A technical scheme intelligent generation system for informationization project bidding

Abstract

The application discloses a technical scheme intelligent generation system for information-based project bidding, relates to the technical fields of intelligent document generation and information-based project bidding, and extracts and structures core elements of a bidding document through a project demand analysis module, stores five types of entities and associated relations based on a field knowledge graph, combines a performance matching algorithm to complete accurate recall of bidding materials, automatically generates an initial scheme according to a subdivided field chapter template by a scheme intelligent generation engine, accounts for project workloads and costs through a multi-model cross-verification mechanism of a cost intelligent calculation module, executes three-level compliance verification and automatic correction through a multi-dimensional compliance verification module, finally outputs a standard bidding technical document, and simultaneously realizes continuous optimization of system capacity through a supporting entity updating and incremental learning mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent document generation and information technology project bidding technology, specifically an intelligent generation system for technical solutions in information technology project bidding. Background Technology

[0002] For organizations deeply rooted in third-party IT consulting services, since commencing operations, they have provided support services for over 500 information system engineering projects, covering four major areas: software testing, software cost estimation, consulting and design, and performance evaluation. In recent years, bidding demands in areas such as e-government and smart cities have continued to expand, project types have become increasingly segmented, and the differences in qualification requirements, service content, and technical standards among different projects have been continuously increasing. As the core document responding to bidding demands, the efficiency, professionalism, and compliance of the bidding technical proposal directly affect the bidding results and are a crucial manifestation of the organization's core competitiveness.

[0003] Traditional bidding technical solutions are prepared manually. The preparers need to break down the requirements of the bidding documents page by page, manually select and match performance and cases from the historical project database, and manually integrate qualification certificates, service processes, technical solutions and other contents. At the same time, they need to manually complete cost calculation and compliance self-check. For example, the existing technology CN115841369A is an intelligent analysis system for bidding agency, but it does not involve compliance verification and cost calculation in the bidding scenario. This model suffers from several intractable problems. First, material matching is inefficient and inaccurate. Manual retrieval can easily miss highly relevant similar project performance, and the matching degree between performance and project industry, scale, and service content is difficult to guarantee. The preparation of a single complete plan usually takes 3 to 5 days, which cannot support the parallel operation of multiple projects during the concentrated bidding period. Second, cost calculation is inconsistent. Manual calculation relies on individual judgment of functional points and workload, and the calculation results of different compilers generally deviate by more than 15%, making it difficult to control the rationality of the quotation. Third, there are blind spots in compliance verification. Manual verification cannot cover all details such as the validity period of qualifications, the validity of standard versions, and the timeliness of performance. The risk of bid rejection due to compliance oversights is a long-term problem. Fourth, the reuse rate of knowledge assets is low. High-quality solutions and project experience are scattered in various documents and cannot be structured, accumulated, and called upon. Personnel changes will directly lower the overall preparation level.

[0004] In summary, existing manual compilation methods and general document generation tools cannot meet the professional compilation needs of bidding proposals in the field of information technology consulting services, and it is difficult to simultaneously meet the business goals of high efficiency, high accuracy, and high compliance. Therefore, an intelligent generation system for this scenario is needed to improve the technology from the dimensions of material matching, cost calculation, and compliance verification. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent technical solution generation system for information technology project bidding. This system can extract and structure the core elements of bidding documents through a project requirements analysis module, rely on five types of entities and their relationships stored in a domain knowledge graph, and combine performance matching algorithms to accurately recall bidding materials. An intelligent solution generation engine automatically generates initial solutions according to subdivided domain chapter templates. The system calculates project workload and costs through a multi-model cross-validation mechanism in a cost intelligent calculation module, and performs three levels of compliance verification and automatic correction through a multi-dimensional compliance verification module. Finally, it outputs standardized bidding technical documents, while also incorporating entity update and incremental learning mechanisms to continuously optimize system capabilities.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent generation system for technical solutions in information technology project bidding, the system comprising the following components: Project Requirements Analysis Module: The output end is connected to the input end of the intelligent solution generation engine and the intelligent cost calculation module respectively, and is used to extract and output the core elements of the project; Domain Knowledge Graph Module: This module is connected to the intelligent solution generation engine, the multi-dimensional compliance verification module, and the intelligent cost calculation module, respectively, and is used to store five types of entities and their associated mapping relationships: qualifications, performance, technical standards, service processes, and typical cases. Intelligent solution generation engine: The output end connects to the input end of the multi-dimensional compliance verification module, recalls physical materials, and generates an initial technical solution according to the chapter template; Multi-dimensional compliance verification module: Corrects the intelligent generation engine of the output feedback connection solution, which is used to perform three types of compliance verifications: qualification, performance, and technical standards, and trigger automatic correction; The intelligent cost calculation module outputs a solution intelligent generation engine to match the calculation model and output project workload and cost details. Output module: Connects to the final output of the intelligent solution generation engine, used to complete document formatting and output according to bidding specifications.

[0007] Furthermore, the project requirements analysis module incorporates a built-in word segmentation thesaurus for the information-based bidding and tendering field. This thesaurus contains 11,700 structured terms, covering four main categories: qualification names, service types, technical standards, and deliverables. The terms are constructed from 487 historical project documents and 230 current industry standard documents after terminology extraction, deduplication, and manual verification. Upon receiving the requirements document, the module first performs multi-format text extraction, compatible with file parsing interfaces for PDF, DOCX, and WPS document formats. It extracts the full text and filters headers, footers, watermarks, and garbled characters. Then, it further... Word segmentation and stop word filtering are performed on the word segmentation lexicon to remove common function words and meaningless characters. Then, a domain entity recognition model based on BERT-base architecture is fine-tuned to locate core elements. The model uses 320 manually annotated tender documents as the fine-tuning dataset. The annotated entities include six categories: project type, service scope, qualification requirements, delivery requirements, construction period, and budget range. After recognition, semantic normalization is performed on the elements, and standard terms in the lexicon are matched to complete the expression calibration. Finally, a structured element dataset containing three types of fields: element code, element value, and source page number is output.

