Method and device for generating unified bid inviting file, electronic equipment and storage medium

Abstract

The invention relates to a method and device for generating a unified bid inviting file, electronic equipment and a storage medium. The method for generating the unified bid invitation file comprises the following steps: S1, designing a bid invitation file structured processing model, and setting the bid invitation file according to a model structure of modules and attributes; s2, designing an intelligent learning model of the bid invitation file, extracting feature vectors from the preprocessed data by utilizing a machine learning model, establishing a rule base conforming to business logic, integrating rules and policies of bid invitation parties, ensuring the compliance of the bid invitation file, and following corresponding laws and regulations and standards; and S3, designing a bid invitation file intelligent generation model, and generating a bid invitation file according to a purchase demand, a technical specification book and the like. According to the method for generating the unified bid invitation file, different types of bid invitation file model libraries can be established according to different types of projects; common problems and terms in the historical bid invitation file are analyzed through artificial intelligence, intelligent prompting and early warning are performed in the new bid invitation file, and benefits of enterprises are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, electronic device and storage medium for generating unified bidding documents. Background Art

[0002] The preparation of bidding documents is a tedious and time-consuming task, involving a lot of data collation, formatting and compliance checks. The traditional way of preparing bidding documents is not only inefficient but also prone to errors. Especially when project information is complex and bidding requirements are strict, the need to generate bidding documents through artificial intelligence technology is becoming increasingly prominent.

[0003] The goal is to enable computer systems to continuously improve their performance and predictive capabilities by learning and adapting to data. The core of deep learning is to build models, train models through large amounts of data and algorithms, enable them to discover patterns and laws from data, and use these laws to make decisions and predictions. Unlike traditional programming, deep learning is a data-based autonomous learning method that does not require clear rules to be defined in advance, but determines the next step by learning the data itself. Figure 7 The figure shows the fusion mode of deep learning technology and Transformer model in AIGC (Artificial Intelligence Generated Content). This learning method enables deep learning to adapt to various environments and scenarios and find the best solution. At present, the fields where deep learning models are widely used mainly include: computer vision, natural language processing, audio processing, time series analysis, shopping and entertainment, automation and robotics, financial technology, etc.

[0004] Therefore, there is an urgent need for a solution that can use artificial intelligence technology to quickly learn and analyze a large number of historical bidding documents, extract key information such as the format, terms, requirements, etc. of the bidding documents, and automatically apply them to new bidding documents, so as to reduce human errors and improve work efficiency. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the traditional way of compiling bidding documents is not only inefficient but also prone to errors.

[0006] In order to solve the above technical problems, according to one aspect of the present invention, a method for generating a unified bidding document is provided, comprising the following steps: S1, designing a structured processing model for bidding documents, setting up the bidding documents according to the model structure of modules and attributes; S2, designing an intelligent learning model for bidding documents, using a machine learning model to extract feature vectors from preprocessed data, establishing a rule base that conforms to business logic, integrating the rules and policies of the tendering party, ensuring the compliance of the bidding documents, and complying with relevant laws and regulations and standards; S3, designing an intelligent generation model for bidding documents, intelligently generating bidding documents based on procurement requirements, technical specifications, etc.

[0007] According to an embodiment of the present invention, step S1 may include the following steps: S11, by analyzing and disassembling the bidding documents, designing a "module + attribute" model structure, extracting information from the document so that all documents in the bidding process can be quickly constructed into a structured text, the information includes hierarchical structure, text content, style information; S12, information classification and annotation method, classifying and annotating the extracted key information, and classifying it into corresponding fields or attributes for subsequent data processing and analysis; S13, file content parsing model, parsing the content of the bidding documents, extracting key information including supplier information, product specifications, technical requirements, and contract terms; S14, setting risks Point detection mode, to detect risk points of key information in the bidding documents, so as to facilitate early warning and processing when risks such as potential contract disputes, technical risks, and intellectual property requirements are found; S15, file structured storage method, to store the parsed and extracted key information in a database or data warehouse to form a structured data mart for subsequent data analysis and mining; S16, through the structured processing model of the above content, a structured processing model for bidding documents is established, and bidding documents are automatically processed and analyzed, key information is extracted, and operations including classification, labeling, deduplication, integration, and risk point detection are performed, so as to improve the processing efficiency and accuracy of bidding documents and provide strong support for the bidding work of enterprises.

