Intelligent bidding document generation system and method

Abstract

The invention discloses an intelligent bidding document generation system and method. The system comprises an intelligent document generation module for automatically generating bidding document content based on an input bidding document; the automatic optimization and content proofreading module is used for automatically verifying and correcting grammar, logic and format of the bidding document content; the multi-target evaluation and scoring module dynamically determines a weight coefficient of each evaluation dimension based on content analysis of the bidding document, and performs comprehensive quantitative scoring on the bidding document according to the weight coefficient; the intelligent auxiliary review module is used for generating structured review feedback based on the evaluation result; and the automatic export module is used for outputting the final document. Wherein the intelligent document generation module intelligently selects a bidding document template through a matching algorithm and historical efficiency data, schedules a large language model, and coordinates a plurality of sub-task nodes through an AI workflow to cooperatively generate contents. According to the invention, full-process automation and intelligentization of bidding document generation, optimization, evaluation and feedback are realized, and the bidding document making efficiency, standardization and quality are improved.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and artificial intelligence, specifically to an intelligent tender document generation system and method, which is particularly suitable for automated document generation and review in complex project bidding. Background Technology

[0002] With the rapid development of artificial intelligence technology, the traditional tender document preparation and review process has gradually revealed its inefficiency and the possibility of human error. Especially when dealing with massive amounts of data and complex documents, traditional methods often struggle to quickly and effectively generate high-quality tender documents that meet bidding requirements. Existing tender document preparation systems typically rely on manual input and rule-based generation, resulting in inconsistent document quality, significant time consumption, and difficulty in handling complex tasks. Although some advanced tools have introduced text generation and optimization technologies, these systems have not yet achieved full automation and intelligence, often relying on expert input and failing to flexibly adapt to the needs of different projects.

[0003] The current mainstream methods for preparing tender documents include: Rule-driven tender document generation: This method uses preset templates or process-driven generation methods. While simple and easy to implement, it is poorly adaptable to complex projects and cannot effectively handle diverse and changing requirements.

[0004] Template-driven tender document creation: This method uses fixed-format templates to quickly populate relevant data. While it speeds up tender document creation, it lacks personalization and flexibility, making it difficult to meet customized needs.

[0005] AI-based document generation: Some systems attempt to generate tender documents using natural language generation technologies (such as GPT, Qwen, Deepseek and other large language models), but these systems still rely on human intervention, cannot be fully automated, and have limited effectiveness in dealing with complex project requirements.

[0006] The existing methods for preparing tender documents have the following drawbacks: Low level of automation: Traditional methods still rely on a lot of manual input, which is inefficient and prone to errors, especially when the content of the tender document is large or the requirements are complex, making it difficult to maintain consistency and high quality.

[0007] Lack of flexibility and intelligence: Existing systems typically use fixed templates or rules, which cannot be flexibly adjusted and customized for different types of tender documents, resulting in unsatisfactory results.

[0008] Poor content consistency: When multiple people participate in the preparation of a tender document, inconsistencies in content style and format may occur, affecting the overall quality.

[0009] Inadequate evaluation system: Existing systems typically lack a comprehensive evaluation mechanism for the quality of tender documents, and are unable to comprehensively assess the completeness, accuracy, and compliance of tender content from multiple dimensions.

[0010] Based on this, the present invention proposes an intelligent tender document generation system and method, which aims to optimize the efficiency and quality of tender document production through intelligent algorithms and automation technology, and solve the defects in the existing technology. Summary of the Invention

[0011] The purpose of this invention is to provide an intelligent tender document generation system and method to solve the problems mentioned in the above-mentioned technical background.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent tender document generation system, comprising: The intelligent document generation module is used to automatically generate tender document content based on the input tender documents; The automatic optimization and content proofreading module is used to automatically check and correct the grammar, logical consistency and format of the tender document content; The multi-objective evaluation and scoring module is used to comprehensively evaluate and quantitatively score the tender document content after it has been processed by the automatic optimization and content proofreading module. The multi-objective evaluation and scoring module is configured to dynamically determine the weight coefficients of each evaluation dimension based on the content analysis of the tender document, and calculate the comprehensive score based on the dynamic weights and preset scoring rules. The intelligent assisted review module is used to generate structured review feedback based on the evaluation results of the multi-objective evaluation and scoring module. The automated export module is used to export the optimized tender document content and the review feedback into a document in a specified format.

[0013] In a preferred embodiment, the intelligent document generation module includes: The tender document parsing unit is used to extract key information from the tender document and structure it into standardized data. The template selection unit is used to select a suitable tender document template from a preset template library based on the key information. The template selection unit is configured to: obtain a candidate template set by calculating the matching degree between the tender document and the templates in the template library, the candidate template set including at least one candidate template, and determine the final template from the candidate template set by comprehensively considering the relevant data of the candidate templates and the bidding results of historical projects, as well as the demand preferences for technology or cost focus analyzed from the tender document. The content generation unit is used to schedule a large language model and combine it with an AI workflow to generate tender content based on the selected template and the standardized data; the AI ​​workflow is configured to decompose the tender generation task into multiple sub-task nodes with dependencies for scheduling and execution.

[0014] In a preferred embodiment, the template library is constructed according to industry classification and project size classification; wherein, the industry classification includes construction engineering, IT services and equipment procurement; and the project size classification includes large projects, medium projects and small projects.

[0015] The templates for construction projects include content such as project construction, schedule planning, and project budget; the templates for IT services include requirements for software development, hardware procurement, technical solutions, and after-sales service; and the templates for equipment procurement include content such as equipment technical specifications, supplier qualifications, delivery time, and price.

[0016] In a preferred embodiment, the template selection unit includes: The feature extraction subunit is used to extract the core keywords in the bidding documents through a first natural language processing model (such as the TF-IDF algorithm) and simultaneously parse the deep semantics of the bidding documents through a second natural language processing model (such as the BERT semantic model); the keyword features and semantic features are fused to generate a comprehensive feature vector of the bidding documents; The candidate generation subunit is used to calculate the matching degree between the comprehensive feature vector of the tender document and the feature vector of each template pre-stored in the template library using a similarity matching algorithm (such as cosine similarity); sort the templates from high to low according to the matching degree, and select the templates with matching degree higher than a preset threshold to form an initial candidate template set; The intelligent decision-making subunit is used to execute the following intelligent decision-making process to determine the final selected tender template when multiple templates exist in the initial candidate template set: a) Obtain the static performance attributes of each candidate template, including: the success rate of the candidate template in historical projects, the average review score of the tender documents generated based on the candidate template, and the average cost deviation of historical projects that applied the template. b) Extract dynamic demand preference features from the tender documents. Specifically, using natural language processing technology, the total frequency of occurrence of the first set of preset keywords (such as "innovation", "patent", "originality", "technological advantage") associated with the technological innovation dimension and the total frequency of occurrence of the second set of preset keywords (such as "cost", "budget", "quotation", "control", "saving") associated with the cost control dimension are statistically analyzed. Based on the total frequency of occurrence of the first set of preset keywords and the total frequency of occurrence of the second set of preset keywords, a quantitative preference vector is generated to characterize the degree of emphasis on technological innovation and cost control in the tender documents. c) Apply a predefined decision rule model and output the final template from the candidate template set based on the static performance attributes and dynamic demand preferences; The decision rule model is configured as follows: When the quantitative preference vector representing the emphasis on technological innovation in the tender documents is higher than the first threshold, the technical efficiency index of each candidate template is calculated according to the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, and the candidate template with the highest technical efficiency index is selected; where α and β are preset weight coefficients, and α + β = 1; When the quantitative preference vector represents the emphasis on cost control in the tender document, which is higher than the second threshold, the candidate template with the lowest average cost deviation in historical projects is selected first.

[0017] In a preferred embodiment, the large language model is the Qwen-32B model, deployed in a Docker environment on the Dify platform.

[0018] In a preferred embodiment, the AI ​​workflow includes: The task decomposition unit is used to decompose the tender document generation task into multiple sub-task nodes with dependencies, based on the structured content of the tender document and the requirements of the tender document. The task allocation and scheduling unit is used to manage the execution order, triggering conditions, and computing resource allocation of each subtask node; The data flow and processing unit is used to manage data transfer, format conversion, and consistency processing between various subtask nodes; The results integration unit is used to integrate the outputs of each subtask node into a complete tender document.

[0019] In a more preferred embodiment, the task decomposition unit is configured to decompose the tender document generation task into at least the following six sub-task nodes: technical solution writing node, budget preparation node, personnel allocation node, project management node, compliance check node, and format proofreading node. Among them, the technical solution writing node, budget preparation node, personnel allocation node and project management node are defined as content generation nodes that can be executed in parallel, and the compliance check node and format proofreading node are defined as quality control nodes that depend on the output of the content generation node. The task allocation and scheduling unit is triggered based on the following logical management node: For a content generation node, the triggering condition is that the input data it depends on is ready; the input data includes the structured requirements parsed from the tender document and the output data of the upstream node (i.e., the node that the current node depends on and that was executed before it); For a quality control node, the trigger condition is that the output data of one or more content generation nodes corresponding to it is ready.

[0020] In a more preferred embodiment, the task allocation and scheduling unit is configured to dynamically calculate the scheduling priority and resource quota of each sub-task node based on the task complexity factor and the task urgency factor. Task complexity factor: Pre-configured based on node type, with technical solution writing nodes and budget preparation nodes configured to have higher computational priority and more resource quotas; Task urgency factor: dynamically calculated based on the relative relationship between the node creation time and the total deadline for tender generation calculated by the system, thereby increasing the scheduling priority of nodes that are close to timeout.

[0021] In a more preferred embodiment, the data transfer and processing unit is configured as follows: When transmitting data between nodes, the output data of the upstream node is automatically converted according to the input requirements of the downstream node (specifically, the format conversion is for the constituent elements of the tender document, such as: converting text data to DOCX or PDF format, converting table data to Excel format, and converting chart data to PNG or SVG format). During data transmission, data consistency verification is performed, which includes checking for logical conflicts in cross-node data and comparing the consistency of key values. When a contradiction is found during verification, the problematic data and the context of the contradiction are fed back to the relevant node or system log.

[0022] In a more preferred embodiment, the AI ​​workflow further includes an exception handling unit configured to execute a tiered exception handling strategy: For temporary errors such as insufficient computing resources or network connection interruptions, an automatic retry mechanism is initiated; If there is a data format error or a business logic error, the task is triggered to roll back to the nearest success checkpoint of the node or the upstream dependent node, and rescheduled for execution after the input conditions are corrected. When automatic retry or rollback fails, a structured manual intervention request report is generated, which includes the error type, the node where it occurred, the associated data context, and an error root cause inference based on log analysis.

[0023] In a more preferred embodiment, the AI ​​workflow is configured to execute according to the following process: The technical solution writing node and the budget preparation node are executed in parallel to generate the technical solution document and the budget table, respectively. The personnel configuration node and the project management node are executed in parallel or sequentially to generate team configuration documents and project management plans; Once the output data of each content generation node (technical solution writing, budget preparation, personnel allocation, project management) is ready, the compliance check node is triggered to independently verify the compliance of the output content of that content generation node. After all the outputs of the content generation nodes have passed compliance verification, the result integration unit summarizes the verified outputs of the content generation nodes into a draft of the tender document and triggers the format proofreading node. The format proofreading node performs unified typesetting and format optimization on the draft of the tender document to generate a tender document with a standardized format. The result integration unit receives the format specification document output by the format proofreading node and generates a complete tender document.

[0024] In a preferred embodiment, the automatic optimization and content proofreading module includes: The grammar and spelling check unit integrates spelling check tools (such as Hunspell or Aspell), combines a basic dictionary with a domain-specific dictionary for matching, and checks and corrects spelling and basic grammar errors in the tender document text. The logical consistency check unit integrates a text similarity model (such as BERT or RoBERTa) to calculate the semantic similarity and contradiction between different chapters, paragraphs and sentences in the tender document, and performs a full-document logical consistency check. For example, it can identify description contradictions between the technical solution chapter and the implementation solution chapter, detect whether the logical support relationship between the project objectives and subsequent resource allocation is reasonable, and find repetitive arguments across chapters. The automated format verification and correction unit parses the tender document structure using a document structure parsing tool (such as Apache POI, python-docx, or PyPDF2) and automatically performs verification and correction based on a pre-stored format specification rule library associated with the selected tender template. The verification and correction items include: standardization of chapter title numbering sequence and font style, uniformity of table and figure caption format, and standardized settings of headers, footers, and page numbers.

