before generating results, the system performs post-processing, including content quality verification, logical consistency verification, and compliance review. For user-uploaded private database data, the system isolates and protects privacy. Users cannot access data from other companies' private databases, nor can they download data from their own company's private databases. This prevents user data from being misused, tampered with, or leaked. Generated results are also implicitly processed within the system before being output to the user in the form of text and a Word document.

[0135] 8. Monitoring logs and records:

[0136] Throughout the entire process, the system monitors each stage and records relevant logs. Monitoring and logging are used to track system performance, detect anomalies, and conduct security audits to ensure the security and stability of the algorithm.

[0137] (1) Semantic understanding of bidding documents based on fine-tuning large models

[0138] Deep learning and large language models (LLMs) have been widely applied in the field of natural language processing (NLP) and have achieved remarkable results. Compared with traditional text processing methods (such as rule-based text parsing and statistical learning methods), large language models can handle more complex semantic relationships, adapt to larger-scale data training, and extract deeper features from them, thereby improving text comprehension. With the increase in training data and continuous improvement of model architecture, large language models have become able to understand and generate text content with strong semantic depth. They are widely used in fields such as machine translation, text generation, and sentiment analysis, and have achieved breakthrough success.

[0139] This invention utilizes fine-tuning technology for large language models, combined with years of accumulated data cleaning results, to construct a high-quality dataset for training, deeply optimizing the semantic understanding of bidding documents. Through large-model intelligent analysis algorithms, the system can fully automatically parse the terms and complex demand descriptions in bidding documents, achieving accurate understanding and extraction of key information. This invention effectively addresses the inefficiency and error-proneness of traditional manual analysis methods, significantly improving the accuracy and automation of information analysis, and providing technical support for the rapid processing and intelligent generation of bidding documents.

[0140] (2) Key information extraction and demand analysis based on large language models

[0141] a. Key information extraction

[0142] This method leverages the deep semantic understanding capabilities of a large language model, combined with a fine-tuned and optimized model, to perform multi-level analysis of the text in bidding documents, accurately extracting key information. This key information includes, but is not limited to, project name, bidding scope, technical requirements, timelines, contract terms, and scoring criteria.

[0143] Method Overview: Based on a multi-task learning mechanism, a dedicated labeling system is constructed. Through named entity recognition (NER) and relation extraction (RE) technology, key information is extracted and associated by category to form structured data output.

[0144] Key technologies: ① Named Entity Recognition (NER): Accurately label entities (such as dates, amounts, and organization names) in bidding documents using deep learning models (such as BERT and RoBERTa). ② Multi-round semantic correction: Utilizing a feedback loop based on model generation results, the accuracy and completeness of extracted information are automatically optimized.

[0145] b. Bidding document requirements analysis

[0146] Based on the extraction results of key information, the present invention further conducts an in-depth analysis of the demand description of the bidding documents to form a structured demand model. And perform the following steps: ① Demand classification and priority division: Classify the bidding requirements according to dimensions such as technology, business, and contract terms, and combine the model prediction and scoring mechanism to determine the importance and priority of the requirements. ② Technical requirement mapping: Through deep semantic analysis, intelligently match the technical requirements with the company's existing solutions or technical capabilities to generate customized technical response plans. ③ Logical consistency check: Verify the logical relationship between the demand terms to ensure that there are no contradictions or omissions, and provide an accurate basis for the preparation of bidding documents.

[0147] (3) Fully automatic generation of bidding documents; e.g. Figure 4

[0148] This paper builds an automated generation system based on a large language model (LLM) and combines it with multimodal generation technology to achieve fully automatic, one-click generation of bid documents. Based on in-depth analysis of the bidding documents and demand models, the system generates complete bid documents covering multiple dimensions, including technical, commercial, qualification, and price.

[0149] The module is divided into the following core steps:

[0150] 1. Content Generation

[0151] Technical solution generation: Combining the technical requirements in the demand model and the company's solution library, the technical response content is automatically generated through a large language model to ensure accurate and professional content.