[0008] According to an embodiment of the present invention, step S2 may include the following steps: S21, based on the enterprise's bidding and procurement history, deep learning technology is applied to build a bidding strategy library, and a preliminary judgment is made on the rationality of the bidding. The preliminary judgment on the rationality of the bidding includes the bidding price limit judgment. The bidding price limit judgment reminds the unreasonable price limit and generates a price limit reference based on the dimensions of the same type, homogeneity, same stage, same batch, and same time period of the subject matter; S22, historical data learning process design, extracting and refining key elements through natural language processing technology, and forming a bidding template knowledge strategy library according to the application field, background, use, attributes and other characteristics, and pre-processing these files, including text cleaning, word segmentation, and part-of-speech tagging to remove irrelevant characters and formatted text; the historical bidding documents after data pre-processing are parsed and sliced, and stored in a unified vector database; S23, rule engine model design, using NLP (Natural Language The AI ​​​​is used to analyze the text using natural language processing (Natural Language Processing) technology to extract key features including the type, scale, technical requirements, and bidding conditions of the bidding project. The key features actively help AI understand the structure and content of the bidding documents and select a suitable machine learning model, where the machine learning model is a deep learning model; extract feature vectors from the preprocessed data, establish a rule library that conforms to the business logic, integrate the rules and policies of the tenderer, ensure the compliance of the bidding documents, and follow the relevant regulations and standards; S24, algorithm optimization design, includes two aspects. On the one hand, through machine learning algorithms, the system can continuously learn and optimize the content of the bidding documents to improve its quality and adaptability; on the other hand, combined with the results of manual revisions to the bidding documents, the model is continuously iterated and optimized based on feedback and data from actual use.

[0009] According to an embodiment of the present invention, step S3 may include the following steps: S31, template generation, using a pre-selected tender document template, according to tender requirements and standards, automatically filling and quickly generating a unified format and standardized content, automatically extracting key information from technical materials such as tender requirements, supplier information, product specifications, etc., to adapt to different tender scenarios; S32, forming a standardized procurement strategy library, unifying according to qualification performance requirements, bidding section division, and evaluation rules; S33, structuring the demand document, performing structured entry and maintenance of technical specifications and scoring rules; S34, structuring the tender document, the system provides templates and cases, realizes one-click generation of tender documents, and supports manual compilation; modular design is used to deconstruct laws, management specifications and historical tender documents into reusable modules, automatically or manually selects applicable modules according to input conditions, and reconstructs and generates standard tender documents; S35, intelligent assisted review, performing key point comparison, including on-screen comparison of review elements, corresponding chapters of technical specifications, and modification records; performing online editing, modifying the draft file online, and automatically planning the version number when saving; performing AI intelligent retrieval, including natural language retrieval of historical similar projects, tender information, and internal specification requirements.

[0010] According to an embodiment of the present invention, in step S14, risk point detection may use data mining technology to analyze the content of structured bidding documents and identify possible problems, including data cleaning, data transformation, data mining implementation process, pattern evaluation and knowledge representation steps, with the goal of discovering patterns and relationships hidden in the data, including statistical techniques, association rules, decision trees, neural networks, to identify possible problems, and provide intelligent review and suggestion functions, using algorithms including the following: classification, finding the common characteristics of a group of data objects in the database and dividing them into different classes according to the classification pattern, used to map the data items in the database to A given category; applied to customer classification, customer attribute and feature analysis, customer satisfaction analysis, customer purchase trend prediction, used for data classification in historical bidding document learning, including procurement-related requirements for its procurement type, procurement scale, and procurement method; regression analysis, realizing the temporal characteristics of attribute values ​​in the transaction database, generating a function that maps data items to a real-valued prediction variable, discovering the dependency between variables or attributes, including the trend characteristics of data sequences, the prediction of data sequences, and the correlation between data; used for feature value judgment and prediction of corresponding bidding documents, corresponding to the analysis and reasoning of bidding documents in the same type of procurement business; Clustering is to divide a set of data into several categories according to similarities and differences, so that the similarity between data belonging to the same category is as large as possible, and the similarity between data in different categories is as small as possible; it is applied to the classification of potential customer groups, potential customer background analysis, and potential market segmentation in bidding documents; association rules are rules used to describe the relationship between data items in the database. According to the appearance of certain items in a transaction, it can be derived that other items also appear in the same transaction, that is, the association or relationship hidden in the data; in customer relationship management, by mining the large amount of data in the company's customer database, it is used to find interesting relationships from a large number of records. association, find out the key factors that affect the business operation effect; extract features, combine feature analysis with vector database, extract feature formulas about these data from a set of data in the vector database, and these feature formulas express the overall characteristics of the data set; including procurement personnel extracting features from historical bidding activities to obtain a series of reasons and main characteristics for bidding failures. These features can effectively prevent bidding failures caused by unreasonable preparation of bidding documents; change and deviation analysis is used to find meaningful differences between observed results and reference quantities; pay attention to unexpected rules, and mine unexpected rules for application to aspects including discovery, analysis, identification, evaluation and early warning of abnormal information.