[0008] Furthermore, the five types of entities in the domain knowledge graph module are all configured with standardized attribute fields. The qualification entity includes four types of attributes: qualification name, level, validity period, and scope of application. The performance entity includes five types of attributes: owner unit, project amount, completion time, service content, and project results. The technical standard entity includes four types of attributes: standard number, version number, release time, and applicable scenarios. The service process entity includes three types of attributes: process steps, corresponding deliverables, and applicable project types. The typical case entity includes three types of attributes: project background, service plan, and implementation effect. The attribute association between entities is automatically established based on the same attribute value. The scenario association is established based on the sub-domain and industry category to which the project belongs. The time sequence association is established based on the project completion time and the standard release time. When a new entity is added, the three types of association degree with the existing entities are automatically calculated. When the association degree exceeds 0.7, the corresponding association relationship is automatically established. The association degree is calculated by weighted summation of attribute overlap and semantic similarity. The weight of attribute overlap is 0.6, and the weight of semantic similarity is 0.4.

[0009] Furthermore, the domain knowledge graph module incorporates a performance matching degree calculation algorithm, which calculates the overall matching degree between a single historical performance record and the current project using the following formula: ,in The overall matching score ranges from 0 to 1. To ensure industry relevance, pre-trained industry semantic word vectors are used to calculate the cosine similarity between the current project and the categories of historical performance. The category system adopts the three-level category standard of the National Economic Industry Classification. When matching across categories, similarity is calculated based on semantic vectors, with values ​​ranging from 0 to 1. The overlap in scale is calculated by dividing the length of the intersection between the budget range and the historical performance amount by the total length of the budget range. A value of 0 is taken when the historical performance amount exceeds the upper or lower limit of the budget range. The value range is 0-1. The service content coverage is calculated as the ratio of the number of overlapping service content items between the two projects to the total number of service items in the current project, with a value ranging from 0 to 1. This is a time decay factor, determined by the difference between the year the performance was completed and the current year. The value increases by 0.05 for every 1 year (1 year), and is fixed at 0.3 when the difference exceeds 6 years. The value ranges from 0 to 0.3. This is a project attribute adjustment coefficient. It varies depending on whether the project involves classified information. For classified projects with prior experience in handling classified matters, the value is 1.05; for projects without prior experience in handling classified matters, the value is 0.8; and for non-classified projects, the value is 1. , , , These are the weighting coefficients. The value is 0.35. The value is 0.3. The value is 0.35. The value is 0.1. The weight coefficient is determined by training a logistic regression model based on 487 historical winning bid data. The training uses the winning bid result as the dependent variable and the four types of matching factors as independent variables. The model fit is 0.78.

[0010] Furthermore, the intelligent cost calculation module incorporates a workload correction algorithm. For software projects, after calculating the baseline workload using the function point method, the adjusted total workload is calculated using the following formula: ,in This represents the adjusted total project workload, expressed in person-months. The baseline workload per function point is calculated by multiplying the total number of function points in the project by the baseline workload per function point. The baseline workload per function point is set at 0.012 person-months per function point. This value is determined based on benchmark data from the Chinese software industry and measured data from 126 completed cost estimation projects of the enterprise. The calibration sample covers four types of projects: software testing, cost estimation, consulting and design, and performance evaluation. This is a software complexity factor, calculated and normalized by weighting the number of project functional modules and the number of external interfaces. The weight for the number of modules is 0.6, and the weight for the number of interfaces is 0.4. The number of modules is linearly reduced by an increment of 0.1 for every 10 modules, and the number of interfaces is linearly reduced by an increment of 0.1 for every 5 interfaces. The value ranges from 0 to 1. The development difficulty factor is calculated and normalized by weighting the technology stack maturity level and the proportion of customization requirements. The maturity level has a weight of 0.5, and the customization ratio has a weight of 0.5. The maturity level is divided into three levels, corresponding to base values ​​of 0.2, 0.5, and 0.8 respectively. The customization ratio is directly converted according to the actual proportion, and the value range is 0-1. The service type adjustment factor is 0.92 for software testing projects, 0.85 for software cost estimation projects, 1.05 for consulting and design projects, and 0.98 for performance evaluation projects. , The calibration coefficients are set to 0.12 and 0.08 respectively. The average relative error between the calibrated calculation results and the actual final cost is 7.6%.

[0011] Furthermore, the multi-dimensional compliance verification module performs a three-level progressive verification. The qualification compliance verification compares each qualification name, level, and validity period required by the tender documents. If all three match, it is marked as compliant; if any one is missing, it is marked as non-compliant. Simultaneously, it retrieves alternative qualifications of the same level and scope of application from the knowledge graph and pushes replacement materials. The performance matching degree verification sets a passing threshold of 0.6, which is determined based on the lowest performance matching degree statistics of historical winning projects. It filters the top 5 performances with the highest matching degree and verifies the consistency between the owner unit attributes, project completion time, and service content. If there is any case of confidentiality issues without valid proof or completion time exceeding 6 years, it is marked as invalid and automatically supplemented from the candidate performance library by matching degree. The technical standard compliance verification compares the standard number with the current version number, marks expired or abolished standards, and pushes the latest version entries. All verification results generate a problem list with chapter position markings. For compliant items, the solution engine automatically triggers the corresponding content replacement, preserving the original paragraph structure and expression style without manual intervention.