[0025] In a preferred embodiment, the multi-objective evaluation and scoring module calculates the comprehensive score of the tender document according to the following formula: Overall score = w1 × completeness score of tender document + w2 × compliance score + w3 × rationality score of technical solution + w4 × cost-effectiveness score + w5 × technical risk assessment; Among them, the weight coefficients w1 to w5 are dynamic weight coefficients that are dynamically determined based on the analysis of the current tender documents, and w1+w2+w3+w4+w5=1; The weighting coefficients w1 to w5 are obtained by performing the following process: (1) Analyze the scoring criteria section of the tender document, identify all explicit scoring items and their corresponding scores; map each scoring item to five evaluation dimensions: completeness of tender document, compliance, rationality of technical solution, cost-effectiveness, and technical risk assessment, and accumulate the scores mapped to the same dimension; normalize the accumulated scores of all dimensions to obtain an explicit weighted benchmark vector W_base = (w1_base, w2_base, w3_base, w4_base, w5_base), where the sum of each component is 1; (2) Perform natural language processing on the full text of the tender document. Based on the pre-set evaluation dimension-keyword mapping knowledge base, identify and statistically analyze the occurrence of preset keywords associated with each evaluation dimension. Calculate the total keyword weight score for each evaluation dimension using a weighted statistical algorithm (such as TF-IDF), and then normalize the result to obtain an implicit demand intensity vector I = ( i 1, i 2, i 3, i 4, i 5), where each component represents the intensity of the tenderer's potential attention to each dimension, and the sum of the components is 1; (3) For each evaluation dimension x, x takes values ​​from 1 to 5, calculate its demand compensation value Δ. x = i x - w x _base; Different weighting adjustment strategies are applied based on the compensation value: If Δ x If a preset reinforcement threshold T is set, then this dimension is determined to be a dimension with implicit strong demand, and its initial weight w is determined. x 'Calculated as: w x ' = w x _base+λ·Δ x , where λ> 1 is the preset enhancement coefficient; Otherwise, its initial weight w x 'Calculated as: w x ' = w x_base + μ·Δ x , where 0 <μ≤1 is the preset conventional compensation coefficient; (4) Normalize the initial weights w1', w2', w3', w4', w5' obtained by calculating the five evaluation dimensions according to the above rules, so that their sum is 1. The final output is the dynamic weight coefficients w1, w2, w3, w4, w5.

[0026] In the above content, the preset enhancement threshold T, enhancement coefficient λ, and conventional compensation coefficient μ can be adaptively configured according to the type of bidding project or the training results of historical data.

[0027] In a more preferred embodiment, the tender document completeness score is achieved by evaluating whether the tender document includes the required sections and the completeness of the content of the required sections; the required sections include technical solutions, project management, personnel allocation, quality assurance, cost analysis, risk assessment, time schedule and appendices.

[0028] In one embodiment, the scoring rules for the completeness of the technical solution section include: A complete analysis including project background, technical approach, implementation plan, and acceptance criteria will receive 20-25 points. A score of 15-19 is awarded for a lack of acceptance criteria or an insufficient description of the technical approach and implementation plan. Only a description of the technical route is provided, with other parts missing. This section will receive 10-14 points. Content that is severely missing or completely irrelevant will receive a score of 0-9.

[0029] In one embodiment, the scoring rules for the completeness of the project management section include: A complete description of the project management framework, project plan, schedule, and management methods will earn 12-15 points. The project management structure and schedule are unclear and lack key components, scoring 8-11 points; The project management section is severely lacking, scoring 4-7 points. If the content is completely missing or irrelevant, you will receive 0-3 points.

[0030] In one embodiment, the completeness scoring rules for the staffing section include: Clearly listing the project leader and core team members, and providing detailed background and responsibilities, earns 8-10 points; Missing team members or unclear job descriptions will receive a score of 5-7. The staffing is severely incomplete, scoring 0-4 points.

[0031] In one embodiment, the scoring rules for the completeness of the quality assurance section include: A comprehensive description of quality control measures and management system that conforms to industry standards will earn 8-10 points. The description of quality control measures is incomplete or insufficient, and the score is 5-7. If the quality assurance section is missing or irrelevant, a score of 0-4 will be awarded.

[0032] In one embodiment, the scoring rules for the completeness of the cost analysis section include: A detailed breakdown of the cost structure and budget allocation, which aligns with the actual needs of the project, earns 8-10 points. Unclear cost structure or incomplete budget allocation: 5-7 points; A score of 0-4 is awarded if the cost analysis section is missing or completely irrelevant.

[0033] In one embodiment, the scoring rules for the completeness of the risk assessment section include: A detailed description of project risks and countermeasures, along with specific solutions, will earn 8-10 points. Incomplete risk descriptions or lack of corresponding measures will receive a score of 5-7. A score of 0-4 is awarded if the risk assessment section is missing or irrelevant.

[0034] In one embodiment, the scoring rules for the completeness of the time schedule section include: Clearly listing the detailed timeline of the project and attaching a progress chart will earn 8-10 points; The timeline is not clearly described and the progress chart is missing, so it scores 5-7 points. The timeline section is severely lacking, so it scores 0-4 points.

[0035] In one embodiment, the completeness scoring rules for appendices and other required sections include: The supplementary documents and attachments are complete and meet the additional requirements of the tender document, earning 8-10 points; Incomplete attachments or unmet additional requirements will receive 5-7 points. The appendix is ​​severely missing, so it will receive 0-4 points.

[0036] In one embodiment, the bid integrity score is also based on the degree of compliance with the word count requirements of each chapter; wherein, 5 points are deducted for technical solution chapters with fewer than 1,000 words and 10 points are deducted for chapters with fewer than 500 words.

[0037] In one embodiment, the tender document completeness score is also based on the degree of compliance with required diagrams; wherein, 5 points are deducted for a lack of a technical roadmap or implementation plan diagram in the technical solution section, and 3 points are deducted for poor quality diagrams.

[0038] In a more preferred embodiment, the compliance scoring is achieved by automatically comparing the tender document content with a pre-set database of regulatory clauses; the database includes national laws and regulations, industry standards, and specific requirements of the tender documents; the automated comparison is performed using the TF-IDF algorithm or the BERT model for text similarity analysis. The system identifies violations in the tender document content and their severity based on the comparison results.

[0039] In one embodiment, the compliance score is deducted based on the severity of the violation: Minor violations: If the violation does not have a significant impact on the actual implementation of the project, 1-3 points will be deducted; Moderate violation: The violation affects the feasibility or compliance of the project, but does not affect the overall compliance of the project, and deducts 4-7 points; Serious violation: Violation of the terms that leads to the project's inability to be executed or significant legal risks will result in a deduction of 8-10 points.

[0040] In a more preferred embodiment, the cost-benefit score is achieved in the following manner: The project budget is compared with industry benchmark data to calculate the budget deviation. Based on this budget deviation, the scoring rules could be as follows: a budget deviation within ±5% earns 90-100 points; a budget deviation within ±10% earns 80-90 points; a budget deviation within ±20% earns 70-80 points; and a budget deviation exceeding 20% ​​results in a deduction of points. Analyze whether the proportions of labor costs, material costs, and equipment costs in the total cost are within the reasonable range for the corresponding industry. Based on the comparison and analysis results, cost optimization items are identified and optimization suggestions are generated.

[0041] In one embodiment, the reasonable proportion range is, for example, 30%-40% for software development projects and 20%-40% for material costs; if the project involves equipment procurement, then equipment costs account for 60%-70%.

[0042] In a more preferred embodiment, the technical risk assessment is achieved in the following manner: By analyzing the tender document text using natural language processing technologies (such as BERT) and automatically identifying technical feasibility risks, implementation time risks, resource dependence risks, or integration risks based on preset keyword matching; For each identified risk, assess its impact score (1-10 points) and probability of occurrence score (1-10 points). Calculate the risk value for each risk using the formula: Risk Value = Impact Score × Probability of Occurrence Score. Set a risk threshold (e.g., 25 points), and items with a risk value exceeding this threshold are defined as high-risk items; Scoring is based on the number of high-risk items: 0-1 high-risk items score 90-100 points; 2 high-risk items score 70-89 points; 3 high-risk items score 50-69 points; 4 or more high-risk items score below 50 points.

[0043] In a preferred embodiment, the intelligent auxiliary review module, based on the evaluation results of the multi-objective evaluation and scoring module, parses the tender document text using natural language processing technology and compares it with a pre-set review standard knowledge base and regulatory clause database to automatically generate a structured review report containing problem identification, compliance judgment, and specific modification suggestions.

[0044] Secondly, the present invention provides an intelligent tender document generation method, which is applied to a system as described in any of the first aspects, the method comprising the following steps: The tender documents are received, and natural language processing technology is used to parse the tender documents, extract key information, and structure the key information into standardized data. Based on the aforementioned key information, tender document content is automatically generated through a large language model and AI workflow, including: automatically selecting a suitable tender document template from a preset template library based on project features extracted from the tender documents using a matching algorithm; and triggering the execution of the AI ​​workflow to generate tender document content based on the selected tender document template and the standardized data, wherein the AI ​​workflow is configured to schedule the large language model and coordinate the collaborative execution of multiple sub-task nodes. The generated tender document content undergoes automated syntax, logic consistency, and format verification and correction. Based on the content analysis of the tender documents, the weight coefficients of multiple evaluation dimensions are dynamically determined as dynamic weight coefficients. Based on the dynamic weight coefficients and the preset scoring rules, the optimized tender documents are comprehensively evaluated and quantitatively scored. Based on the results of the quantitative scoring, structured review feedback is generated; Export the final tender document and the review feedback as a document in the specified format.

[0045] In a more preferred embodiment, the step of automatically selecting a suitable tender template from a preset template library using a matching algorithm includes: The tender document is converted into a feature vector, and the matching degree between the feature vector and the feature vector of each template in the template library is calculated to obtain a candidate template set; When there are multiple candidate templates in the candidate template set with a matching degree higher than a preset threshold, the following decision is made: Obtain the static performance attributes of each candidate template, including: the success rate of the candidate template in historical projects, the average review score of the tender documents generated based on the candidate template, and the average cost deviation of historical projects that applied the template. Extract dynamic demand preference features from the tender documents, including: using natural language processing technology to count the total frequency of occurrence of the first set of preset keywords associated with the technological innovation dimension and the total frequency of occurrence of the second set of preset keywords associated with the cost control dimension; and generating a quantitative preference vector to characterize the degree of emphasis on technological innovation and cost control in the tender documents based on the total frequency of occurrence of the first set of preset keywords and the second set of preset keywords. A predefined decision rule model is applied to select the final template from the candidate template set based on the static performance attributes and the dynamic demand preferences; The decision rule model is configured as follows: When the quantitative preference vector representing the emphasis on technological innovation in the tender documents is higher than the first threshold, the technical efficiency index of each candidate template is calculated according to the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, and the candidate template with the highest technical efficiency index is selected; where α and β are preset weight coefficients, and α + β = 1; When the quantitative preference vector represents the emphasis on cost control in the tender document, which is higher than the second threshold, the candidate template with the lowest average cost deviation in historical projects is selected first.

[0046] In a more preferred embodiment, the step of scheduling the large language model and coordinating the collaborative execution of multiple sub-task nodes in the AI ​​workflow includes: The task of generating tender documents is decomposed into multiple sub-task nodes with dependencies. These sub-task nodes include a technical solution writing node, a budget preparation node, a personnel allocation node, a project management node, a compliance check node, and a format proofreading node. Among them, the technical solution writing node, budget preparation node, personnel allocation node, and project management node are defined as content generation nodes. The technical solution writing node and the budget preparation node are executed in parallel. The personnel configuration node and the project management node can be scheduled and executed in parallel or sequentially. After each content generation node has finished outputting data, the compliance check node is triggered to perform compliance verification on the output content of that content generation node. After all the outputs of the content generation nodes have passed compliance verification, the format proofreading node is triggered to perform typesetting and format optimization on the integrated tender draft; Integrate the effective outputs from each node to generate a complete tender document.