[0152] Business quotation generation: Based on business needs and company pricing rules, combined with the built-in quotation algorithm, a quotation list is automatically generated, supporting multiple versions to meet different bid evaluation criteria.

[0153] Contract clause writing: Automatically generate contract clause response content based on the contract requirements in the bidding documents to ensure that the clause content is standardized and compliant, while also supporting differentiated analysis and supplementation of clauses.

[0154] 2. Multimodal Enhanced Expression

[0155] Utilizing multimodal generation technology, we generate charts, flow charts, and visual data for bidding documents by combining text and graphics, intuitively presenting technical solutions and business advantages, and enhancing the competitiveness of the documents.

[0156] 3.Logic verification and optimization

[0157] Logical consistency check: Through deep semantic analysis, it automatically verifies the consistency of technical, commercial and contractual content in the bidding documents to avoid duplication or contradictions.

[0158] Content integrity check: Automatically check the content of the generated bidding documents according to the terms and conditions of the bidding documents to ensure the integrity and coverage of the documents.

[0159] 4. Personalization and version management

[0160] Intelligent personalized customization: Supports generation of personalized bidding document versions based on different evaluation criteria or customer preferences, thereby improving bidding success rate.

[0161] Version management and comparison: The system has built-in version management function, which supports the generation, comparison and optimization of multiple versions of files, providing users with flexible adjustment space.

[0162] 5. Final file generation and formatting

[0163] The system supports output in multiple document formats (such as Word, PDF) and automatically beautifies the format, including table of contents generation, page number addition, font and layout optimization, to ensure the professional appearance of the document.

[0164] Through the above modules, the present invention realizes the automation of the entire process from demand analysis to bidding document generation, greatly improves the efficiency and quality of document preparation, and provides enterprises with an efficient and intelligent solution.

[0165] The technical solution of this invention can be widely applied to intelligent generation of bidding documents in all industries, specifically covering the following six core areas:

[0166] 1. Engineering construction field

[0167] Applicable scenarios: general construction contracting bidding, engineering design bidding, supervision service bidding, PPP project bidding, etc.

[0168] Technical Value: Automatically parses the bill of quantities and generates quotation documents that comply with the "Construction Project Bill of Quantities Pricing Specifications"; intelligently plans construction equipment configuration plans based on the BIM model, and automatically verifies the project manager's registered construction engineer qualifications and social security payment records, solving the pain points of traditional bidding documents such as "disconnection between technical solutions and quotations" and "incomplete responses to personnel qualifications."

[0169] 2. Government procurement

[0170] Applicable scenarios: Goods procurement bidding (such as medical equipment, office supplies), service procurement bidding (such as software development, property services), engineering procurement bidding

[0171] Technical value: Automatically generate bidding documents that meet the format requirements of platforms such as "Government Procurement Cloud"; for confidential projects, intelligently generate technical solutions with watermark encryption to ensure that the responsiveness of commercial terms 100% covers procurement needs.

[0172] 3. Information Technology

[0173] Applicable scenarios: software development bidding, system integration bidding, cloud computing service bidding, information security project bidding

[0174] Technology value: Automatically generate bidding documents and intelligently match the enterprise's intellectual property database (patents / software copyrights) to build technology competitiveness analysis;

[0175] 4. Medical device field

[0176] Applicable scenarios: Medical equipment procurement bidding, in vitro diagnostic reagent bidding, medical information project bidding

[0177] Technical value: Automatically identify the clinical evaluation requirements in the "Medical Device Supervision and Administration Regulations" and generate complete bidding documents including product registration certificate, test report, and after-sales service system;

[0178] 5. Transportation

[0179] Applicable scenarios: highway engineering bidding, port and waterway project bidding, logistics service bidding, transportation informationization project bidding

[0180] Technical value: Automatically plan personnel and machinery scheduling plans (such as the number of pavers and rollers).