[0012] Furthermore, the intelligent cost calculation module incorporates a multi-model adaptation and cross-validation mechanism. The module simultaneously employs three calculation models: function point method, analogy method, and inference method. The system automatically determines the primary calculation model based on project characteristics. When the number of function points in a project is ≥500, the function point method is used as the primary model; when the number of similar historical projects is ≥12, the analogy method is used as the primary model; and for maintenance service projects, the inference method is used as the primary model. During the calculation process, the other two models are executed simultaneously as verification references. When the maximum relative deviation of the calculation results from the three models is ≤10%, the weighted average of the three results is taken as the final cost, with the primary model's result as the benchmark. The primary model has a weight of 0.6, and the other two models each have a weight of 0.2. When the deviation exceeds 10%, abnormal results are automatically marked and a manual review prompt is triggered. The cost results are divided into four categories: direct labor costs, direct non-human costs, indirect labor costs, and indirect non-human costs. A detailed cost breakdown is generated and embedded according to cost composition rules.

[0013] Furthermore, the domain knowledge graph module is configured with an entity linkage update unit. After a project is completed, the project's qualification usage records, performance data, technical standard citation records, service process execution data, and solution results are automatically entered into the graph. Corresponding entity nodes are added and automatically established with existing nodes according to association rules. At the same time, the matching weight of the corresponding entity is adjusted based on the bidding results. The matching weight of the entity corresponding to the winning bid is increased by 5% of the original value, and the matching weight of the entity corresponding to the unsuccessful bid is decreased by 2% of the original value. The cumulative adjustment of the weight of a single entity in a single natural month shall not exceed 20% of the original value. The weight value range is fixed at 0.2-1.5. When it exceeds the range, it is automatically truncated to the boundary value. The qualification entity has a built-in validity period reminder unit, which marks the qualification entity that is about to expire 30 days in advance. The technical standard entity is connected to the public data interface of the National Standard Document Sharing Service Platform, and the current status and version information of the standard are synchronized once a month, and the corresponding entity is updated or abolished in a timely manner.

[0014] Furthermore, the intelligent solution generation engine incorporates a four-category sub-domain chapter template library, covering four project scenarios: software testing, software cost estimation, consulting and design, and performance evaluation. Each template has a preset fixed chapter order and corresponding material filling positions. The filling positions are marked according to material type and priority. When generating a solution, the corresponding template is first matched according to the project type. Then, the corresponding material with the highest matching degree is retrieved from the knowledge graph according to the priority of qualifications, service processes, technical standards, performance, and typical cases, and filled into the corresponding position of the template. The transition sentences between chapters are implemented using a domain sentence template library. The template library contains 287 sets of categorized transition sentences, stored according to the chapter connection scenario. When called, the corresponding sentences of the preceding and following chapter themes are matched to complete the connection. After generation, terminology consistency verification is performed. Based on the synonym entity mapping relationship of the knowledge graph, non-standard expressions are uniformly replaced with standard terms to ensure consistent terminology throughout the text.

[0015] Furthermore, the system is configured with an incremental learning optimization unit that receives manual modifications to the generated scheme. It automatically extracts the material category and expression method corresponding to the modified location, and stores them in the optimization sample library after semantic deduplication and validity verification. The validity verification rule is that the modified content involves the replacement of core materials or structural adjustment. Simple text punctuation modifications are not counted as valid samples. Every 50 valid modification samples are accumulated, the system automatically triggers the recalibration of the knowledge graph entity matching weights and the supplementation and update of the sentence template library. The weight calibration adopts the gradient descent method, with the matching ranking of the manually modified materials as the objective function, and adjusts the matching weight parameters of the corresponding entities. The learning step size is set to 0.01. After calibration, a weight update report is generated. It officially takes effect after manual review and confirmation. Before it takes effect, the system still runs with the original weight parameters to ensure the stability of the generated scheme quality.

[0016] Compared with existing technologies, this intelligent generation system for technical solutions in information technology project bidding has the following advantages: I. This invention constructs a domain knowledge graph containing five types of entities, including qualifications, performance, and technical standards. Combined with a performance matching algorithm that incorporates confidentiality and scenario attribute correction, it achieves accurate recall and automatic template filling of bidding materials. This can compress the preparation cycle of a single proposal to the hour level, while reducing the impact of differences in the experience of the preparation personnel on the quality of the proposal, and ensuring the professionalism and consistency of the proposal outputs from multiple business lines.

[0017] Second, this invention is equipped with a multi-model-adaptive cost calculation mechanism and a three-level progressive compliance verification link, which enables accurate calculation of project workload and cost, as well as full-dimensional automatic compliance verification of qualifications, performance and technical standards. It can control the average relative error of cost calculation within 8%, effectively avoiding the risk of bid rejection due to issues such as non-compliance with qualifications or expired standards.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 This is an overall flowchart of a technical solution intelligent generation system for information technology project bidding. Figure 2 A flowchart of the element extraction module for a technical solution intelligent generation system for information technology project bidding. Figure 3 This is a flowchart of a three-level progressive compliance verification process for a technical solution intelligent generation system for information technology project bidding. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] Example 1

[0023] like Figure 1 As shown in this example, this embodiment uses the bidding scenario of a city's urban brain full-domain command and dispatch system project cost assessment service to illustrate the overall operation process of the system. The procuring party issues a formal tender document. The project adopts a 1+16+N architecture. The tender content is a third-party cost assessment service for the full-domain command and dispatch module, with a budget range of 1.1 million to 1.6 million yuan. Bidders are required to have the service capability level qualification of a software cost assessment agency, the solution must comply with the national standard for software cost measurement, the quotation calculation must be traceable, and they must provide performance data of similar smart city projects in the past 6 years. This embodiment uses an intelligent technical solution generation system for information technology project bidding to automatically generate the bidding technical solution for this project, to illustrate the specific operation process and implementation details of the system.