[0047] In a more preferred embodiment, the step of dynamically determining the weight coefficients of each evaluation dimension based on the content analysis of the tender documents includes the following steps: (1) Analyze the scoring criteria section of the tender document, identify all explicit scoring items and their corresponding scores; map each scoring item to five evaluation dimensions: completeness of tender document, compliance, rationality of technical solution, cost-effectiveness, and technical risk assessment, and accumulate the scores mapped to the same dimension; normalize the accumulated scores of all dimensions to obtain an explicit weighted benchmark vector W_base = (w1_base, w2_base, w3_base, w4_base, w5_base), where the sum of each component is 1; (2) Perform natural language processing on the full text of the tender document. Based on the pre-set evaluation dimension-keyword mapping knowledge base, identify and statistically analyze the occurrence of preset keywords associated with each evaluation dimension. Calculate the total keyword weight score for each evaluation dimension using a weighted statistical algorithm (such as TF-IDF), and then normalize the result to obtain an implicit demand intensity vector I = ( i 1, i 2, i 3, i 4, i 5), where each component represents the intensity of the tenderer's potential attention to each dimension, and the sum of the components is 1; (3) For each evaluation dimension x, x takes values ​​from 1 to 5, calculate its demand compensation value Δ. x = i x - w x _base; Different weighting adjustment strategies are applied based on the compensation value: If Δ x If a preset reinforcement threshold T is set, then this dimension is determined to be a dimension with implicit strong demand, and its initial weight w is determined. x 'Calculated as: w x ' = w x _base+λ·Δ x , where λ> 1 is the preset enhancement coefficient; Otherwise, its initial weight w x 'Calculated as: w x ' = w x _base + μ·Δ x , where 0 <μ≤1 is the preset conventional compensation coefficient; (4) Normalize the initial weights w1', w2', w3', w4', w5' obtained by calculating the five evaluation dimensions according to the above rules, so that their sum is 1. The final output is the dynamic weight coefficients w1, w2, w3, w4, w5.

[0048] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the intelligent tender generation method as described in any of the second aspects.

[0049] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The technical solution of this invention can improve the automation level and overall quality of tender document preparation. Through an intelligent generation module based on a large language model and AI workflow, the system can automatically parse tender documents and generate structured tender document content, reducing the workload of manual writing and integration, and helping to improve efficiency, especially when dealing with tender documents with a lot of content.

[0050] Through the automatic optimization and proofreading module, the system can perform grammatical error correction, logical contradiction detection, and format standardization on tender documents, effectively solving common problems such as inconsistent styles and disordered formats in multi-person collaboration, thereby improving the overall professional quality of the documents.

[0051] The system uses a multi-dimensional evaluation model that includes completeness, compliance, technical rationality, cost-effectiveness, and technical risks to quantitatively score tender documents, thereby helping users identify omissions and potential deficiencies in the content and providing a reference for targeted optimization. Attached Figure Description

[0052] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the architecture of the intelligent tender document generation system according to a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of the template selection unit according to a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the AI ​​workflow according to a preferred embodiment of the present invention; Figure 4 This is a schematic diagram of the intelligent bid document generation method of the intelligent bid document generation system according to a preferred embodiment of the present invention. Detailed Implementation

[0054] To make the above and other features and advantages of the present invention clearer, the invention will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art and are exemplary only, not restrictive. Example

[0055] Figure 1 This is a schematic block diagram of the architecture of an intelligent tender document generation system according to a preferred embodiment of the present invention. Figure 1 As shown, the system specifically includes the following modules: an intelligent document generation module, an automatic optimization and content proofreading module, a multi-objective evaluation and scoring module, an intelligent auxiliary review module, and an automated export module. The intelligent document generation module automatically generates bid content based on the input tender document; the automatic optimization and content proofreading module automatically verifies and corrects the grammar, logical consistency, and format of the bid content; the multi-objective evaluation and scoring module comprehensively evaluates and quantitatively scores the bid content processed by the automatic optimization and content proofreading module; the multi-objective evaluation and scoring module is configured to dynamically determine the weight coefficients of each evaluation dimension based on the content analysis of the tender document, and calculate the comprehensive score based on the dynamic weights and preset scoring rules; the intelligent auxiliary review module generates structured review feedback based on the evaluation results of the multi-objective evaluation and scoring module; and the automated export module exports the optimized bid content and the review feedback as a document in a specified format. These modules work collaboratively to achieve full automation from tender document parsing to high-quality bid generation and evaluation.

[0056] The following sections provide detailed introductions to some of the modules: 1. Intelligent document generation module The intelligent document generation module is the core of the system, responsible for automatically generating a structured draft tender document based on the tender documents uploaded by the user. Specifically, it includes a tender document parsing unit, a template selection unit, and a content generation unit. The tender document parsing unit extracts key information from the tender document and structures it into standardized data. The template selection unit selects a suitable tender document template from a pre-set template library based on the key information. This template selection unit is configured to: obtain a candidate template set by calculating the matching degree between the tender document and templates in the template library; and determine the final template from the candidate template set by comprehensively considering data related to historical project bidding results and the technical or cost-focused requirements analyzed from the tender document. The content generation unit, based on the selected template and the standardized data, schedules a large language model and combines it with an AI workflow to generate the tender document content. The AI ​​workflow is configured to decompose the tender document generation task into multiple dependent sub-task nodes for scheduled execution.

[0057] 1.1 Tender Document Analysis Unit For example, a user uploads a tender document (usually in PDF or DOCX format) for a "Smart Park Data Center Construction" project. The tender document parsing unit first uses document parsing tools (such as PyPDF2 or python-docx) to extract the raw text. Then, a pre-trained natural language processing model is used to perform in-depth parsing of the text, extracting key structured information, including: project description, technical requirements, budget scope, qualification requirements, scoring criteria, and delivery time. This information is then organized into a standard data structure (e.g., JSON format) for subsequent modules to use.

[0058] 1.2 Template Selection Unit The template selection unit is responsible for selecting the most suitable tender document template for the current bidding project from the preset template library.

[0059] The template library is a key component of the system, responsible for selecting appropriate templates for different types of tender documents. The template library is constructed and categorized according to the following criteria: By Industry: Based on the industry type involved in the tender documents, the templates in the template library are categorized by industry. Common industries include: ① Construction Engineering: mainly includes engineering construction, schedule planning, and project budgeting; ② IT Services: covers software development, hardware procurement, technical solutions, and after-sales service requirements; ③ Equipment Procurement: mainly involves equipment technical specifications, supplier qualifications, delivery time, and price.

[0060] Classification by Project Size: Based on the size of the project, common classification criteria include: Large projects: Involving multiple fields and multiple suppliers, usually requiring a more complex tender document structure, including detailed project planning and risk assessment; Medium projects: Suitable for medium-sized projects, with templates including technical solutions, budget allocation, and delivery cycles; Small projects: Relatively simple, mainly including simplified content such as project requirements, delivery time, and cost control.

[0061] Each template is designed with industry characteristics and project scale in mind to ensure that it can accurately match different bidding requirements.

[0062] like Figure 2 As shown in this embodiment, the template selection unit includes a feature extraction subunit, a candidate generation subunit, and an intelligent decision-making subunit.

[0063] 1.2.1 Feature Extraction Subunit The feature extraction subunit is used to extract core keywords from the parsed tender document data using a first natural language processing model (such as the TF-IDF algorithm) and simultaneously analyze the deep semantics of the tender document using a second natural language processing model (such as the BERT semantic model); the keyword features and semantic features are then fused to generate a comprehensive feature vector of the tender document.

[0064] Feature extraction is a crucial step in the template selection mechanism, aiming to extract key information from tender documents and accurately identify project requirements. The specific process for feature extraction is as follows: Text preprocessing includes stop word removal, punctuation cleaning, and case conversion.

[0065] Keyword extraction: The core keywords in the tender documents are extracted using the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm. TF-IDF helps the system identify important terms in the tender documents, such as project type, budget, and technical requirements.

[0066] Semantic feature extraction: Semantic analysis is performed using large language models (such as BERT) to obtain deep semantic information from the tender documents. The BERT model can help the system understand the specific requirements of the project, such as key factors like budget range, technical difficulty, and delivery time.

[0067] Feature vector representation: Finally, the system will convert the tender document into a feature vector based on the extracted keywords and semantic information, and match it with templates in the template library.

[0068] 1.2.2 Candidate Generation Subunits The candidate generation subunit uses a cosine similarity algorithm to calculate the matching degree between the comprehensive feature vector of the tender document and the feature vectors of each template pre-stored in the template library. Templates are sorted from high to low matching degree, and those with matching degrees higher than a preset threshold (e.g., 0.7) are selected to form an initial candidate template set. For example, two candidate templates may be selected: template A (matching degree 0.88) and template B (matching degree 0.85).

[0069] 1.2.3 Intelligent Decision-Making Subunit The intelligent decision-making subunit is used to execute the following intelligent decision-making process to determine the final selected tender template when there are multiple candidate templates (two or more) in the initial candidate template set: a) Obtain the static performance attributes of each candidate template, including: the success rate of the candidate template in historical projects, the average review score of the tender documents generated based on the candidate template, and the average cost deviation of historical projects that applied the template.

[0070] b) Extract dynamic demand preference features from the tender documents. Specifically, using natural language processing technology, the total frequency of occurrence of the first set of preset keywords (such as "innovation", "patent", "originality", "technological advantage", etc.) associated with the "technological innovation" dimension and the total frequency of occurrence of the second set of preset keywords (such as "cost", "budget", "quotation", "control", "saving") associated with the "cost control" dimension are statistically analyzed. Based on the total frequency of occurrence of the first set of preset keywords and the total frequency of occurrence of the second set of preset keywords, a quantitative preference vector is generated to characterize the degree of emphasis on technological innovation and cost control in the tender documents.

[0071] c) Apply a predefined decision rule model, based on the static performance attributes and dynamic demand preferences, to output the final template from the candidate template set; wherein, the decision rule model is constructed based on the following business principles: when the bidding party focuses on technological innovation, the advancement and feasibility of the technical solution are key to winning the bid, so the template with the best overall performance (winning rate and solution quality) in historical technology projects should be selected first; when the bidding party focuses on cost control, the accuracy and controllability of the budget are core, so the template with the smallest historical cost execution deviation should be selected first.

[0072] Specifically, the decision rule model is configured as follows: When the quantitative preference vector representing the emphasis on technological innovation in the tender documents is higher than the first threshold, the technical efficiency index of each candidate template is calculated according to the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, and the candidate template with the highest technical efficiency index is selected; where α and β are preset weight coefficients, and α + β = 1; When the quantitative preference vector represents the emphasis on cost control in the tender document, which is higher than the second threshold, the candidate template with the lowest average cost deviation in historical projects is selected first.

[0073] It should be noted that when applying the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, the historical success rate and the average review score can be calculated using the same numerical scale. For example, the average review score on a percentage basis can be normalized to the same interval [0, 1] as the historical success rate.

[0074] Hypothesis: A user uploads a tender document for a "Smart Park Data Center Construction" project.

[0075] The system, through the feature extraction subunit, parsed the keywords in the document, which included "cloud computing", "green energy saving" and "redundant design", and semantically emphasized "technological advancement and reliability".

[0076] After matching with the template library, an initial candidate template set is generated, which includes: Template A (Technical): Matching degree 0.88, historical success rate 35%, average review score 85 points, but historical projects have an average cost overrun of 8%.

[0077] Template B (Cost-based): Matching degree 0.85, historical success rate 28%, average review score 80 points, historical project average cost control is good (deviation +2%).

[0078] The system analyzes the entire tender document through an intelligent decision-making subunit, counting the total frequency of keywords related to "technological innovation" (such as "innovation," "intelligent," and "advanced technology") and keywords related to "cost control" (such as "budget," "savings," and "cost-effectiveness"). The analysis reveals that technical keywords appear extremely frequently, and the technical solution accounts for as much as 60% of the score in the tender document's evaluation criteria, indicating that the tendering party prioritizes technological innovation.