[0181] 6. Energy and chemical industry

[0182] Applicable scenarios: Power project bidding, petrochemical project bidding, new energy project bidding (photovoltaic / wind power)

[0183] Technical Value: Automatically matches enterprise safety production licenses and special operations personnel certificates; for photovoltaic EPC projects, intelligently generates equipment selection lists (photovoltaic modules, inverter technical parameters) and associates them with quotation analysis.

[0184] (2) Related product forms

[0185] Based on the technical solution of the present invention, the core functions of the titanium bidding product are formed, covering the full process intelligent needs from data management to file generation:

[0186] 1. Titanium Bidding (AI intelligent digital solution one-click generation platform)

[0187] Functional positioning: A one-stop bid generation system for bidding agencies and bidders, with one-click bid document generation, integrated knowledge graph, strategy engine, and format response as three core modules

[0188] Core Competencies:

[0189] (1) Support intelligent parsing and generation of bidding documents for 37 sub-sectors in 12 major industries

[0190] (2) Built-in 68 standard document template libraries (including official templates from the Ministry of Housing and Urban-Rural Development, the Ministry of Finance, etc.), supporting custom template import

[0191] (3) Technical advantages: Average bid generation time is ≤ 0.5 hours, format compliance rate is 100%, and the winning rate is increased by 39% (verified by 5,000+ projects)

[0192] Figure 5 The system demonstrated its ability to fully structure tender document templates, enabling one-click filling of content, including project numbers, bidder information, and qualification checklists, ensuring that the response document is highly consistent with the tender format. This process eliminates the need for manual duplication and effectively improves document preparation efficiency and compliance.

[0193] Figures 6 to 8 The specific implementation of automatic extraction and intelligent configuration functions is reflected in turn: Figure 6 Provides structured identification of bidding terms and clause-to-clause correspondence response results; Figure 7 It realizes the automatic capture and tabular output of vehicle information. Figure 8 It supports the intelligent integration and classified display of personnel information (such as certificates, identity documents, labor contracts, etc.), greatly improving the integrity and accuracy of bid preparation.

[0194] Figure 9 It reflects the system's ability to support the output of both text and pictures in content generation. It is suitable for modular text and picture combination content such as bidding plans, technical architectures, and implementation plans. It supports automatic retrieval of illustrations and structural relationship diagrams, and solves the problem of chaotic editing structure of traditional Word versions of bidding documents.

[0195] Figure 10 and Figure 11 The system automatically extracts chapter titles and page numbers based on the preceding text structure, enabling fully automatic table of contents generation. Reference document templates can also be quickly replaced with content tailored to the current project, ensuring document professionalism and consistent layout.

[0196] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0197] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A fully automatic digital solution generation method based on large model algorithm development, characterized in that: The following steps are involved: Step 1, raw data input: The system receives the original data provided by the user in text format; The document format can be PDF or Word; the quality and integrity of the data are crucial at this stage; Step 2, data preprocessing; Step 3, data enhancement; Step 4: Preview the tender document outline; Step 5, secondary data enhancement: The system will further enhance the data through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or select the system's public library data to enhance the versatility of the results. Step 6, document generation algorithm; Step 7, result output; Step 8, monitor logs and records: Throughout the entire process, the system will monitor each stage and record relevant logs; Monitoring and logging are used to track system performance, detect anomalies, and conduct security audits.

2. The fully automatic digital solution generation method based on large model algorithm development as claimed in claim 1, characterized in that: The data preprocessing: The raw data undergoes preprocessing, including text content extraction, data cleaning, and data enhancement; 1) Text content extraction: Using document analysis algorithms, the text content of the bidding documents is extracted and processed, and then the extracted text content is cleaned; 2) Data cleaning: Text analysis models are used to remove duplicate data, correct format inconsistencies, and fill in missing values. Text data cleaning involves removing irrelevant words and special characters. 3) Data enhancement: After data cleaning, the data is further enhanced. The text data is extracted by extracting the information of each title node, and then reformatted through the text analysis model to obtain the content of each part.