[0011] According to an embodiment of the present invention, text structuring technology may be used in step S15 to convert unstructured text data into structured data so that it can be understood and processed by a computer system; by processing the text data including word segmentation, part-of-speech tagging, syntactic analysis, and semantic analysis, the text data is converted into a computer-readable format, thereby facilitating the computer to process and mine the text data; for structured processing of bidding documents to improve the accuracy and efficiency of natural language processing, help enterprises better process and analyze large amounts of text data, and improve the efficiency of information acquisition and utilization of enterprises; adopt the following processing: text preprocessing, cleaning and preprocessing of documents, which includes unified character encoding, eliminating redundant and repeated content, removing special characters and HTML tags, handling spelling errors, performing word segmentation, identifying and removing stop words, segmenting, sentence segmentation, and converting text to lowercase, to ensure that subsequent extraction work is provided with clean, structured and Consistent data foundation; entity recognition, not limited to Transformer-based models such as BERT, but also includes its various refined domain-specific versions and variants suitable for low-resource languages; the model is trained to recognize various entities in documents, including but not limited to names of people, organizations, places, dates, and other domain-specific terms such as products, equipment names, and equipment parameters; combining knowledge graphs and external dictionaries or databases to enhance the model's contextual understanding capabilities for more accurate entity recognition; relation extraction, which combines knowledge graphs, external relational databases, and context-enhanced methods to ensure accurate capture of multiple connections between entities in complex texts; information normalization, which is used to unify diverse representations into a standard format to ensure data consistency and comparability; taking dates as an example, whether it is "1Jan2022" or "01 / 01 / 2022", they are converted into a unified standard. But this is just the tip of the iceberg. Normalization also involves variants of place names, conversion of currency units, processing of synonyms and near-synonyms, etc.In addition, deep learning and transfer learning strategies are also introduced to help the model automatically learn and respond to emerging new representation variants, ensuring continuous and efficient normalization of information; context understanding and long text processing, by using models including Longformer and BigBird to process long text sequences that exceed the limitations of traditional models, to ensure that each part of the document is evaluated in the appropriate context, which not only improves the accuracy of information extraction, but also captures subtle associations and hints in the document; error handling and fault tolerance mechanisms, when there are errors, ambiguities or various noises in the document, these factors may lead to deviations and inaccuracies in information extraction; in order to improve robustness and reliability, error handling and fault tolerance mechanisms have become an indispensable part; involving post-processing and rule-based error correction, an integrated method including Bagging and Boosting is used to merge the predictions of multiple models, and semi-supervised learning is used to learn from partially labeled data. Error handling and fault tolerance mechanisms are used to build a powerful, adaptable and reliable system that can continuously provide high-quality information extraction in complex and noisy data environments.

[0012] According to an embodiment of the present invention, the rule engine technology may be used in step S23, which is developed from the reasoning engine and is a component embedded in the application program, used to separate business decisions from the application program code and use predefined semantic modules to write business decisions; the rule engine accepts data input, interprets business rules, and makes business decisions based on business rules; it is used to formulate and manage rules, and automatically review bidding documents; the rule engine can automatically identify possible problems and provide suggestions based on business rules and standards, and the rule engine includes the following processing: fact processing is a way of representing facts in the rule engine, using the data structure of the fact table; the fact table is a relational data structure, used to store attribute value pairs of facts; fact processing includes operations of inserting, updating, deleting and querying facts; rule processing is a way of representing rules in the rule engine, using the rule table (rule table); the rule table is a relational data structure used to store the conditions and results of the rules; rule processing includes the operations of inserting, updating, deleting and querying rules; decision processing is the core function of the rule engine, which is used to realize automated logical processing based on rules and facts; decision processing includes the operations of generating, evaluating and executing decisions; among which, decision generation is the process of building a decision tree based on facts and rules; decision evaluation is the process of evaluating the feasibility and advantages of a decision based on a decision tree; decision execution is the process of realizing automated decision-making based on a decision tree.