[0079] The system's pre-set decision rule model is as follows: If the tender documents emphasize technological innovation, then the technical effectiveness index of each candidate template is calculated. Specifically, after normalizing the average review score (out of 100), the technical effectiveness index is calculated according to the formula: Technical Effectiveness Index = α × Historical Success Rate + β × Average Review Score. In this example, let α = 0.6 and β = 0.4. The average review score of template A (85 points) is normalized to 0.85, and the score of template B (80 points) is normalized to 0.8. Then, the technical effectiveness index of template A = 0.6 × 0.35 + 0.4 × 0.85 = 0.55, and the technical effectiveness index of template B = 0.6 × 0.28 + 0.4 × 0.8 = 0.488. If cost control is emphasized, then the candidate template with the lowest historical average cost deviation is selected first. In this example, it is determined that the tendering party emphasizes technological innovation, and template A has a higher technical effectiveness index; therefore, the system ultimately selects template A as the generation basis.

[0080] 1.3 Content Generation Unit The content generation unit, based on the selected template and the parsed structured bidding requirements, schedules a large language model to generate the specific content of each part of the tender document. As a specific implementation in this embodiment, the scheduled large language model can be the Qwen-32B model and can be deployed in a Docker environment on the Dify platform. The generation process is coordinated and managed through a structured AI workflow.

[0081] like Figure 3 As shown in this embodiment, the AI ​​workflow includes: a task decomposition unit, a task allocation and scheduling unit, a data transfer and processing unit, and a result integration unit.

[0082] The system includes a task decomposition unit, which decomposes the tender document generation task into multiple dependent sub-task nodes based on the structured content of the tender document and the requirements of the tender documents; a task allocation and scheduling unit, which manages the execution order, triggering conditions, and computing resource allocation of each sub-task node; a data flow and processing unit, which manages the data transfer, format conversion, and consistency processing between each sub-task node; and a result integration unit, which integrates the outputs of each sub-task node into a complete tender document.

[0083] 1.3.1 Task Decomposition Unit The task decomposition unit is configured to decompose the tender document generation task into at least the following six sub-task nodes: Technical solution writing stage: Responsible for generating the technical section of the tender document, including project background, technical route, implementation plan, acceptance criteria, etc.

[0084] Budget preparation stage: This involves writing the budget section of the tender document, including cost analysis, fund allocation, and budget rationality.

[0085] Personnel configuration node: Define the project's personnel structure and clarify the responsibilities and backgrounds of the project leader and core team members.

[0086] Project management milestones include project schedule, resource allocation, and risk management.

[0087] Compliance checkpoint: Ensure that the content of the tender documents complies with the requirements of the tender documents and relevant regulations.

[0088] Format proofreading node: Responsible for checking and optimizing the format of the tender documents, such as margins, fonts, and layout.

[0089] The technical solution writing node, budget preparation node, personnel allocation node, and project management node are defined as content generation nodes, which can be executed in parallel. The trigger condition for a content generation node is that the input data it depends on is ready; the input data includes the structured requirements parsed from the tender document and the output data of the upstream node (i.e., the node that the current node depends on and that was executed before it).

[0090] The compliance check node and format proofreading node are defined as quality control nodes, which depend on the output of their corresponding content generation nodes. The trigger condition for a quality control node is that the output data of one or more of its corresponding content generation nodes is ready. For example, after the technical solution writing node is completed, its output immediately triggers the compliance check node to verify that part of the content.

[0091] Specifically, each workflow node represents an independent subtask, and the node's functions, inputs, and outputs are as follows: (1) Technical solution nodes: Function: To write the technical solution section according to the requirements of the tender documents.

[0092] Input: Technical requirements from the tender documents, historical project data, and industry standards.

[0093] Output: Technical solution documents (including background analysis, technical roadmap, implementation plan, etc.).

[0094] (2) Budget preparation milestones: Function: Calculate and write the budget section.

[0095] Input: Project requirements, historical budget data, industry standards.

[0096] Output: Budget sheet (including cost analysis, fund allocation, etc.).

[0097] (3) Staffing node: Function: Identify project team members and write the personnel configuration section.

[0098] Input: Personnel requirements, project scale, and job responsibilities as specified in the tender documents.

[0099] Output: Staffing documents (including team structure, responsibility assignments, etc.).

[0100] (4) Project Management Nodes: Function: Write the project management section, including schedule planning, resource allocation, risk control, etc.

[0101] Input: Time requirements and resource requirements in the tender documents.

[0102] Output: Project management plan (including schedule, resource allocation, risk assessment, etc.).

[0103] (5) Compliance checkpoints: Function: Check whether the tender documents comply with relevant regulations and tender document requirements.

[0104] Input: Draft tender documents, relevant regulations and tender document requirements.

[0105] Output: Compliance report (indicating whether each requirement is met and marking any non-compliant parts).

[0106] (6) Format proofreading node: Function: Check whether the layout and format of the tender documents meet the specifications.

[0107] Input: Draft tender document, format specification template.

[0108] Output: Optimized tender document.

[0109] 1.3.2 Task Allocation and Scheduling Unit The task allocation and scheduling unit dynamically allocates computing resources based on node complexity (e.g., allocating more computing resources to technical solutions and budget nodes) and urgency (prioritizing tasks nearing their deadlines).

[0110] Specifically, the task allocation and scheduling unit executes a multi-factor-based dynamic resource allocation algorithm, which dynamically schedules computing resources according to a predefined strategy: Task complexity factor: Pre-configure weights based on node type; for example, assign higher computing priority and more resource quotas to nodes that write technical solutions and those that compile budgets. Task urgency factor: Based on the relative relationship between the node creation time and the total deadline for bid generation calculated by the system, the scheduling priority of nodes that are about to time out is dynamically calculated and increased.

[0111] 1.3.3 Data Flow and Processing Unit The data transfer and processing unit is configured as follows: When transmitting data between nodes, the output data of the upstream node is automatically converted according to the input requirements of the downstream node (specifically, the format conversion is for the constituent elements of the tender document, including converting text data to DOCX or PDF format, converting table data to Excel format, and converting chart data to PNG or SVG format). During data transmission, data consistency verification is performed, which includes checking for logical conflicts in cross-node data and comparing the consistency of key values. When a contradiction is found during verification, the problematic data and the context of the contradiction are fed back to the relevant node or system log.

[0112] 1.3.4 Result Integration Unit After all nodes have been executed, the result integration unit will integrate the technical solution documents, budget sheets, team introductions, management plans, etc. generated by each node according to the template structure to form a complete initial draft of the tender document.

[0113] In a preferred embodiment, the AI ​​workflow further includes an exception handling unit, which executes the following processing strategy when a node fails: Automatic retry mechanism: For temporary errors (such as insufficient computing resources, network connection interruption, etc.), the system will automatically retry the node, up to 3 times; Error logging: When a node fails to execute continuously, the system will generate a detailed error log, recording the reasons for the failure and the tasks affected; Task rollback: After a task fails, the task is rolled back to the nearest success checkpoint of the current node or an upstream dependent node, resources are reallocated and re-execution is triggered; Manual intervention prompt: If the system cannot automatically correct the error, it will notify the project manager to intervene manually, providing detailed information about the error and a solution.

[0114] In this embodiment, the task allocation and scheduling unit is configured to perform task scheduling and triggering according to the following process: The technical solution writing node and the budget preparation node are executed in parallel to generate the technical solution document and the budget table, respectively. The personnel configuration node and the project management node are executed in parallel or sequentially to generate team configuration documents and project management plans; Once the output data of each content generation node (technical solution writing, budget preparation, personnel allocation, project management) is ready, the compliance check node is triggered to independently verify the compliance of the output content of that content generation node. After all the outputs of the content generation nodes have passed compliance verification, the result integration unit summarizes the verified outputs of the content generation nodes into a draft of the tender document and triggers the format proofreading node. The format proofreading node performs unified typesetting and format optimization on the draft of the tender document to generate a tender document with a standardized format. The result integration unit receives the format specification document output by the format proofreading node and generates a complete tender document.

[0115] 2. Automatic optimization and content proofreading module The automatic optimization and content proofreading module uses intelligent algorithms to perform grammatical, logical, and format checks and optimizations on the initial draft of the tender document, improving its standardization and professionalism. Specifically, it includes a grammar and spelling check unit, a logical consistency check unit, and an automated format verification and correction unit.

[0116] Grammar and spelling check unit: Integrates spelling check tools (such as Hunspell or Aspell), uses basic dictionaries and domain-specific (such as architecture, IT, equipment procurement, etc.) custom dictionaries to check and correct spelling and basic grammar errors in the entire tender document.

[0117] Logical consistency check unit: Integrates a text similarity model (such as BERT or RoBERTa) to perform semantic consistency checks on various parts of the tender document. For example, it checks for contradictions or inconsistencies between chapters and paragraphs, for duplicate content, or for inconsistencies in information.

[0118] Automated Formatting Validation and Correction Unit: This unit uses document structure parsing tools (such as Apache POI, python-docx, or PyPDF2) to parse the document structure of the tender documents, automatically identifying the organization of elements such as chapters, paragraphs, headings, and tables. The system will check whether it conforms to specific formatting requirements (such as the font and size of chapter headings, and the arrangement of tables).

[0119] 3. Multi-objective assessment and scoring module The multi-objective evaluation and scoring module performs multi-dimensional quantitative scoring on the optimized bid. In this embodiment, the system presets five core evaluation dimensions: bid completeness, compliance, rationality of the technical solution, cost-effectiveness, and technical risk assessment. Specifically, the comprehensive score of the bid is calculated according to the following formula: Overall score = w1 × completeness score of tender document + w2 × compliance score + w3 × rationality score of technical solution + w4 × cost-effectiveness score + w5 × technical risk assessment; Among them, the weight coefficients w1 to w5 are dynamic weight coefficients determined dynamically based on the analysis of the current tender documents, and w1+w2+w3+w4+w5=1.

[0120] 3.1 Determination of Dynamic Weights The weighting coefficients w1 to w5 are obtained by performing the following process: (1) Analyze the scoring criteria section of the tender document, identify all explicit scoring items and their corresponding scores; map each scoring item to five evaluation dimensions: completeness of tender document, compliance, rationality of technical solution, cost-effectiveness, and technical risk assessment, and accumulate the scores mapped to the same dimension; normalize the accumulated scores of all dimensions to obtain an explicit weighted benchmark vector W_base = (w1_base, w2_base, w3_base, w4_base, w5_base), where the sum of each component is 1; (2) Perform natural language processing on the full text of the tender document. Based on the pre-set evaluation dimension-keyword mapping knowledge base, identify and statistically analyze the occurrence of preset keywords associated with each evaluation dimension. Calculate the total keyword weight score for each evaluation dimension using a weighted statistical algorithm (such as TF-IDF), and then normalize the result to obtain an implicit demand intensity vector I = ( i 1, i 2, i 3, i 4, i 5), where each component represents the intensity of the tenderer's potential attention to each dimension, and the sum of the components is 1; (3) For each evaluation dimension x, x takes values ​​from 1 to 5, calculate its demand compensation value Δ. x = i x - w x _base; Different weighting adjustment strategies are applied based on the compensation value: If Δ x If a preset reinforcement threshold T is set, then this dimension is determined to be a dimension with implicit strong demand, and its initial weight w is determined. x 'Calculated as: w x ' = w x _base+λ·Δ x , where λ> 1 is the preset enhancement coefficient; Otherwise, its initial weight w x 'Calculated as: w x ' = w x _base + μ·Δ x , where 0 <μ≤1 is the preset conventional compensation coefficient; (4) Normalize the initial weights w1', w2', w3', w4', w5' obtained by calculating the five evaluation dimensions according to the above rules, so that their sum is 1. The final output is the dynamic weight coefficients w1, w2, w3, w4, w5.

[0121] In the above content, the preset enhancement threshold T, enhancement coefficient λ, and conventional compensation coefficient μ can be adaptively configured according to the type of bidding project or the training results of historical data.