[0024] After the system starts, the project requirements analysis module first receives the uploaded tender document PDF. For example... Figure 2 As shown, the module first calls the PDF parsing interface to extract the full text content. It then filters the purchaser's identifier at the top of the document and the page numbers and disclaimers at the bottom using header and footer feature matching, while simultaneously removing watermark characters and scanned gibberish to obtain a clean tender text. Next, the module calls its built-in word segmentation lexicon for the information-based bidding and tendering field to perform word segmentation. This lexicon contains 11,700 structured terms, constructed from 487 historical project documents and 230 current industry standard documents after terminology extraction, deduplication, and manual verification. It covers four major categories of terms: qualification names, service types, technical standards, and deliverables. During word segmentation, common function words and meaningless auxiliary words are filtered out, retaining domain-specific vocabulary. After word segmentation, the module calls a domain entity recognition model fine-tuned based on the BERT-base architecture to locate the core elements in the tender document. This model uses 320 manually annotated tender documents as a fine-tuning dataset, with annotated entities covering six categories: project type, service scope, qualification requirements, delivery requirements, construction period, and budget range. It can accurately identify non-standardized tender requirements. After identifying the elements, the module performs semantic normalization processing, uniformly matching synonyms such as software cost assessment and cost accounting in the bidding documents into the standard term "software cost assessment" in the thesaurus. The final output is a structured element dataset containing three types of fields: element code, element value, and source page number. This clearly defines the project as a software cost assessment, the service scope as the cost assessment of the City Brain module, the qualification requirements as including the service capability level certification of software cost assessment institutions, the deliverable as a cost assessment report, the project duration as 45 days, and the budget as 1.1 million to 1.6 million.

[0025] Structured element data is synchronously transmitted to the intelligent solution generation engine and the intelligent cost calculation module, while the domain knowledge graph module initiates material retrieval. The domain knowledge graph module stores five types of entities: qualifications, performance, technical standards, service processes, and typical cases. Each type of entity is configured with standardized attribute fields. The qualification entity includes four attributes: qualification name, level, validity period, and scope of application. The performance entity includes five attributes: owner, project amount, completion time, service content, and project results. The technical standard entity includes four attributes: standard number, version number, release time, and applicable scenarios. The service process entity includes three attributes: process steps, corresponding deliverables, and applicable project types. The typical case entity includes three attributes: project background, service plan, and implementation effect. Three types of relationships are established between entities: attribute relationships are automatically mapped based on the same attribute values; scenario relationships are mapped based on the sub-field and industry category of the project; and time-series relationships are mapped based on the project completion time and standard release time. When a new entity is added, the three types of relationship with existing entities are automatically calculated. When the relationship exceeds 0.7, a corresponding relationship is automatically established. This threshold is set in conjunction with industry material matching accuracy statistics to ensure that the relationship accuracy remains above 90%. The relationship calculation uses a weighted sum of attribute overlap and semantic similarity, with attribute overlap having a weight of 0.6 and semantic similarity having a weight of 0.4. In this scenario, the module first matches qualification entities such as the service capability level certificate of software cost assessment agencies, and then calls the built-in performance matching degree calculation algorithm to traverse all historical performance entities to calculate the comprehensive matching degree.

[0026] Taking a past performance example, this project involved cost assessment services for the Yantai Big Data Bureau's City Brain project. The project amount was 1.28 million yuan, completed in 2021, and the services covered cost assessments for six major modules. The project was non-confidential. Substituting these values ​​into the formula, the overall matching degree and industry suitability are calculated. The cosine similarity was calculated using pre-trained industry semantic word vectors. Both belong to the third-level category of smart cities. Value: 0.94; Scale overlap Divide the length of the intersection between the budget range and the performance amount by the total length of the budget range. 1.28 million yuan falls within the range of 1.1 million to 1.6 million yuan; the length of the intersection corresponds to the full amount. Value 1; Service content coverage Calculate the ratio of the number of overlapping service items between the two to the total number of service items in the current project. This project has a total of 4 core service items, with 3 overlapping items. The value is 0.75; time decay coefficient. Based on the year difference, 2026 is 5 years apart from 2021, and the value increases by 0.05 for each year of difference. Value: 0.25; Item attribute correction coefficient Since the project is not classified, the value is 1; weighting coefficient 0.35 0.3 0.35 The weight of this group is 0.1. It was determined using a logistic regression model trained on 487 historical winning bids. The training used the winning bid result as the dependent variable and four matching factors as independent variables. The model's goodness of fit was 0.78. Substituting these values ​​into the calculation yields the overall matching degree. The performance score meets the matching criteria. The module iterates through all performance scores according to this rule, filters out the top-ranked performance scores that meet the criteria, and simultaneously recalls the corresponding technical standard entities, service process entities, and typical case entities to complete the material preparation.

[0027] Simultaneously, the intelligent cost estimation module initiated its calculation process. This project has a total of 1260 function points. The module first uses the function point method to calculate the baseline workload, with a baseline workload of 0.012 person-months per function point. This value is determined based on benchmark data from the Chinese software industry and measured data from 126 completed cost estimation projects of the company. The calibration sample covers four types of IT service projects, from which the baseline workload was calculated. Man-Month. Then, the workload adjustment algorithm is called to calculate the adjusted total workload and software complexity factor. The number of functional modules and external interfaces is weighted and calculated. The number of modules is linearly reduced by an increment of 0.1 for every 10 modules. This project has 18 functional modules, resulting in a module component of 0.18 and a weight of 0.6. The number of interfaces is linearly reduced by an increment of 0.1 for every 5 interfaces. This project has 12 external interfaces, resulting in an interface component of 0.24 and a weight of 0.4. The weighted calculation yields... Development difficulty factor The calculation is based on a weighted average of technology stack maturity and the proportion of customization requirements. The technology stack adopts a mature and general architecture, corresponding to maturity level 1, with a base value of 0.2 and a weight of 0.5. Customization requirements account for approximately 30%, which is directly converted to 0.3 with a weight of 0.5. Service type correction factor The corresponding value for software cost items is 0.85. This correction factor is determined based on statistical analysis of the standardization levels and per capita output differences across the four business categories. (Correction Factor) The value is 0.12. The value is 0.08. Substituting this into the formula yields the adjusted total workload. Person and Month.