[0013] According to a second aspect of the present invention, there is provided a device for generating a unified bidding document using the above-mentioned method for generating a unified bidding document, comprising:

[0014] The bidding document structured processing model is used to set the bidding document according to the model structure of modules and attributes;

[0015] Intelligent learning model for bidding documents, which is used to extract feature vectors from pre-processed data by using machine learning models, establish a rule base that conforms to business logic, integrate the rules and policies of the tenderer, ensure the compliance of bidding documents, and comply with relevant laws, regulations and standards;

[0016] The bidding document intelligent generation model is used to intelligently generate bidding documents based on procurement requirements, technical specifications, etc.

[0017] According to a third aspect of the present invention, there is provided an electronic device comprising: a memory, a processor and a program for generating a unified bidding document which is stored in the memory and can be run on the processor, wherein the program for generating a unified bidding document implements the steps of the above-mentioned method for generating a unified bidding document when executed by the processor.

[0018] According to a fourth aspect of the present invention, a computer storage medium is provided, wherein a program for generating a unified bidding document is stored on the computer storage medium, and when the program for generating a unified bidding document is executed by a processor, the steps of the above-mentioned method for generating a unified bidding document are implemented.

[0019] Compared with the prior art, the technical solution provided by the embodiments of the present invention can at least achieve the following beneficial effects:

[0020] According to the method for generating unified bidding documents of the present invention, when preparing bidding documents, different types of bidding document model libraries can be established according to different types of projects, including engineering investment, material procurement, and service projects; common problems and clauses in historical bidding documents can be analyzed through artificial intelligence, and intelligent prompts and warnings can be provided in new bidding documents; it helps the writing personnel to better grasp the key elements of the bidding documents and avoid potential risks and problems; through data mining and analysis, it automatically identifies key information in the bidding documents, such as the name of the bidding project, the tenderer, the bidding agency, the bid deadline, etc., and automatically reviews and verifies this information; and it helps enterprises check the standardization, rigor and comprehensiveness of the bidding documents, so as to avoid mistakes in the bidding process and protect the interests of the enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present invention, but are not intended to limit the present invention.

[0022] Figure 1 is a flow chart showing generation of a unified bidding document according to an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram showing a module structure according to an embodiment of the present invention;

[0024] Figure 3 is a diagram showing a structured processing model of a bidding document according to an embodiment of the present invention;

[0025] Figure 4 It is a diagram showing an intelligent learning model for bidding documents according to an embodiment of the present invention;

[0026] Figure 5 is an overall flow chart showing generation of bidding documents according to an embodiment of the present invention;

[0027] Figure 6 is a flow chart illustrating generation of a bidding document according to an embodiment of the present invention;

[0028] Figure 7 is a diagram showing an example of generating a bidding document according to an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the fusion mode of deep learning technology and the Transformer model in AIGC. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Unless otherwise defined, the technical or scientific terms used herein shall have the common meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the patent application specification and claims of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but indicate the existence of at least one.

[0032] Figure 1 is a flow chart showing generation of a unified bidding document according to an embodiment of the present invention.

[0033] like Figure 1 As shown, the method for generating a unified bidding document includes the following steps:

[0034] S1. Design of the structural processing model of the bidding documents, setting the bidding documents according to the model structure of modules and attributes;

[0035] S2. Design of intelligent learning model for bidding documents. Using machine learning model, extract feature vectors from pre-processed data, establish a rule base that conforms to business logic, integrate the rules and policies of the bidding party, ensure the compliance of bidding documents, and follow relevant laws and regulations and standards;

[0036] S3. Design of intelligent bidding document generation model to intelligently generate bidding documents based on procurement requirements, technical specifications, etc.

[0037] Figure 2 is a schematic diagram showing a module structure according to an embodiment of the present invention; Figure 3 is a diagram showing a structured processing model of a bidding document according to an embodiment of the present invention; Figure 4 The figure shows an intelligent learning model of bidding documents according to an embodiment of the present invention.