[0122] The following example, using the tender document for the "Smart Park Data Center Construction" project in this embodiment, further illustrates the process for determining the aforementioned dynamic weights.

[0123] Step 1: Analyze and display the scoring criteria.

[0124] The system analyzed the bidding documents and found that the "Evaluation Method" section clearly stipulated: Technical solution: 40 points; Business price: 35 cents; Project Management: 15 points; Company Qualification: 10 points; The score for "technical risk" was not mentioned.

[0125] The system maps the above scoring items to five dimensions: Map the “Technical Solution (40 points)” to the dimension of the rationality of the technical solution.

[0126] Map the "Business Quotation (35 points)" to the cost-benefit dimension.

[0127] Map “Project Management (15 points)” to the Project Management section under the Tender Completeness dimension.

[0128] Map "Enterprise Qualification (10 points)" to the compliance dimension.

[0129] The "Technology Risk" dimension was mapped to a score of 0.

[0130] The scores for each dimension are as follows: completeness (15), compliance (10), rationality of technical solution (40), cost-effectiveness (35), and technical risk (0). The total score is 100.

[0131] After normalization, the explicit weighted baseline vector is obtained: W_base = (0.15, 0.10, 0.40, 0.35, 0.00).

[0132] Step 2: Analyze the intensity of implicit demand.

[0133] The system performs natural language processing on the full text of the tender document. An example of a pre-defined evaluation dimension-keyword mapping knowledge base is shown below: Key words for technical risk assessment dimensions: {reliability, stability, redundancy, backup, failure rate, security}; Key words for the rationality of technical solutions: {advanced, innovative, intelligent, efficient}; ...

[0134] Statistical analysis and calculations revealed that terms such as "reliable," "safe," "redundant," and "failure rate" appeared frequently in the tender documents. Through TF-IDF weighted and normalized calculations, the implicit demand intensity vector was obtained: I = (0.08, 0.07, 0.30, 0.25, 0.30).

[0135] This reveals that the bidding party has a strong implicit need for "technical risk assessment". i (5 = 0.30), significantly higher than its explicit benchmark w5_base = 0.00.

[0136] Step 3: Calculate compensation and adjust preliminary weights.

[0137] Assume the system's preset parameters are: enhancement threshold T = 0.15, enhancement coefficient λ = 1.8, and conventional compensation coefficient μ = 0.6.

[0138] Calculate the demand compensation value Δ for the five dimensions. x = i x - w x _base, and apply the corresponding strategy: Integrity (x=1): Δ1 = 0.08 - 0.15 = -0.07 (not greater than T). Calculated using standard compensation: w1' = 0.15 + 0.6 × (-0.07) = 0.108; Compliance (x=2): Δ2 = 0.07 - 0.10 = -0.03 (not greater than T). Calculated using standard compensation: w2' = 0.10 + 0.6 × (-0.03) = 0.082; Technical solution (x=3): Δ3 = 0.30 - 0.40 = -0.10 (not greater than T). Calculated using conventional compensation: w3' = 0.40 + 0.6 × (-0.10) = 0.340; Cost-benefit (x=4): Δ4 = 0.25 - 0.35 = -0.10 (not greater than T). Calculated using conventional compensation: w4' = 0.35 + 0.6 × (-0.10) = 0.290; Technical risk (x=5): Δ5 = 0.30 - 0.00 = 0.30 (> T = 0.15). Trigger reinforcement compensation: w5' = 0.00 + 1.8 × 0.30 = 0.540.

[0139] The initial weight set is obtained as: W' = (0.108, 0.082, 0.340, 0.290, 0.540).

[0140] Step 4: Normalize to obtain the final weights.

[0141] Normalize the initial weight set W' (so that the sum is 1): The total sum is: 0.108 + 0.082 + 0.340 + 0.290 + 0.540 = 1.360 The final dynamic weight coefficients are: w1 = 0.108 / 1.360 ≈ 0.079; w2 = 0.082 / 1.360 ≈ 0.060; w3 = 0.340 / 1.360 ≈ 0.250; w4 = 0.290 / 1.360 ≈ 0.213; w5 = 0.540 / 1.360 ≈ 0.397.

[0142] Final result: W_final ≈ (0.08, 0.06, 0.25, 0.21, 0.40) Through this process, the system intelligently identifies the tenderer's strong implicit need for "technical risk assessment" and accordingly increases its weight from an explicit 0.00 to a final 0.40, making it the most important assessment dimension. Simultaneously, the weights of other dimensions are dynamically adjusted based on the matching of their explicit and implicit needs. This allows the subsequent comprehensive bid scoring model to accurately align with the tenderer's true and complete intentions, guiding the generation of a bid that better meets the tenderer's expectations.

[0143] 3.2 Specific Scoring for Each Dimension After the dynamic weights are determined, the system scores each dimension according to the following scoring criteria: 3.2.1 Evaluation Criteria for Tender Document Completeness Check whether the tender documents contain the following required sections and assess the level of detail in their content.

[0144] The required sections include: technical solutions, project management, staffing, quality assurance, cost analysis, risk assessment, timeline, appendices, and other requirements.

[0145] The scoring criteria for each chapter are as follows: (1) Technical Solution Section (25 points in total) Project background analysis: 5 points; Technical route description: 5 points; Implementation plan: 10 points; Acceptance criteria: 5 points.

[0146] The scoring details are as follows: 20-25 points: Complete, including project background analysis, technical approach, implementation plan, acceptance criteria, etc. 15-19 points: Lack of acceptance criteria, or insufficient description of the technical route and implementation plan; 10-14 points: Only the technical route is described; other parts are missing. 0-9 points: The content is seriously missing or completely irrelevant.

[0147] (2) Project Management Chapter (15 points) Project organizational structure: 5 points; Project schedule: 5 points; Management processes and methods: 5 points.

[0148] The scoring details are as follows: 12-15 points: Provide a complete description of the project management framework, project plan, schedule, and management methods; 8-11 points: The project management structure and schedule are unclear, and key components are missing; 4-7 points: The project management section is severely lacking; 0-3 points: Completely missing or irrelevant.

[0149] (3) Staffing section (10 points) Project Leader: 3 points; Core Team Members: 4 points; Personnel Responsibilities and Professional Background: 3 points.

[0150] The scoring details are as follows: 8-10 points: Clearly list the project leader and core team members, and provide detailed background and responsibilities; 5-7 points: Some team members are missing or their responsibilities are not clearly described; 0-4 points: Staffing is severely incomplete.

[0151] (4) Quality Assurance Section (10 points) Quality control measures: 5 points; Quality management system description: 5 points.

[0152] The scoring details are as follows: 8-10 points: Provides a comprehensive description of quality control measures and management systems, conforming to industry standards; 5-7 points: The description of quality control measures is incomplete or insufficient; 0-4 points: Quality assurance section is missing or irrelevant.

[0153] (5) Cost Analysis Section (10 points) Cost structure: 5 points; Budget allocation: 5 points.

[0154] The scoring details are as follows: 8-10 points: The cost structure and budget allocation are detailed and meet the actual needs of the project; 5-7 points: Unclear cost structure or incomplete budget allocation; 0-4 points: Cost analysis is missing or completely irrelevant.

[0155] (6) Risk Assessment Section (10 points) Risk identification: 4 points; Response measures: 6 points.

[0156] The scoring details are as follows: 8-10 points: Describe the project risks and countermeasures in detail, and provide specific solutions; 5-7 points: Risk description is incomplete or lacks corresponding countermeasures; 0-4 points: Risk assessment section is missing or irrelevant.

[0157] (7) Time Schedule Planning Section (10 points in total) Detailed timeline: 5 points; Project implementation schedule: 5 points.

[0158] The scoring details are as follows: 8-10 points: Clearly list the detailed timeline of the project and attach a progress chart; 5-7 points: The timeline is not clearly described, and the progress chart is missing; 0-4 points: The timeline is severely lacking.

[0159] (8) Appendix and other requirements sections (10 points in total) Related attachments: 5 points; Compliance with additional requirements of the tender documents: 5 points.

[0160] The scoring details are as follows: 8-10 points: The supplementary documents and attachments are complete and meet the additional requirements of the tender documents; 5-7 points: The attachments are incomplete, or some additional requirements are not met; 0-4 points: The appendix is ​​severely missing.

[0161] In one embodiment, the tender document completeness score is also based on the degree to which the word count requirements for each chapter are met. Each chapter should meet the minimum word count requirement, and points will be deducted if the word count requirement is not met. For example, the technical solution chapter must have at least 1,500 words; less than 1,000 words will result in a deduction of 5 points, and less than 500 words will result in a deduction of 10 points; the project management chapter must have at least 1,000 words; less than 800 words will result in a deduction of 3 points, and less than 500 words will result in a deduction of 6 points; the personnel configuration chapter must have at least 500 words; less than 500 words will result in a deduction of 2 points, and less than 200 words will result in a deduction of 5 points; the quality assurance chapter must have at least 500 words; less than 400 words will result in a deduction of 2 points, and less than 200 words will result in a deduction of 5 points, etc.

[0162] In one embodiment, the tender document completeness score is also based on the degree of compliance with required charts and diagrams. The tender document must contain a certain number of charts and diagrams, and the content and quality of these charts and diagrams have a significant impact on the completeness score. For example, the technical solution section must contain at least one technical roadmap and one implementation plan diagram; a missing technical roadmap or implementation plan diagram deducts 5 points, and poor-quality or unclear charts deduct 3 points. The project management section must contain at least one project schedule diagram; a missing project schedule diagram deducts 3 points, and unclear or non-standard format charts deduct 2 points. The risk assessment section must contain at least one risk assessment matrix; a missing risk assessment matrix deducts 3 points, and poor-quality charts deduct 2 points, etc.

[0163] 3.2.2 Compliance Scoring Criteria The goal of compliance scoring is to ensure that the tender documents comply with relevant laws, policies, and tender document requirements. The scoring criteria involve a step-by-step review of each section of the tender documents to ensure compliance with laws, industry standards, and tender document requirements.

[0164] The compliance scoring is achieved by automatically comparing the tender document content with a pre-built database of regulatory clauses; the database of regulatory clauses includes national laws and regulations, industry standards, and specific requirements of the tender documents; the automated comparison is completed by text similarity analysis using the TF-IDF algorithm or the BERT model.

[0165] In one embodiment, the compliance score is deducted based on the severity of the violation: Minor violations: The violation does not have a significant impact on the actual execution of the project, and 1-3 points will be deducted; for example, some attachments were not provided in accordance with the format requirements of the tender documents, but the content of the attachments met the requirements; Moderate violations: Violations affect the feasibility or compliance of the project, but do not affect the overall compliance of the project, resulting in a deduction of 4-7 points; for example, failure to provide certain legal supporting documents may affect the approval process, but does not affect the execution of the project; Serious violations that result in project failure or significant legal risks will result in a deduction of 8-10 points. For example, tender documents may not comply with relevant labor law provisions, fail to protect employees' basic rights, or contract terms may not comply with contract law.

[0166] After verifying the regulatory clauses, the system can generate a detailed compliance report, listing the compliance check results and violations for each regulatory clause. The report should include: the regulatory clause being checked, whether it meets the requirements (compliant / non-compliant), the specific content of the violation, and a description of the points deducted.

[0167] Based on the results of regulatory clause checks, the system can provide intelligent correction suggestions to help bidders revise their bids to ensure compliance with all requirements. For example, if a missing legal document is found, the system can prompt, "Please provide relevant labor law compliance documents."

[0168] 3.2.3 Cost-benefit scoring criteria The goal of cost-benefit scoring is to assess the rationality of a project's resource allocation and budget control, and whether it can maximize cost-effectiveness. Scoring criteria include analysis of the project budget and cost structure, and the generation of optimization recommendations.

[0169] In one embodiment, the cost-benefit assessment is performed through the following process: (1) Verification of the reasonableness of the project budget The system automatically compares the total project budget listed in the tender documents with a pre-set industry benchmark database and calculates the budget deviation rate. The specific scoring rules are as follows: A budget deviation rate within ±5% is considered highly reasonable and scores 90-100 points. A budget deviation rate within ±10% is considered reasonable, and a score of 80-90 points is awarded. Budget deviation rate within ±15% is considered basically reasonable, with a score of 70-80 points; Budget deviation rate within ±20%, score 60-70 points; Budget deviation rate exceeding 20% ​​is considered unreasonable, resulting in a score below 60 points, and points will be deducted as appropriate.