[0028] After the calculations were completed, the module initiated multi-model cross-validation. Since this project has more than 500 function points, the function point method was used as the main model, simultaneously invoking the analogy and extrapolation methods. The analogy method, matching data from 12 similar projects of the same scale, yielded 13.52 person-months. The extrapolation method, based on staffing and service cycles, yielded 12.89 person-months. The maximum relative deviation among the three results was approximately 4.8%, less than the 10% threshold, which was set based on industry cost error tolerance and historical project settlement deviation statistics. Subsequently, a weighted average was calculated with the main model having a weight of 0.6 and the other models each having a weight of 0.2, to obtain the final workload result. This was then broken down into four categories: direct labor costs, direct non-human costs, indirect labor costs, and indirect non-human costs, generating a detailed cost breakdown section which was then transmitted to the intelligent solution generation engine.

[0029] After receiving the element data, materials, and cost information, the intelligent solution generation engine first matches the software cost estimation project chapter template. This template has pre-set fixed chapters such as project overview, evaluation basis, evaluation method, evaluation process, personnel configuration, qualifications and performance, cost details, and service advantages. Each chapter has corresponding material filling positions, and the material types are marked with priority. The engine fills the corresponding positions with materials of qualifications, service processes, technical standards, performance, and typical cases in order of priority. The transition sentences between chapters call the domain sentence template library. The template library contains 287 sets of categorized transition sentences extracted from historical high-quality solutions, stored according to the chapter connection scenario. In this case, the matching of corresponding sentences such as project overview to evaluation basis and evaluation method to evaluation process completes the connection. After generating the initial draft, the engine performs terminology consistency verification. Based on the synonym entity mapping relationship of the knowledge graph, it uniformly replaces non-standard expressions such as cost calculation and expense accounting in the text with standard cost estimation terms, resulting in the initial technical solution.

[0030] like Figure 3 As shown, the initial solution was transmitted to the multi-dimensional compliance verification module, where a three-level progressive verification was performed. The qualification compliance verification compared each item with the qualification name, level, and validity period required by the tender documents. The software cost assessment agency qualification required in this case matched all three attributes and was marked as compliant. The performance matching degree verification used 0.6 as the passing threshold. This threshold was determined based on historical winning project statistics; performance with a matching degree below 0.6 had a winning rate of less than 15%. The top 5 performances with the highest matching degree that met the standard were selected, and the owner's attributes, completion time, and service content were verified. All selected performances were within their 6-year validity period, with no cases of invalid confidentiality or desensitization, and were marked as valid. The technical standard compliance verification compared the referenced standard numbers (such as GB / T36964) with the current version numbers, confirming that all were valid versions with no expiration. The verification generated a list of issues with chapter location markers. No non-compliant items were found in this case. The module triggered the solution engine to confirm the content. If non-compliant items were found, the module would simultaneously push corresponding replacement materials, and the engine would automatically replace the content at the corresponding location, preserving the original paragraph structure and expression style.

[0031] After verification, the final solution is transmitted to the output module. The module matches the format template for government procurement projects, adjusts the font size, paragraph spacing, headers and footers, automatically generates a directory structure with jump function, and simultaneously generates editable DOCX format and unmodifiable PDF format documents. The content and format of the two types of documents are completely consistent and can be directly used for bid submission.

[0032] In summary, this embodiment uses the bidding process for the cost assessment of a city brain project as a specific scenario. It fully executes the entire process of requirements analysis, material retrieval, cost calculation, scheme generation, compliance verification, and output. The system can complete the preparation of a complete technical scheme within 1 hour, which is much faster than the 3 to 5 days preparation cycle of manual preparation. The average relative error between the cost calculation result and the actual project completion cost can be controlled within 8%. The compliance verification covers all core risk points, effectively reducing the quality fluctuations and bid rejection risks of manual preparation. This verifies the feasibility and practicality of the system in actual bidding scenarios.

[0033] Example 2

[0034] In a bidding scenario for a third-party cost assessment service for a city's urban brain full-domain command and dispatch system, the procuring entity's tender documents specified a project budget range of 1.1 million to 1.6 million yuan. The bidding entity was required to hold a software cost assessment agency qualification, provide similar smart city performance within the past 6 years, and the bidding proposal needed to include a complete and traceable cost calculation process. In this embodiment, this system was used to complete the fully automated compilation of the bidding technical proposal, and the accuracy of the algorithm calculation data was verified simultaneously.

[0035] Operators upload PDF tender documents to the system. The system then initiates the project requirements analysis module to perform document preprocessing. The module automatically identifies and filters PDF headers, footers, and page watermarks, removing garbled characters to obtain clean text. It then calls upon a built-in domain-specific word segmentation lexicon of 11,700 entries for filtering. This lexicon is constructed based on 487 government and enterprise tender documents and 230 IT industry standards that have been manually verified. After removing function words, it is input into the BERT-base entity recognition model for fine-tuning. This model, trained on 320 labeled tender documents, can identify six core elements: project type, qualifications, construction period, and budget. Semantic normalization is performed on synonymous expressions such as cost accounting and cost estimation in the tender documents, uniformly calibrating them to the standard term "software cost assessment." The system outputs a structured dataset with encoding, element content, and original page numbers. The dataset is labeled with the project type as "software cost assessment," the budget range as 1.1 million to 1.6 million yuan, the mandatory qualification as a "software cost assessment agency service capability level certificate," and the deliverable as a cost assessment report.