[0038] like Figures 2 to 4 As shown, step S1 includes the following steps:

[0039] S11. By analyzing and disassembling the bidding documents, a "module + attribute" model structure is designed to extract information from the documents so that all documents in the bidding process can be quickly constructed into a structured text, including hierarchical structure, text content, and style information;

[0040] S12, information classification and annotation method, classify and annotate the extracted key information and classify it into corresponding fields or attributes to facilitate subsequent data processing and analysis;

[0041] S13, document content parsing model, parses the content of the bidding document and extracts key information including supplier information, product specifications, technical requirements, and contract terms;

[0042] S14. Set up a risk point detection mode to detect risk points of key information in the bidding documents, so as to facilitate early warning and handling when risks including potential contract disputes, technical risks, and intellectual property requirements are found;

[0043] S15, file structured storage method, storing the parsed and extracted key information in a database or data warehouse to form a structured data mart for subsequent data analysis and mining;

[0044] S16. Through the structured processing model of the above content, a structured processing model for bidding documents is established to automatically process and analyze the bidding documents, extract key information and perform operations including classification, labeling, deduplication, integration, and risk point detection, so as to improve the processing efficiency and accuracy of bidding documents and provide strong support for the bidding work of enterprises.

[0045] According to an embodiment of the present invention, step S2 includes the following steps:

[0046] S21. Based on the enterprise's bidding and procurement history, deep learning technology is applied to build a bidding strategy library to make a preliminary judgment on the rationality of the bidding, which includes the bidding price limit judgment. The bidding price limit judgment reminds unreasonable price limits and generates price limit references based on dimensions including the same type, homogeneity, same stage, same batch, and same time period of the subject matter;

[0047] S22. Design of historical data learning process, extracting and refining key elements through natural language processing technology, forming a knowledge strategy library of bidding templates according to application fields, backgrounds, uses, attributes and other characteristics, and preprocessing these documents, including text cleaning, word segmentation, and part-of-speech tagging to remove irrelevant characters and formatted text; parsing and slicing the historical bidding documents after data preprocessing, and storing them in a unified vector database;

[0048] S23. Design of rule engine model, using NLP (Natural Language Processing) technology to analyze text and extract key features including type, scale, technical requirements and bidding conditions of bidding projects, which actively help AI understand the structure and content of bidding documents and select appropriate machine learning models, where the machine learning model is a deep learning model; extract feature vectors from pre-processed data, establish a rule base that conforms to business logic, integrate the rules and policies of the tenderer, ensure the compliance of bidding documents, and follow relevant laws and regulations and standards;

[0049] S24. Algorithm optimization design includes two aspects. On the one hand, through machine learning algorithms, the system can continuously learn and optimize the content of the bidding documents to improve their quality and adaptability; on the other hand, combined with the results of manual revisions to the bidding documents, the model is continuously iterated and optimized based on feedback and data from actual use.

[0050] According to one or some embodiments of the present invention, step S3 includes the following steps:

[0051] S31, template generation, using pre-selected tender document templates, automatically filling and quickly generating uniform formats and standardized content according to tender requirements and standards, automatically extracting key information from technical materials such as tender requirements, supplier information, product specifications, etc., to adapt to different tender scenarios;

[0052] S32. Form a standardized procurement strategy library, unify the qualification and performance requirements, bidding divisions, and review rules;

[0053] S33, Structuring of requirement documents, structured entry and maintenance of technical specifications and scoring rules;

[0054] S34. The bidding documents are structured. The system provides templates and cases to realize one-click generation of bidding documents and supports manual compilation. It adopts modular design to deconstruct laws, management specifications and historical bidding documents into reusable modules, automatically or manually selects applicable modules according to input conditions, and reconstructs and generates standard bidding documents.

[0055] S35, intelligent assisted review, comparison of key points, including on-screen comparison of review elements, corresponding chapters of the technical specification, and modification records; online editing, online modification of draft files, automatic planning of version numbers when saving; AI intelligent retrieval, including natural language retrieval of historical similar projects, bidding information, and internal specification requirements.

[0056] According to one or some embodiments of the present invention, in step S14, risk point detection uses data mining technology to analyze the content of structured bidding documents and identify possible problems, including data cleaning, data transformation, data mining implementation process, pattern evaluation and knowledge representation steps, with the goal of discovering patterns and relationships hidden in the data, including statistical techniques, association rules, decision trees, neural networks, to identify possible problems and provide intelligent review and suggestion functions, using algorithms including the following:

[0057] Classification, finding out the common features of a group of data objects in the database and dividing them into different classes according to the classification model, used to map the data items in the database to a given category through the classification model; applied to customer classification, customer attribute and feature analysis, customer satisfaction analysis, customer purchase trend prediction, used for data classification in the study of historical bidding documents, including procurement-related requirements for its procurement type, procurement scale, and procurement method;