[0170] (2) Analysis of the rationality of cost structure The system analyzes the proportion of key cost items such as labor costs, material costs, and equipment costs in the total cost of the tender document and determines whether they are within the reasonable range for the corresponding industry. For example, for software development projects, the reasonable proportions in the industry are usually: 30%-40% for labor costs and 20%-40% for material costs; for equipment procurement projects, equipment costs usually account for 60%-70%.

[0171] The scoring rules for cost structure rationality are as follows: All key cost items are within the industry standard range, scoring 90-100 points; If only one cost item deviates from the standard range by less than 10%, the score is 80-90 points; If only one cost item deviates from the standard range by more than 10%, the score is 70-80 points; If multiple cost percentages deviate from the standard range simultaneously, the score will be 60-70 points, and corresponding points will be deducted.

[0172] (3) Cost optimization suggestion generation and scoring The system automatically identifies cost items that may contain waste or unreasonable expenditures by comparing the details in the project budget with industry benchmarks and historical project data. These items are marked as "optimization opportunities," and specific optimization suggestions are generated. For example: If labor costs account for too high a proportion, the system suggests optimizing the team structure or considering partial outsourcing; If material costs are too high, the system suggests optimizing the supply chain or bulk purchasing. If the equipment investment is substantial, the system suggests assessing the feasibility of leasing instead of purchasing.

[0173] The quality of the generated optimization suggestions is also included in the score: The system successfully identified significant areas for optimization and provided effective and specific suggestions, scoring 80-100 points. If no obvious areas for improvement are identified, or if the generated suggestions are not sufficiently reasonable, the score will be 60-80 points. No optimization suggestions were generated, and the score was below 60.

[0174] (4) Example of comprehensive cost-benefit score To make it easier to understand, we will take an IT service project as an example to illustrate the comprehensive scoring process of the cost-benefit dimension.

[0175] Assume a project's total budget is 2 million yuan, of which: labor costs are 1 million yuan (50%), material costs are 400,000 yuan (20%), and equipment costs are 500,000 yuan (25%). Industry benchmarks are: labor costs 40%, material costs 20%-40%, and equipment costs 25%-30%.

[0176] The system analysis is as follows: a) Budget rationality: The total budget deviates from the average budget of similar projects in the industry (assuming it is RMB 1.9 million) by about 5.3%, which is within the ±10% range. This item scores 85 points.

[0177] b) Cost structure: The proportion of labor costs (50%) exceeds the industry standard upper limit (40%) by 10 percentage points, triggering the rule of "one cost deviating by more than 10%", and the score for this item is 75 points; the proportion of material and equipment costs meets the standard.

[0178] c) Optimization Suggestion: The system identified that the proportion of labor costs was too high and automatically generated an optimization suggestion: "Review the staffing and outsource some non-core module development tasks to reduce the proportion of labor costs." The suggestion is specific and feasible, and this item scores 90 points.

[0179] Finally, the system calculates a comprehensive cost-benefit score for this dimension based on preset weights (e.g., budget rationality 40%, cost structure rationality 40%, optimization suggestions 20%) or a dynamic weighting algorithm. Based on the scoring example above, the comprehensive score is likely to be in the range of 80-85 points, and the system will also provide feedback to the user regarding "human resource cost optimization" as a key improvement point.

[0180] 3.2.4 Technical Risk Assessment Standards The goal of technical risk assessment is to systematically identify and quantify the potential risks in the technical solutions presented in the tender documents, and to objectively score their severity. This technical risk assessment is achieved through the following process: (1) Risk identification and classification The system utilizes natural language processing techniques (such as the BERT model) to deeply analyze the technical solutions and implementation schemes sections of the tender document, and automatically identifies potential risks by matching them against a pre-defined risk keyword library (such as "unverified" and "external dependencies"). The identified risks are automatically categorized into the following predefined risk types: Technical feasibility risk: refers to whether the technical route, method or tool adopted is mature and reliable, and whether there are technical obstacles that are difficult to achieve; Implementation time risk: This refers to whether factors such as the complexity of the technical solution and resource dependence may cause project delays. Resource dependency risk: refers to whether a project depends on certain key resources or external suppliers; Integration risks refer to the interface, protocol, or data compatibility issues that may arise when integrating a new system or component with a customer's existing environment or system.

[0181] (2) Risk quantification For each identified risk, the system will conduct a quantitative assessment based on a pre-set rule knowledge base. This assessment includes two dimensions: impact level (I) and probability of occurrence (P).

[0182] The Impact Level (I) is used to assess the magnitude of the negative impact that the risk would have on project objectives (such as functionality, performance, and delivery) should it occur. A scale of 1-10 is used. 1-3 points: Minor impact, causing only minor functional deviations or slight delays.

[0183] 4-6 points: Moderate impact, resulting in some major functions not meeting standards or moderate delays in the project.

[0184] 7-10 points: Significant impact, leading to failure of core project objectives, significant delays, or legal / safety liabilities.

[0185] The probability of occurrence (P) is used to assess the likelihood of the risk actually occurring within the project lifecycle. A 1-10 scale is used. 1-3 points: Low probability, based on existing information and technical conditions, the possibility of occurrence is extremely low.

[0186] 4-6 points: Medium probability, there is a certain possibility of it happening, and it needs to be paid attention to.

[0187] 7-10 points: High probability, based on the existing conditions, it is very likely to happen.

[0188] The system calculates the risk value for each risk using the formula: Risk Value (R) = Impact Level (I) × Probability of Occurrence (P). The higher the risk value, the more priority the risk needs to be given to addressing it.

[0189] (3) Risk scoring rules The system defines risk items whose risk value (R) exceeds a preset threshold (e.g., R > 25) as "high-risk items". The final score for the technical risk dimension is determined based on the number of identified "high-risk items", according to the following rules: 0-1 High Risk: Score 90-100. This indicates that the risk control of the technical solution in the tender document is excellent, or that only an acceptablely minor risk exists.

[0190] Two high-risk items: scores 70-89. This indicates that the solution carries moderate risks that require attention, and targeted optimization is recommended.

[0191] Three high-risk items: scores 50-69. This indicates that the proposed solution carries a high risk and may pose a substantial threat to project execution; revision is strongly recommended.

[0192] Four or more high-risk items: a score below 50 indicates significant risks and major hidden dangers, requiring thorough evaluation and major modifications.

[0193] (4) Example of evaluation process The following example of a hypothetical IT system integration project illustrates the complete process of the risk assessment and scoring described above: After analyzing the tender documents, the system identified and quantified the following three risks: Technical feasibility risk: Impact level I = 8 points (significant impact), probability of occurrence P = 7 points (high probability), risk value R1 = 8 × 7 = 56; Implementation time risk: Impact level I = 6 points (medium impact), probability of occurrence P = 6 points (medium probability), risk value R2 = 6 × 6 = 36; Resource dependence risk: Impact level I = 4 points (slight impact), probability of occurrence P = 4 points (low probability), risk value R3 = 4 × 4 = 16.

[0194] Assume the system's preset high-risk threshold is R > 25.

[0195] Risk item 1 (R1=56) > 25, is judged as a high-risk item.

[0196] Risk item 2 (R2=36) > 25, and is therefore classified as a high-risk item.

[0197] Risk item 3 (R3=16) ≤ 25, was not identified as a high-risk item.

[0198] The system identifies two high-risk items. According to the scoring rules, the corresponding score range for these two high-risk items is 70-89 points. The system can further analyze the specific distribution of risk values ​​to provide a more precise score (e.g., 78 points).

[0199] The system automatically generates structured review feedback, such as: "Technical risk assessment score: 78 points. Two high risks were identified: 1) Feasibility risk in the application of new technologies (risk value 56); 2) Delay risk due to tight project schedule (risk value 36). It is recommended to supplement the technology verification plan with a detailed schedule buffer and communication mechanism." 3.2.5. Example of Comprehensive Assessment and Report Generation To illustrate the actual workflow and output of the multi-objective evaluation and scoring module more specifically, the following example uses the automated review of a "Industrial Visual Positioning and Inspection System" project proposal to demonstrate how the system generates a complete structured evaluation report. In this example, it is assumed that the evaluation dimension weights determined by the system's dynamic analysis are: w1=30%, w2=20%, w3=20%, w4=15%, w5=15%.

[0200] Step 1: Independent analysis and scoring of each dimension Based on a pre-set scoring rule library, the system automatically analyzes the tender document content across five dimensions, generating the following sub-results: (1) Bid document completeness score (weight w1, assumed to be 30% after dynamic calculation) Analysis and scoring criteria: System testing revealed that the tender documents covered core content such as technical solutions (vision system functions, positioning accuracy, interface protocols) and safety protection design (emergency stop button, safety circuits), but key sections were missing: a detailed project schedule (timelines for development, testing, deployment, etc.) was not provided; detailed cost estimates and budget allocations were lacking (such as the specific cost composition of equipment procurement and software development); and specific terms for after-sales service and technical support (such as response time and maintenance period) were not clearly defined.

[0201] Rating: 75 / 100.

[0202] System-generated optimization suggestions: Supplement the project schedule (it is recommended to present it in Gantt chart format).

[0203] Add a detailed cost breakdown (e.g., hardware costs account for approximately 30%, software development accounts for 40%, and human resources implementation accounts for 30%).

[0204] A new chapter has been added on after-sales service and technical support.

[0205] (2) Compliance score (weight w2, assumed to be 20%) Analysis and scoring criteria: The system confirmed by comparing the tender documents with the regulatory database that the tender documents mentioned and complied with several key standards: SMEMA (Surface Mount Device Communication Protocol) interface; GigE Vision and USB3 Vision mainstream industrial camera protocols; safety design (emergency stop button, equipotential grounding) complies with IEC 62004-1 standard.

[0206] Rating: 85 / 100.

[0207] The system generates optimization suggestions: It is recommended to clearly list the certification information (such as CE and FCC certificate numbers) obtained by key equipment in the tender documents to further enhance the credibility of compliance proof.

[0208] (3) Rationality score of technical solution (weight w3, assumed to be 20%) Analysis and scoring criteria: The system evaluation concluded that the technical solution achieved a good balance between innovation and maturity, possessing significant advantages: measurement accuracy (±0.05mm) is superior to the industry average standard (±0.1mm); it adopts a modular, multi-functional positioning tool design; and it is compatible with mainstream PLCs and industrial communication protocols. However, the system also identified potential weaknesses due to insufficient justification: it did not provide detailed explanations of how the core hardware selection (specific models and parameters of the camera and light source) met the claimed high-precision requirements; and it lacked performance verification data under complex application scenarios (such as varying lighting conditions).

[0209] Rating: 80 / 100.

[0210] System-generated optimization suggestions: Add a hardware parameter specification table to list key parameters (such as camera resolution, frame rate, and light source wavelength).

[0211] Supplement with third-party test reports or typical customer case data (e.g., "99.9% recognition accuracy was achieved on a certain automotive parts production line").

[0212] (4) Cost-benefit score (weight w4, assumed to be 15%) Evaluation and scoring criteria: System analysis revealed that although the tender document mentioned cost control ideas such as "modular design to save secondary development costs" and "custom algorithm encapsulation to reduce development cycle", it did not provide any specific cost data and analysis, making it impossible to verify the rationality of the budget and evaluate the return on investment.

[0213] Rating: 60 / 100.

[0214] System-generated optimization suggestions: Provide itemized cost estimates (e.g., hardware procurement approximately ¥200,000, software development approximately ¥150,000, project implementation and labor approximately ¥100,000).

[0215] Conduct a simple return on investment analysis (e.g., "After the implementation of the solution, it is expected to save ¥500,000 in manual inspection costs annually, with an investment payback period of approximately X months").