[0036] Structured elements are simultaneously pushed to the intelligent solution generation engine and the intelligent cost calculation module, triggering a retrieval of domain knowledge graph materials. The knowledge graph contains five standardized entities: qualifications, performance, technical standards, service processes, and typical cases. Each entity has fixed attribute fields. Entities are categorized into three types of relationships: attribute, scenario, and time sequence. Adding a new entity automatically calculates the relevance score, with a relevance threshold of 0.7 set based on the company's historical material matching accuracy. At this threshold, the material relevance accuracy consistently exceeds 90%. The relevance score is calculated using a weighted average of attribute overlap (0.6) and semantic similarity (0.4). The system calls the performance matching formula to iterate through all historical performance entities, selecting the Yantai City Brain cost project for calculation: Industry Adaptability. Based on the cosine of the word vectors from the third-level industry categories, the value is 0.94, indicating a high degree of overlap in scale. Since the performance amount of 1.28 million yuan falls within the budget range, the value is taken as 1, and the service content coverage is... Taking the intersection ratio of entries as 0.75, the attenuation coefficient corresponding to a time difference of 5 years is... =0.25, correction factor for non-classified projects Weight , , , The logistic regression was trained using 487 winning bids, with a goodness of fit of 0.78. The overall matching degree was calculated. ≈0.88, which is higher than the performance qualification threshold of 0.6. 0.6 is taken from the lowest matching statistical value of historical winning projects. If the winning rate of projects is lower than this threshold, the system will automatically select the top 5 performance records with the highest matching degree and store them in the material pool.

[0037] The intelligent cost estimation module performs quantitative calculations. The project has a total of 1260 functional points, with a single-function baseline workload of 0.012 person-months, derived from measured and calibrated data from 126 completed projects. =15.12 person-months; software complexity factor Based on 18 modules and 12 interfaces, the weighted average is 0.204, representing the development difficulty factor. Based on a mature technology stack, 30% customization requirements are converted to a software cost adjustment factor of 0.25. =0.85, calibration coefficient =0.12、 =0.08, substituting into the workload correction formula, we get ≈13.17 person-months. The module initiates cross-validation of three models, with the function point method as the main model. The analogy method and the inference method are used to calculate the results simultaneously, which are 13.52 person-months and 12.89 person-months, respectively. The maximum relative deviation among the three is 4.8%, which is lower than the 10% deviation threshold. The final workload is obtained by weighting the main model with a weight of 0.6 and the others with a weight of 0.2, and the cost is split into four categories to generate a detailed cost list.

[0038] The intelligent solution generation engine matches the software's cost estimation-specific chapter templates. These templates include fixed fill spaces for project overview, evaluation basis, cost calculation, and similar performance records. It retrieves graphic materials based on priority according to qualifications, standards, performance, and case studies. It uses 287 sets of scenario-based transition sentences to complete paragraph connections. After completion, synonyms are uniformly validated to eliminate terminological ambiguity, and an initial technical solution is output. The initial solution is then sent to a multi-dimensional compliance verification module for three-level verification. All qualifications, performance records, and cited standards are found to be compliant, with no expired or mismatched content, and no need for automatic correction. Upon completion of verification, the deliverables module automatically generates standardized DOCX and PDF dual-format bidding documents.

[0039] Data validation section A comparative test was conducted on a city brain project of similar scale, comparing manual and system-based compilation. Manual compilation took 3.5 days, with a relative error of 16.3% in cost calculation and an average matching degree of only 0.62 for manually selected projects. The system, on the other hand, took 52 minutes for the entire process, with an average relative error of 7.6% in cost calculation and an average matching degree of 0.86 for automatically selected projects. Compliance verification showed no omissions of risk points. Quantitative data proves that the system has a significant improvement in compilation efficiency, calculation accuracy, and material matching quality.

[0040] In summary, this embodiment fully operates the six core modules of the system, fully calculates the performance matching and workload using two innovative algorithms, verifies the implementation effect of multi-model cross-validation and graph entity association mechanisms, and completes quantitative verification through manual comparison with data. It can efficiently output compliant and accurately calculated cost-related bidding schemes, and is fully adapted to the needs of smart city cost assessment bidding business.

[0041] Example 3

[0042] In a bidding scenario for third-party software testing services for a map data warehouse information system of a provincial natural resources department, the bidding documents require testing covering all dimensions of functionality, security, performance, and usability. Bidding units must have dual CNAS and CMA testing qualifications and at least three similar geographic information system testing performance records in the past 6 years. This embodiment relies on the system to automatically generate software testing bidding technical solutions, focusing on verifying the recognition accuracy of the requirements analysis module and the full-process control capabilities of multi-dimensional compliance verification.

[0043] Operators upload Word-format bidding documents to the system. The project requirements analysis module initiates multi-format text parsing, automatically filtering headers, unit watermarks, and garbled page numbers. It utilizes a 11,700-word segmentation lexicon covering software testing, surveying and mapping informatization, and testing qualification-specific terminology to complete word segmentation and stop word removal. The segmentation results are fed into a BERT-base fine-tuning entity recognition model, trained on 320 government and enterprise testing-related bidding documents, accurately extracting five core elements: project type, qualification requirements, testing scope, schedule, and delivery report. Semantic normalization is performed on synonyms such as "security testing" and "information security testing" in the documents, uniformly calibrating them to the standard term "information security testing." A structured element dataset is output, labeled with the project type as software testing, mandatory qualifications including CNAS laboratory accreditation and CMA inspection and testing qualifications, testing scope covering seven major categories of map system tests, and the delivery result as a complete software testing report.