[0058] Regression analysis realizes the temporal characteristics of attribute values ​​in the transaction database, generates a function that maps data items to a real-valued prediction variable, and discovers the dependencies between variables or attributes, including the trend characteristics of data series, the prediction of data series, and the correlation between data; it is used to judge and predict the characteristic values ​​of corresponding bidding documents, and corresponds to the analysis and reasoning of bidding documents in the same type of procurement business;

[0059] Clustering is to divide a set of data into several categories according to similarities and differences, so that the similarity between data belonging to the same category is as large as possible, and the similarity between data in different categories is as small as possible; it is applied to the classification of potential customer groups, potential customer background analysis, and potential market segmentation in bidding documents;

[0060] Association rules are rules used to describe the relationship between data items in a database. The appearance of certain items in a transaction can lead to the appearance of other items in the same transaction, that is, the association or relationship hidden between data. In customer relationship management, by mining a large amount of data in the company's customer database, it is used to discover interesting associations from a large number of records and find out the key factors that affect the effectiveness of business operations.

[0061] Extract features. Combine feature analysis with vector database to extract feature expressions about a set of data in the vector database. These feature expressions express the overall characteristics of the data set. This includes extracting features from historical bidding activities by procurement personnel to obtain a series of reasons and main features for bidding failure. These features can effectively prevent bidding failure caused by unreasonable preparation of bidding documents.

[0062] Change and deviation analysis is used to find meaningful differences between observations and reference quantities; focus on unexpected rules and mine unexpected rules for application to aspects including discovery, analysis, identification, evaluation and early warning of abnormal information.

[0063] According to one or some embodiments of the present invention, text structuring technology is used in step S15 to convert unstructured text data into structured data so that it can be understood and processed by a computer system; the text data is converted into a computer-readable format by processing including word segmentation, part-of-speech tagging, syntactic analysis, and semantic analysis, thereby facilitating the computer to process and mine the text data; the structured processing of the bidding documents is used to improve the accuracy and efficiency of natural language processing, help enterprises better process and analyze large amounts of text data, and improve the efficiency of information acquisition and utilization of enterprises; the following processing is used:

[0064] Text preprocessing: cleaning and preprocessing documents, including unified character encoding, eliminating redundant and repeated content, removing special characters and HTML tags, dealing with spelling errors, performing word segmentation, identifying and removing stop words, segmenting, sentence segmentation, and converting text to lowercase, to ensure a clean, structured, and consistent data foundation for subsequent extraction work;

[0065] Entity recognition, not limited to Transformer-based models such as BERT, but also includes its various refined domain-specific versions and variants suitable for low-resource languages; the model is trained to recognize various entities in documents, including but not limited to names of people, organizations, places, dates, and other domain-specific terms such as products, device names, and device parameters; combining knowledge graphs and external dictionaries or databases to enhance the model's contextual understanding capabilities for more accurate entity recognition;

[0066] Relation extraction, which combines knowledge graphs, external relational databases, and context-enhanced methods to ensure that multiple connections between entities are accurately captured in complex texts;

[0067] Information normalization is used to unify diverse representations into a standard format to ensure data consistency and comparability; taking dates as an example, whether it is "1Jan 2022" or "01 / 01 / 2022", they are all converted into a unified standard. But this is just the tip of the iceberg. Normalization also involves variants of place names, conversion of currency units, processing of synonyms and near-synonyms, etc. In addition, deep learning and transfer learning strategies are also introduced to help the model automatically learn and respond to emerging new representation variants, ensuring continuous and efficient normalization of information.

[0068] Contextual understanding and long text processing, using models including Longformer and BigBird to process long text sequences that exceed the limits of traditional models, ensuring that every part of the document is evaluated in the appropriate context, which not only improves the accuracy of information extraction, but also captures subtle connections and hints in the document;

[0069] Error handling and fault tolerance mechanisms. When there are errors, ambiguities or various noises in the document, these factors may lead to deviations and inaccuracies in information extraction. In order to improve robustness and reliability, error handling and fault tolerance mechanisms have become an indispensable part. It involves post-processing and rule-based error correction. An integrated method including Bagging and Boosting is used to merge the predictions of multiple models. Semi-supervised learning is used to learn from partially labeled data. Error handling and fault tolerance mechanisms are used to build a powerful, adaptable and reliable system that can continuously provide high-quality information extraction in complex and noisy data environments.