[0216] (5) Technical risk assessment (weight w5, assumed to be 15%) Analysis and scoring criteria: The system uses natural language processing and keyword matching to identify several technical risks mentioned in the tender proposal but not adequately addressed. These risks mainly include: image distortion may affect accuracy, but no correction solution is provided; for the identification of low-contrast targets, no specific optimization strategy for the OCR algorithm is mentioned; and key hardware (such as industrial cameras) may malfunction, but no redundant backup mechanism is designed.

[0217] Rating: 65 / 100.

[0218] The system generates optimization suggestions: The system suggests supplementing the risk response measures table, for example:

[0219] Step 2: Calculate the weighted composite score The system calculates a weighted comprehensive score based on dynamic weights: Overall score = 75 × 0.30 + 85 × 0.20 + 80 × 0.20 + 60 × 0.15 + 65 × 0.15 = 74.25 points Step 3: Generate a structured review report The system automatically integrates the above analysis to generate a complete review report, the core contents of which include: 1) The overall score is 74.25, and the overall evaluation is "good, with significant room for improvement".

[0220] 2) The scores, weights, and key comments for the five dimensions are clearly displayed in a table format.

[0221] 3) Automatically extract the highlights of the tender documents, such as "advanced technical solutions and good compliance foundation".

[0222] 4) The specific problems and suggestions identified in each dimension are listed in a structured manner and can be directly linked to the original text of the tender document.

[0223] 5) Users can further enhance the professionalism and readability of their tender documents by adding visual elements such as technology roadmaps, cost structure pie charts, and project schedule Gantt charts.

[0224] Example final conclusion: The system report states that by adopting the above optimization suggestions and supplementing and improving the content of the tender document, its predicted comprehensive score can be effectively improved to over 85 points, thereby significantly enhancing its competitiveness in winning the bid.

[0225] This example fully demonstrates how the system integrates intelligent multi-dimensional analysis, quantitative scoring, and actionable feedback suggestions. It not only replaces the tedious and subjective checks in traditional manual reviews, but also uncovers deep-seated logical flaws and risks that are easily overlooked by humans, and generates structured and standardized improvement guidelines, thus enhancing the efficiency and quality of bid self-optimization.

[0226] 4. Intelligent Assisted Review Module The intelligent auxiliary review module parses the tender document text using natural language processing technology and automatically generates a structured review report containing problem identification, compliance assessment, and specific modification suggestions by comparing it with a pre-built review standard knowledge base and regulatory clause database. For example, it might state: "Section 3.2, 'using XX brand servers,' does not include a corresponding quote in the budget details table, posing a risk of cost omissions," and provide modification suggestions.

[0227] 5. Automated Export Module The automated export module will ultimately export the optimized and evaluated tender documents, as well as the detailed review report, into PDF, DOCX, and other formats that meet the requirements of the tendering party with one click. Example

[0228] This embodiment provides a method for generating intelligent tender documents applied to the intelligent tender document generation system described in Embodiment 1. Figure 4As shown, the method includes the following steps: Step S1: Receive and parse the tender documents.

[0229] Users upload bidding documents (such as PDF or DOCX formats) through the system interface. The system uses natural language processing technology to automatically parse the documents, extracting key information including project description, technical requirements, qualification requirements, budget range, scoring criteria, and delivery time, and structuring this information into standardized data (such as JSON format) that can be processed internally.

[0230] Step S2: Intelligent generation of tender document content.

[0231] Based on the key information extracted in step S1, the system automatically generates a draft proposal through a large language model and a structured AI workflow. This step specifically includes: Step S21, Intelligent Template Selection: Based on the project type, budget scale, technical field and industry characteristics extracted from the tender documents, the system automatically selects the most suitable tender document template from the preset template library through the matching algorithm (feature extraction, similarity calculation, intelligent decision-making) as described in Example 1.

[0232] Step S22, Initial Content Generation: Based on the selected template and extracted key information, the large language model (such as Qwen-32B) is used to generate the basic content of each part of the tender document.

[0233] Step S23, Workflow Collaboration Improvement: The AI ​​workflow coordinates the parallel or sequential execution of multiple sub-task nodes (such as technical solution writing, budget preparation, personnel allocation, and project management), triggering quality control nodes such as compliance checks and format verification, ultimately integrating them to generate a complete tender document. The specific process is consistent with that described in "1.3 Content Generation Unit" of Example 1.

[0234] Step S3: Automatic optimization and content proofreading.

[0235] The system performs automated, in-depth optimization and proofreading on the initial draft of the tender document generated in step S2, including: Use spell checkers and domain dictionaries for grammar and spelling correction.

[0236] Use a text similarity model to perform a full-document-wide logical consistency check to identify contradictory or repetitive content.

[0237] Based on the formatting specifications of the selected template, the document's chapter numbering, font style, headers and footers, figure and table captions, and other formatting information are automatically checked and corrected.

[0238] S4. Multi-objective comprehensive evaluation and quantitative scoring.

[0239] The system performs a multi-dimensional quantitative evaluation of the optimized bid. First, as described in "3.1 Determination of Dynamic Weights" in Example 1, the system dynamically determines the weight coefficients (w1 to w5) for five dimensions: bid completeness, compliance, rationality of technical solutions, cost-effectiveness, and technical risk assessment, by analyzing the explicit scoring criteria and implicit textual requirements of the tender documents. Then, based on a pre-set scoring rule library (as described in "3.2 Specific Scores for Each Dimension" in Example 1), the system calculates the bid's score for each dimension. Finally, the overall score is calculated using the formula: Comprehensive Score = Σ (w1 to w5) i ×Dimensional Scoring i ), calculate the final comprehensive score of the tender document.

[0240] S5. Generate intelligent review feedback.

[0241] Based on the evaluation results of step S4, the system uses natural language processing technology to extract key information from the tender document text and compares it with the review standard knowledge base and regulatory clause database to automatically generate a structured review report. This report not only includes the overall score and conclusions but also details the tender document's core strengths, specific issues (located to specific chapters), compliance assessments, and actionable modification suggestions.

[0242] S6, Automated Export.

[0243] The final bid document, which has been optimized, evaluated, and includes a review report, can be exported with one click to a document format that meets the requirements of the bidding party, such as PDF or DOCX.

[0244] The method described in this embodiment significantly improves the efficiency, standardization, and quality of bid document preparation by automating and intelligentizing the entire process of bid document generation, and provides bidders with objective and quantitative optimization guidance.

[0245] In summary, this invention automates the entire tender document generation process by constructing an intelligent document generation module and an AI workflow engine. First, the tender document parsing unit automatically extracts and structures key information using natural language processing technology. Then, the content generation unit schedules a large language model to automatically write the tender document content. The AI ​​workflow decomposes the tender document generation task into multiple logically dependent sub-task nodes, and the task allocation and scheduling unit manages and schedules the execution order and computing resources of these nodes. This automated process reduces the workload of manual input and integration, improves processing efficiency, and lowers the error rate.

[0246] To address the adaptability issue in template selection, this invention introduces an intelligent decision-making mechanism. The template selection unit first filters candidate templates through feature vector similarity matching; its intelligent decision-making subunit further comprehensively analyzes static performance attributes such as historical bidding success rates and average review scores of each candidate template, as well as dynamic demand preferences for technological innovation or cost control derived from analyzing the frequency of keywords in the tender documents. Based on a predefined decision rule model, the system outputs the final selected template based on the aforementioned multi-dimensional information. This method, by integrating static attributes and dynamic preferences for decision-making, transforms the template selection process from matching based on a single feature similarity into a multi-factor fusion screening process, thereby improving the accuracy of adaptation to specific tender requirements.

[0247] To ensure the standardization of the output documents, this invention includes an automatic optimization and content proofreading module. This module integrates spell checking tools, a text similarity model, and a document structure parsing tool to perform the following automated processing on the tender document content: correcting grammatical and spelling errors; detecting logical inconsistencies and semantic repetitions across chapters; and automatically verifying and standardizing the formatting of document elements such as chapter titles, figure captions, headers, and footers based on a formatting rule library associated with the selected template. This ensures the consistency and standardization of the output tender document in terms of both content logic and format presentation.

[0248] Regarding the evaluation system, this invention proposes a multi-objective dynamic evaluation model. This model presupposes five core dimensions: completeness, compliance, rationality of the technical solution, cost-effectiveness, and technical risk assessment. The dynamic determination method for its weight coefficients includes: parsing the scoring criteria sections of the tender document to obtain explicit weight benchmarks; performing natural language processing on the entire tender document to analyze the implicit demand intensity; and combining the two methods to calculate the final dynamic weights using a compensation algorithm. Each evaluation dimension has quantified scoring details. This evaluation system can provide a comprehensive score based on dynamic weights and indicate the focus of tender content optimization.

[0249] Through the synergistic effect of the above-mentioned technical means, the intelligent tender document generation system and method constructed by the present invention realizes the whole process processing from tender document parsing, content generation, quality proofreading to comprehensive evaluation, thereby improving the efficiency and quality of tender document generation.

[0250] The specific embodiments of the present invention have been described in detail above, but they are merely examples, and the present invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent transformations and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.

Claims

1. An intelligent tender document generation system, characterized in that, include: The intelligent document generation module is used to automatically generate tender document content based on the input tender documents; The automatic optimization and content proofreading module is used to automatically check and correct the grammar, logical consistency and format of the tender document content; The multi-objective evaluation and scoring module is used to comprehensively evaluate and quantitatively score the tender document content after it has been processed by the automatic optimization and content proofreading module. The multi-objective evaluation and scoring module is configured to: dynamically determine the weight coefficients of each evaluation dimension based on the content analysis of the tender document, and calculate the comprehensive score based on the dynamic weights and preset scoring rules. The intelligent assisted review module is used to generate structured review feedback based on the evaluation results of the multi-objective evaluation and scoring module. The automated export module is used to export the optimized tender document content and the review feedback into a document in a specified format.

2. The intelligent tender document generation system according to claim 1, characterized in that, The intelligent document generation module includes: The tender document parsing unit is used to extract key information from the tender document and structure it into standardized data. The template selection unit is used to select a suitable tender document template from a preset template library based on the key information. The template selection unit is configured to: obtain a candidate template set by calculating the matching degree between the tender document and the templates in the template library, the candidate template set including at least one candidate template, and determine the final template from the candidate template set by comprehensively considering the relevant data of the candidate templates and the bidding results of historical projects, as well as the demand preferences for technology or cost focus analyzed from the tender document. The content generation unit is used to schedule a large language model and combine it with an AI workflow to generate tender content based on the selected template and the standardized data; the AI ​​workflow is configured to decompose the tender generation task into multiple sub-task nodes with dependencies for scheduling and execution.

3. The intelligent tender document generation system according to claim 2, characterized in that, The template selection unit includes: The feature extraction subunit is used to extract the core keywords in the bidding documents through the first natural language processing model, and at the same time to parse the deep semantics of the bidding documents through the second natural language processing model; the keyword features and semantic features are fused to generate a comprehensive feature vector of the bidding documents; The candidate generation subunit is used to calculate the matching degree between the comprehensive feature vector of the tender document and the feature vector of each template pre-stored in the template library using a similarity matching algorithm; sort the templates from high to low according to the matching degree, and select templates with matching degrees higher than a preset threshold to form an initial candidate template set; The intelligent decision-making subunit is used to execute the following intelligent decision-making process to determine the final selected tender template when multiple templates exist in the initial candidate template set: a) Obtain the static performance attributes of each candidate template, including: the success rate of the candidate template in historical projects, the average review score of the tender documents generated based on the candidate template, and the average cost deviation of historical projects that applied the template. b) Extract dynamic demand preference features from the tender documents, specifically: using natural language processing technology, calculate the total frequency of occurrence of the first set of preset keywords associated with the technological innovation dimension and the total frequency of occurrence of the second set of preset keywords associated with the cost control dimension in the tender documents; based on the total frequency of occurrence of the first set of preset keywords and the total frequency of occurrence of the second set of preset keywords, generate a quantitative preference vector to characterize the degree of emphasis on technological innovation and cost control in the tender documents; c) Apply a predefined decision rule model and output the final template from the candidate template set based on the static performance attributes and dynamic demand preferences; The decision rule model is configured as follows: When the quantitative preference vector representing the emphasis on technological innovation in the tender documents is higher than the first threshold, the technical efficiency index of each candidate template is calculated according to the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, and the candidate template with the highest technical efficiency index is selected; where α and β are preset weight coefficients, and α + β = 1. When the quantitative preference vector represents the emphasis on cost control in the tender document above the second threshold, the candidate template with the lowest average cost deviation in historical projects is selected first.