[0044] Element data is synchronously distributed to the intelligent solution generation engine and the intelligent cost calculation module, while the domain knowledge graph simultaneously initiates material retrieval. The graph stores five types of entities, including all enterprise qualifications such as CNAS and CMA, past surveying and mapping test performance, national software testing standards, standardized testing processes, and typical map system test cases. Entity attributes are fully differentiated by qualification validity period, performance service content, and current standard version. Three types of relationships are automatically calculated for new entities, with a correlation degree threshold of 0.7 to ensure the accuracy of material association. The system traverses all test performance categories, calls the performance matching formula to calculate the matching score, and filters out valid geographic information system test performance with a matching degree higher than 0.6 to store in the material library.

[0045] The intelligent solution generation engine matches software testing-specific chapter templates. These templates pre-set fixed positions for test scope, test procedures, personnel configuration, qualification certificates, and similar cases, filling in content according to the priority of map materials. It calls upon 287 sets of transition sentence templates to complete chapter connections, performs full-text terminology uniformity verification, and outputs an initial test technical solution containing all seven major categories of test content. The initial solution is then sent to a multi-dimensional compliance verification module for three-level progressive verification. Qualification verification checks each item against the CNAS and CMA qualification names, levels, and validity periods; both qualifications are marked as compliant as they are all within their validity period. Performance verification selects the top 5 map system test performances with the highest matching degree; all completion times are within 6 years, and there are no confidential or anonymized projects. Standard verification checks against the GB / T software testing series standard versions; there are no expired or obsolete standards, and no non-compliant items are found, requiring no automatic replacement or correction.

[0046] The intelligent cost calculation module simultaneously calculates the workload of testing services. Based on the project's functional points and test case scale, it substitutes them into the workload correction formula and simultaneously initiates cross-validation of the three models: functional points, analogy, and inference. The calculation deviation is less than 10%. It then breaks down the costs into four categories: manpower, equipment, and indirect costs to generate a test service quotation section embedding scheme. After all content is validated, it is pushed to the output module, automatically adapting to the bidding format specifications of the natural resources system, and generating an editable document with a jump table of contents and an encrypted PDF bid document.

[0047] Data validation section Three similar geographic information system tender documents were selected for a comparative test of the accuracy of requirements analysis. The average omission rate of manually extracted core elements was 18.6%, and terminology standardization took about 40 minutes. The omission rate of entity recognition elements in this system was only 2.1%, and semantic normalization was completed automatically, with a single document parsing time of no more than 2 minutes. The compliance verification process showed an average of 2 to 3 omissions in qualifications and standards during manual self-checks, while the system achieved zero compliance omissions in its three-level verification, fully verifying the stability and reliability of the requirements analysis and multi-level compliance verification modules.

[0048] In summary, this embodiment fully replicates the entire process of generating software testing project solutions, and fully implements all technical mechanisms, including project requirement analysis, word segmentation and recognition, graph material retrieval, three-level compliance verification, and multi-model cost calculation. It is verified through real bidding scenarios in natural resource surveying and mapping informatization, and can comprehensively cover the requirements for developing testing solutions in all dimensions such as functionality and security, significantly reducing the risk of bid rejection caused by omissions of human factors and compliance oversights.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A technical solution intelligent generation system for information technology project bidding, characterized in that, The system includes the following components: Project Requirements Analysis Module: The output end is connected to the input end of the intelligent solution generation engine and the intelligent cost calculation module respectively, and is used to extract and output the core elements of the project; Domain Knowledge Graph Module: This module is connected to the intelligent solution generation engine, the multi-dimensional compliance verification module, and the intelligent cost calculation module, respectively, and is used to store five types of entities and their associated mapping relationships: qualifications, performance, technical standards, service processes, and typical cases. Intelligent solution generation engine: The output end connects to the input end of the multi-dimensional compliance verification module, recalls physical materials, and generates an initial technical solution according to the chapter template; Multi-dimensional compliance verification module: Corrects the intelligent generation engine of the output feedback connection solution, which is used to perform three types of compliance verifications: qualification, performance, and technical standards, and trigger automatic correction; The intelligent cost calculation module outputs a solution intelligent generation engine to match the calculation model and output project workload and cost details. Output module: Connects to the final output of the intelligent solution generation engine, used to complete document formatting and output according to bidding specifications.

2. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The project requirements analysis module has a built-in word segmentation lexicon for the field of information-based bidding and tendering. After receiving the requirements document, the module first performs multi-format text extraction, extracts the full text, and filters headers, footers, watermarks, and garbled characters. Then, it performs word segmentation and stop word filtering based on the word segmentation lexicon. Subsequently, it uses a domain entity recognition model based on the BERT-base architecture for fine-tuning to locate core elements. The model uses 320 manually annotated bidding documents as a fine-tuning dataset. After recognition, it performs semantic normalization on the elements, matches standard terms in the lexicon to complete the expression calibration, and finally outputs a structured element dataset containing three types of fields: element code, element value, and source page number.

3. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The domain knowledge graph module has five types of entities, each with standardized attribute fields. The qualification entity includes four attributes: qualification name, level, validity period, and scope of application. The performance entity includes five attributes: owner, project amount, completion time, service content, and project results. The technical standard entity includes four attributes: standard number, version number, release time, and applicable scenarios. The service process entity includes three attributes: process steps, corresponding deliverables, and applicable project types. The typical case entity includes three attributes: project background, service plan, and implementation effect. Attribute associations between entities are automatically mapped based on shared attribute values. Scenario associations are mapped based on the project's sub-domain and industry category. Time-series associations are mapped based on the project completion time and standard release time. When a new entity is added, the three types of association with existing entities are automatically calculated. If the association score exceeds 0.7, a corresponding association is automatically established. The association score calculation uses a weighted sum of attribute overlap and semantic similarity, with an attribute overlap weight of 0.6 and a semantic similarity weight of 0.