[0070] According to one or some embodiments of the present invention, the rule engine technology used in step S23 is developed from the reasoning engine, which is a component embedded in the application program, used to separate business decisions from the application program code, and use predefined semantic modules to write business decisions; the rule engine accepts data input, interprets business rules, and makes business decisions based on business rules; it is used to formulate and manage rules, and automatically review bidding documents; the rule engine can automatically identify possible problems and provide suggestions based on business rules and standards, and the rule engine includes the following processing:

[0071] Fact processing is the way to represent facts in the rule engine, using the data structure of the fact table. The fact table is a relational data structure used to store attribute-value pairs of facts. Fact processing includes operations such as inserting, updating, deleting, and querying facts.

[0072] Rule processing is the way to represent rules in the rule engine, using the data structure of the rule table. The rule table is a relational data structure used to store the conditions and results of the rules. Rule processing includes the operations of inserting, updating, deleting and querying rules.

[0073] Decision processing is the core function of the rule engine, which is used to realize automated logical processing based on rules and facts. Decision processing includes the generation, evaluation and execution of decisions. Among them, decision generation is the process of building a decision tree based on facts and rules; decision evaluation is the process of evaluating the feasibility and advantages of decisions based on decision trees; decision execution is the process of realizing automated decision-making based on decision trees.

[0074] According to a second aspect of the present invention, there is provided a device for generating a unified bidding document using the above-mentioned method for generating a unified bidding document, comprising:

[0075] The bidding document structured processing model is used to set the bidding document according to the model structure of modules and attributes;

[0076] Intelligent learning model for bidding documents, which is used to extract feature vectors from pre-processed data by using machine learning models, establish a rule base that conforms to business logic, integrate the rules and policies of the tenderer, ensure the compliance of bidding documents, and comply with relevant laws, regulations and standards;

[0077] The bidding document intelligent generation model is used to intelligently generate bidding documents based on procurement requirements, technical specifications, etc.

[0078] Figure 5 is an overall flow chart showing generation of bidding documents according to an embodiment of the present invention; Figure 6 is a flow chart illustrating generation of a bidding document according to an embodiment of the present invention; Figure 7 is a diagram showing an example of generating a bidding document according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the fusion mode of deep learning technology and the Transformer model in AIGC.

[0079] According to the method for generating unified bidding documents of the present invention, when preparing bidding documents, different types of bidding document model libraries can be established according to different types of projects, including engineering investment, material procurement, and service projects; common problems and clauses in historical bidding documents can be analyzed through artificial intelligence, and intelligent prompts and warnings can be provided in new bidding documents; it helps the writing personnel to better grasp the key elements of the bidding documents and avoid potential risks and problems; through data mining and analysis, it automatically identifies key information in the bidding documents, such as the name of the bidding project, the tenderer, the bidding agency, the bid deadline, etc., and automatically reviews and verifies this information; and it helps enterprises check the standardization, rigor and comprehensiveness of the bidding documents, so as to avoid mistakes in the bidding process and protect the interests of the enterprises.

[0080] According to another aspect of the present invention, there is provided a device for generating a unified bidding document, comprising: a memory, a processor, and a program for generating a unified bidding document stored in the memory and executable on the processor, wherein the program for generating a unified bidding document implements the steps of the above-mentioned method for generating a unified bidding document when executed by the processor.

[0081] According to the present invention, a computer storage medium is also provided.

[0082] The computer storage medium stores a program for generating unified bidding documents, and when the program for generating unified bidding documents is executed by the processor, the steps of the method for generating unified bidding documents are implemented.

[0083] Among them, the method implemented when the unified bidding document generating program running on the processor is executed can refer to the various embodiments of the unified bidding document generating method of the present invention, and will not be repeated here.

[0084] The present invention also provides a computer program product.

[0085] The computer program product of the present invention comprises a program for generating a unified bidding document, and when the program for generating a unified bidding document is executed by a processor, the steps of the method for generating a unified bidding document as described above are implemented.

[0086] Among them, the method implemented when the unified bidding document generating program running on the processor is executed can refer to the various embodiments of the unified bidding document generating method of the present invention, and will not be repeated here.

[0087] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0088] The above description is merely an exemplary embodiment of the present invention and is not intended to limit the protection scope of the present invention. The protection scope of the present invention is determined by the appended claims.