4. The intelligent tender document generation system according to claim 2, characterized in that, The large language model is the Qwen-32B model, deployed in a Docker environment on the Dify platform.

5. The intelligent tender document generation system according to claim 2, characterized in that, The AI ​​workflow includes: The task decomposition unit is used to decompose the tender document generation task into multiple sub-task nodes with dependencies, based on the structured content of the tender document and the requirements of the tender document. The task allocation and scheduling unit is used to manage the execution order, triggering conditions, and computing resource allocation of each subtask node; The data flow and processing unit is used to manage data transfer, format conversion, and consistency processing between various subtask nodes; The results integration unit is used to integrate the outputs of each subtask node into a complete tender document.

6. The intelligent tender document generation system according to claim 5, characterized in that, The task decomposition unit is configured to decompose the tender document generation task into at least the following six sub-task nodes: technical solution writing node, budget preparation node, personnel allocation node, project management node, compliance check node, and format proofreading node; Among them, the technical solution writing node, budget preparation node, personnel allocation node and project management node are defined as content generation nodes that can be executed in parallel, and the compliance check node and format proofreading node are defined as quality control nodes that depend on the output of the content generation node. The task allocation and scheduling unit is triggered based on the following logical management node: For a content generation node, the triggering condition is that the input data it depends on is ready; the input data includes the structured requirements parsed from the tender document and the output data of the upstream node; For a quality control node, the trigger condition is that the output data of one or more content generation nodes corresponding to it is ready.

7. The intelligent tender document generation system according to claim 6, characterized in that, The data transfer and processing unit is configured as follows: When transmitting data between nodes, the output data of the upstream node is automatically converted according to the input requirements of the downstream node. During data transmission, data consistency verification is performed, which includes checking for logical conflicts in cross-node data and comparing the consistency of key values. When a contradiction is found during verification, the problematic data and the context of the contradiction are fed back to the relevant nodes or system logs.

8. The intelligent tender document generation system according to claim 6, characterized in that, The task allocation and scheduling unit is configured to schedule and trigger tasks according to the following process: The technical solution writing node and the budget preparation node are executed in parallel to generate the technical solution document and the budget table, respectively. The personnel configuration node and the project management node are executed in parallel or sequentially to generate team configuration documents and project management plans; Once the output data of each content generation node is ready, the compliance check node is triggered to perform independent compliance verification on the output content of that content generation node. After all the outputs of the content generation nodes have passed compliance verification, the result integration unit summarizes the verified outputs of the content generation nodes into a draft of the tender document and triggers the format proofreading node. The format proofreading node performs unified typesetting and format optimization on the draft of the tender document to generate a tender document with a standardized format. The result integration unit receives the format specification document output by the format proofreading node and generates a complete tender document.

9. The intelligent tender document generation system according to claim 1, characterized in that, The automatic optimization and content proofreading module includes: The grammar and spelling check unit integrates a spelling check tool, combines a basic dictionary with a domain-specific dictionary for matching, and checks and corrects spelling and basic grammar errors in the tender document text. The logical consistency check unit integrates a text similarity model to calculate the semantic similarity and inconsistency between different chapters, paragraphs and sentences in the tender document, and performs a full-document logical consistency check. The format automation verification and correction unit parses the tender document structure using a document structure parsing tool and automatically performs verification and correction based on a pre-stored format specification rule library associated with the selected tender template. The verification and correction items include: standardization of chapter title numbering sequence and font style, uniformity of table and figure caption format, and standardized setting of headers, footers, and page numbers.

10. The intelligent tender document generation system according to claim 1, characterized in that, The multi-objective evaluation and scoring module calculates the comprehensive score of the tender document according to the following formula: Overall score = w1 × completeness score of tender document + w2 × compliance score + w3 × rationality score of technical solution + w4 × cost-effectiveness score + w5 × technical risk assessment; Among them, the weight coefficients w1 to w5 are dynamic weight coefficients that are dynamically determined based on the analysis of the current tender documents, and w1+w2+w3+w4+w5=1; The weighting coefficients w1 to w5 are obtained by performing the following process: (1) Analyze the scoring criteria section of the tender document, identify all explicit scoring items and their corresponding scores; map each scoring item to five evaluation dimensions: completeness of tender document, compliance, rationality of technical solution, cost-effectiveness, and technical risk assessment, and accumulate the scores mapped to the same dimension; normalize the accumulated scores of all dimensions to obtain an explicit weighted benchmark vector W_base = (w1_base, w2_base, w3_base, w4_base, w5_base), where the sum of each component is 1; (2) Natural language processing is performed on the full text of the tender document. Based on the pre-set evaluation dimension-keyword mapping knowledge base, the occurrence of preset keywords associated with each evaluation dimension is identified and statistically analyzed. A weighted statistical algorithm is used to calculate the total keyword weight score for each evaluation dimension, and after normalization, an implicit demand intensity vector I = ( i 1, i 2, i 3, i 4, i 5), where each component represents the intensity of the tenderer's potential attention to each dimension, and the sum of the components is 1; (3) For each evaluation dimension x, x takes values ​​from 1 to 5, calculate its demand compensation value Δ. x = i x - w x _base; Different weighting adjustment strategies are applied based on the compensation value: If Δ x If a preset reinforcement threshold T is set, then this dimension is determined to be a dimension with implicit strong demand, and its initial weight w is determined. x 'Calculated as: w x '= w x _base+λ·Δ x , where λ> 1 is the preset enhancement coefficient; Otherwise, its initial weight w x 'Calculated as: w x ' = w x _base + μ·Δ x , where 0 <μ≤1 is the preset conventional compensation coefficient; (4) Normalize the initial weights w1', w2', w3', w4', w5' obtained by calculating the five evaluation dimensions according to the above rules, so that their sum is 1. The final output is the dynamic weight coefficients w1, w2, w3, w4, w5.

11. The intelligent tender document generation system according to claim 1, characterized in that, The intelligent auxiliary review module, based on the evaluation results of the multi-objective evaluation and scoring module, parses the tender document text using natural language processing technology and compares it with a pre-set review standard knowledge base and regulatory clause database, automatically generating a structured review report that includes problem identification, compliance judgment, and specific modification suggestions.

12. A method for generating intelligent tender documents, characterized in that, The method is applied to the system as described in any one of claims 1-11, and the method includes the following steps: The tender documents are received, and natural language processing technology is used to parse the tender documents, extract key information, and structure the key information into standardized data. Based on the aforementioned key information, tender document content is automatically generated through a large language model and AI workflow, including: automatically selecting a suitable tender document template from a preset template library based on project features extracted from the tender documents using a matching algorithm; and triggering the execution of the AI ​​workflow to generate tender document content based on the selected tender document template and the standardized data, wherein the AI ​​workflow is configured to schedule the large language model and coordinate the collaborative execution of multiple sub-task nodes. The generated tender document content undergoes automated grammar, logical consistency, and format verification and correction. Based on the content analysis of the tender documents, the weight coefficients of multiple evaluation dimensions are dynamically determined as dynamic weight coefficients. Based on the dynamic weight coefficients and the preset scoring rules, the optimized tender documents are comprehensively evaluated and quantitatively scored. Based on the results of the quantitative scoring, structured review feedback is generated; Export the final tender document and the review feedback as a document in the specified format.

13. The intelligent tender document generation method according to claim 12, characterized in that, The step of automatically selecting a suitable tender template from a preset template library using a matching algorithm includes: The tender document is converted into a feature vector, and the matching degree between the feature vector and the feature vector of each template in the template library is calculated to obtain a candidate template set; When there are multiple candidate templates in the candidate template set with a matching degree higher than a preset threshold, the following decision is made: Obtain the static performance attributes of each candidate template, including: the success rate of the candidate template in historical projects, the average review score of the tender documents generated based on the candidate template, and the average cost deviation of historical projects that applied the template. Extract dynamic demand preference features from the tender documents, including: using natural language processing technology to count the total frequency of occurrence of the first set of preset keywords associated with the technological innovation dimension and the total frequency of occurrence of the second set of preset keywords associated with the cost control dimension; and generating a quantitative preference vector to characterize the degree of emphasis on technological innovation and cost control in the tender documents based on the total frequency of occurrence of the first set of preset keywords and the second set of preset keywords. A predefined decision rule model is applied to select the final template from the candidate template set based on the static performance attributes and the dynamic demand preferences; The decision rule model is configured as follows: When the quantitative preference vector representing the emphasis on technological innovation in the tender documents is higher than the first threshold, the technical efficiency index of each candidate template is calculated according to the formula: Technical Efficiency Index = α × Historical Success Rate + β × Average Review Score, and the candidate template with the highest technical efficiency index is selected; where α and β are preset weight coefficients, and α + β = 1. When the quantitative preference vector represents the emphasis on cost control in the tender document above the second threshold, the candidate template with the lowest average cost deviation in historical projects is selected first.

14. The intelligent tender document generation method according to claim 12, characterized in that, The steps of scheduling a large language model and coordinating the collaborative execution of multiple sub-task nodes in the AI ​​workflow include: The task of generating tender documents is decomposed into multiple sub-task nodes with dependencies. These sub-task nodes include a technical solution writing node, a budget preparation node, a personnel allocation node, a project management node, a compliance check node, and a format proofreading node. Among them, the technical solution writing node, budget preparation node, personnel allocation node, and project management node are defined as content generation nodes. The technical solution writing node and the budget preparation node are executed in parallel. The personnel configuration node and the project management node can be scheduled and executed in parallel or sequentially. After each content generation node has finished outputting data, the compliance check node is triggered to perform compliance verification on the output content of that content generation node. After all the outputs of the content generation nodes have passed compliance verification, the format proofreading node is triggered to perform typesetting and format optimization on the integrated tender draft; Integrate the effective outputs from each node to generate a complete tender document.

15. The intelligent tender document generation method according to claim 12, characterized in that, The dynamic determination of the weighting coefficients for each evaluation dimension based on the content analysis of the tender documents includes the following steps: (1) Analyze the scoring criteria section of the tender document, identify all explicit scoring items and their corresponding scores; map each scoring item to five evaluation dimensions: completeness of tender document, compliance, rationality of technical solution, cost-effectiveness, and technical risk assessment, and accumulate the scores mapped to the same dimension; normalize the accumulated scores of all dimensions to obtain an explicit weighted benchmark vector W_base = (w1_base, w2_base, w3_base, w4_base, w5_base), where the sum of each component is 1; (2) Natural language processing is performed on the full text of the tender document. Based on the pre-set evaluation dimension-keyword mapping knowledge base, the occurrence of preset keywords associated with each evaluation dimension is identified and statistically analyzed. A weighted statistical algorithm is used to calculate the total keyword weight score for each evaluation dimension, and after normalization, an implicit demand intensity vector I = ( i 1, i 2, i 3, i 4, i 5), where each component represents the intensity of the tenderer's potential attention to each dimension, and the sum of the components is 1; (3) For each evaluation dimension x, x takes values ​​from 1 to 5, calculate its demand compensation value Δ. x = i x - w x _base; Different weighting adjustment strategies are applied based on the compensation value: If Δ x If a preset reinforcement threshold T is set, then this dimension is determined to be a dimension with implicit strong demand, and its initial weight w is determined. x 'Calculated as: w x ' = w x _base+λ·Δ x , where λ> 1 is the preset enhancement coefficient; Otherwise, its initial weight w x 'Calculated as: w x ' = w x _base + μ·Δ x , where 0 <μ≤1 is the preset conventional compensation coefficient; (4) Normalize the initial weights w1', w2', w3', w4', w5' obtained by calculating the five evaluation dimensions according to the above rules, so that their sum is 1. The final output is the dynamic weight coefficients w1, w2, w3, w4, w5.

16. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the intelligent tender document generation method as described in any one of claims 12-15.

Images