4.

4. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The domain knowledge graph module has a built-in performance matching degree calculation algorithm. It calculates the overall matching degree between a single historical performance record and the current project using the following formula: ,in For overall matching accuracy, For industry compatibility, For the degree of overlap in scale, To improve service content coverage, The time decay coefficient, This is a correction factor for project attributes. , , , These are the weighting coefficients.

5. The intelligent generation system for technical solutions of information technology project bidding as described in claim 1, characterized in that, The intelligent cost calculation module has a built-in workload correction algorithm. For software projects, after calculating the baseline workload using the function point method, the adjusted total workload is calculated using the following formula: ,in To adjust the total workload of the project, As the baseline function point workload, For software complexity factor, To develop difficulty factors, For service type adjustment coefficient, , This is the calibration coefficient.

6. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The multi-dimensional compliance verification module performs a three-level progressive verification. For qualification compliance verification, each qualification is compared against the three attributes required in the tender documents: name, level, and validity period. If all three match, the qualification is marked as compliant; if any one is missing, it is marked as non-compliant. Simultaneously, alternative qualifications of the same level and scope of application are retrieved from the knowledge graph and replacement materials are pushed. For performance matching verification, a pass / fail threshold of 0.6 is set. This threshold is determined based on the lowest performance matching score of historical winning projects. The top 5 performances with the highest matching scores that meet the standard are selected. The owner unit attributes, project completion time, and service content of the performances are verified. If there is any case of confidentiality issues without valid proof or completion time exceeding 6 years, the performance is marked as invalid and automatically supplemented from the candidate performance library by matching score. For technical standard compliance verification, the standard number is compared with the current version number. Expired or abolished standards are marked and the latest version is pushed. All verification results generate a problem list with chapter location markings. For compliant items, the solution engine automatically triggers the corresponding content replacement.

7. The intelligent generation system for technical solutions of information technology project bidding as described in claim 1, characterized in that, The intelligent cost calculation module incorporates a multi-model adaptation and cross-validation mechanism. The module simultaneously supports three calculation models: function point method, analogy method, and inference method. The system automatically determines the primary calculation model based on project characteristics. When the number of function points in a project is ≥500, the function point method is used as the primary model; when the number of similar historical projects is ≥12, the analogy method is used as the primary model; and for operation and maintenance service projects, the inference method is used as the primary model. During the calculation process, the other two models are executed simultaneously as verification references. When the maximum relative deviation of the calculation results from the three models is ≤10%, the weighted average of the three results is taken as the final cost, with the primary model's result as the benchmark. The primary model has a weight of 0.6, and the other two models each have a weight of 0.

2. When the deviation exceeds 10%, abnormal results are automatically marked and a manual review prompt is triggered. The cost results are broken down into four categories: direct labor costs, direct non-human costs, indirect labor costs, and indirect non-human costs. A detailed cost breakdown is generated and embedded according to cost composition rules.

8. The intelligent generation system for technical solutions of information technology project bidding as described in claim 1, characterized in that, The domain knowledge graph module is configured with an entity linkage update unit. After a project is completed, the project's qualification usage records, performance data, technical standard reference records, service process execution data, and solution deliverables are automatically entered into the graph. Corresponding entity nodes are added and automatically established with existing nodes according to association rules. At the same time, the matching weight of the corresponding entity is adjusted based on the bidding results. The matching weight of the entity corresponding to the winning bid is increased by 5% of the original value, and the matching weight of the entity corresponding to the unsuccessful bid is decreased by 2% of the original value. The cumulative adjustment of the weight of a single entity in a single natural month shall not exceed 20% of the original value. The weight value range is fixed at 0.2-1.

5. When it exceeds the range, it is automatically truncated to the boundary value. The qualification entity has a built-in validity period reminder unit, which marks the qualification entity that is about to expire 30 days in advance.

9. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The intelligent generation engine for the proposed solution has a built-in template library of four sub-domains, covering four project scenarios: software testing, software cost estimation, consulting and design, and performance evaluation. Each template has a fixed chapter order and corresponding material filling positions. The filling positions are marked according to material type and priority. When generating a solution, the corresponding template is first matched according to the project type. Then, the corresponding material with the highest matching degree is retrieved from the knowledge graph according to the priority of qualifications, service processes, technical standards, performance, and typical cases, and filled into the corresponding position of the template. The transition sentences between chapters are implemented using a domain sentence template library. The template library contains 287 sets of categorized transition sentences, which are stored according to the chapter connection scenario. When called, the corresponding sentences of the themes of the preceding and following chapters are matched to complete the connection. After generation, terminology consistency verification is performed. Based on the synonym entity mapping relationship of the knowledge graph, non-standard expressions are uniformly replaced with standard terms.

10. The intelligent generation system for technical solutions in information technology project bidding as described in claim 1, characterized in that, The system is configured with an incremental learning optimization unit that receives manual modifications to the generated scheme. It automatically extracts the material category and expression method corresponding to the modified location, and stores them in the optimization sample library after semantic deduplication and validity verification. The validity verification rule is that the modified content involves the replacement of core materials or structural adjustment. Simple text punctuation modifications are not counted as valid samples. Every 50 valid modification samples are accumulated, the system automatically triggers the recalibration of the knowledge graph entity matching weights and the supplementation and update of the sentence template library. The weight calibration adopts the gradient descent method, with the matching ranking of the manually modified materials as the objective function, and adjusts the matching weight parameters of the corresponding entities. The learning step size is set to 0.

01. After calibration, a weight update report is generated. It officially takes effect after manual review and confirmation. Before it takes effect, the system still runs with the original weight parameters.

Images