3. The fully automatic digital solution generation method based on large model algorithm development as claimed in claim 1, characterized in that: The data augmentation: After feature extraction, the system will enhance the data; this is achieved through data expansion. Users view the content obtained through algorithm analysis and can choose to modify and fill in the content themselves; or if no changes are made, the algorithm will enhance itself. Finally, through the document analysis algorithm and text analysis model, the outline of the bidding document is obtained.

4. The fully automatic digital solution generation method based on large model algorithm development as claimed in claim 1, characterized in that: Preview of the outline of the bidding documents: The system will provide the bidding document outline to the user for preview. After verification, the user can enter the bidding document generation module. If there is an error, the user can also choose to restore with one click, return to the bidding document extraction module, and make modifications.

5. The fully automatic digital solution generation method based on large model algorithm development as claimed in claim 1, characterized in that: The document generation algorithm: The content of each module verified by users is assembled and filled through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies to ensure that the generated content complies with legal and ethical standards.

6. The fully automatic digital solution generation method based on large model algorithm development as claimed in claim 1, characterized in that: The resulting output is: Before generating results, the system will perform post-processing, including content quality verification, logical consistency verification, and compliance review; for the proprietary library data uploaded by users, the system will perform private isolation and privacy processing. Users cannot obtain other companies' proprietary library data, nor can they download their own company's proprietary library data; this can prevent user data from being abused, tampered with, or leaked; the generated results will also be implicitly processed within the system and then output to the user in the form of text and a Word document.

7. A fully automatic digital solution generation system based on large model algorithm development that implements the fully automatic digital solution generation method based on large model algorithm development as described in any one of claims 1 to 6, characterized in that: The fully automatic digital solution generation system developed based on the large model algorithm includes: The data input module is used for the system to receive raw data provided by users. The data format is text; the document format can be PDF or Word. The quality and integrity of the data are crucial at this stage. Data preprocessing module, used for preprocessing of raw data, including text content extraction, data cleaning, and data enhancement; The data enhancement module is used to enhance the data after feature extraction. This is achieved through data expansion. Users can view the content obtained through algorithm analysis and choose to modify and fill in the content themselves. Alternatively, the algorithm will automatically enhance the content without making any changes. Finally, the document analysis algorithm and text analysis model are used to obtain the tender document outline. The preview module is used to provide the user with a preview of the tender document outline. After verifying that it is correct, the user can enter the tender document generation module. If there is an error, the user can also choose to restore it with one click, returning to the tender document extraction module to make further changes. The secondary data enhancement module is used to improve the robustness of the model. The system will further enhance the data. This is achieved through data expansion. Users can choose to upload their own library data to enhance the independence of the generated results, or they can choose the system's public library data to enhance the versatility of the results. The document generation algorithm module is used to assemble and fill the content of each module verified by users through algorithms and text processing models. At the same time, the algorithm will apply intervention strategies (content review and filtering) to ensure that the generated content complies with legal and ethical standards; The result output module is used to perform post-processing before generating results, including content quality verification, logical consistency verification, and compliance review. The system will isolate and protect the privacy of user-uploaded private library data. Users cannot access other companies' private library data or download their own company's private library data. This prevents user data from being abused, tampered with, or leaked. The generated results will also be implicitly processed within the system and then output to the user in the form of text and Word documents. The monitoring module is used to monitor each stage of the entire process and record relevant logs. Monitoring and logging are used to track system performance, detect anomalies, and conduct security audits to ensure the security and stability of the algorithm.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the fully automatic digital solution generation method based on large model algorithm development as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for generating a fully automatic digital solution based on large model algorithm development as claimed in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the fully automatic digital solution generation system developed based on the large model algorithm as described in claim 7.

Images