Claims

1. A method for generating a unified bidding document, comprising the following steps: S1. Design of the structural processing model of the bidding documents, setting the bidding documents according to the model structure of modules and attributes; S2. Design of intelligent learning model for bidding documents. Using machine learning model, extract feature vectors from pre-processed data, establish a rule base that conforms to business logic, integrate the rules and policies of the bidding party, ensure the compliance of bidding documents, and follow relevant laws and regulations and standards; S3. Design of intelligent bidding document generation model to intelligently generate bidding documents based on procurement requirements, technical specifications, etc.

2. The method for generating a unified bidding document according to claim 1, wherein: Step S1 includes the following steps: S11. By analyzing and disassembling the bidding documents, a "module + attribute" model structure is designed to extract information from the documents so that all documents in the bidding process can be quickly constructed into a structured text, including hierarchical structure, text content, and style information; S12, information classification and annotation method, classify and annotate the extracted key information and classify it into corresponding fields or attributes to facilitate subsequent data processing and analysis; S13, document content parsing model, parses the content of the bidding document and extracts key information including supplier information, product specifications, technical requirements, and contract terms; S14. Set up a risk point detection mode to detect risk points of key information in the bidding documents, so as to facilitate early warning and handling when risks including potential contract disputes, technical risks, and intellectual property requirements are found; S15, file structured storage method, storing the parsed and extracted key information in a database or data warehouse to form a structured data mart for subsequent data analysis and mining; S16. Through the structured processing model of the above content, a structured processing model for bidding documents is established to automatically process and analyze the bidding documents, extract key information and perform operations including classification, labeling, deduplication, integration, and risk point detection to improve the processing efficiency and accuracy of bidding documents.

3. The method for generating a unified bidding document according to claim 1, wherein: Step S2 includes the following steps: S21. Based on the enterprise's bidding and procurement history, deep learning technology is applied to build a bidding strategy library to make a preliminary judgment on the rationality of the bidding, which includes the bidding price limit judgment. The bidding price limit judgment reminds unreasonable price limits and generates price limit references based on dimensions including the same type, homogeneity, same stage, same batch, and same time period of the subject matter; S22. Design of historical data learning process, extracting and refining key elements through natural language processing technology, forming a bidding template knowledge strategy library according to application fields, backgrounds, usage, attributes and other characteristics, and preprocessing these documents, including text cleaning, word segmentation, and part-of-speech tagging to remove irrelevant characters and formatted text; parsing and slicing the historical bidding documents after data preprocessing, and storing them in a unified vector database; S23. Design of rule engine model, using NLP technology to analyze text and extract key features including type, scale, technical requirements and bidding conditions of bidding projects, which actively help AI understand the structure and content of bidding documents and select appropriate machine learning models, where the machine learning model is a deep learning model; extract feature vectors from pre-processed data, establish a rule base that conforms to business logic, integrate the rules and policies of the tenderer, ensure the compliance of bidding documents, and follow relevant laws and regulations and standards; S24. Algorithm optimization design includes two aspects. On the one hand, through machine learning algorithms, the system can continuously learn and optimize the content of the bidding documents to improve their quality and adaptability; on the other hand, combined with the results of manual revisions to the bidding documents, the model is continuously iterated and optimized based on feedback and data from actual use.

4. The method for generating a unified bidding document according to claim 1, wherein: Step S3 includes the following steps: S31, template generation, using pre-selected tender document templates, automatically filling and quickly generating uniform formats and standardized content according to tender requirements and standards, automatically extracting key information from technical materials such as tender requirements, supplier information, product specifications, etc., to adapt to different tender scenarios; S32. Form a standardized procurement strategy library, unify the qualification and performance requirements, bidding divisions, and review rules; S33, Structuring of requirement documents, structured entry and maintenance of technical specifications and scoring rules; S34. The bidding documents are structured. The system provides templates and cases to enable one-click generation of bidding documents and supports manual compilation. Adopt modular design to deconstruct laws, management regulations and historical bidding documents into reusable modules, automatically or manually select applicable modules according to input conditions, and reconstruct and generate standard bidding documents; S35, intelligent assisted review, comparison of key points, including on-screen comparison of review elements, corresponding chapters of the technical specification, and modification records; online editing, online modification of draft files, automatic planning of version numbers when saving; AI intelligent retrieval, including natural language retrieval of historical similar projects, bidding information, and internal specification requirements.

5. The method for generating unified bidding documents as claimed in claim 2, wherein: In step S14, risk point detection uses data mining technology to analyze the content of structured bidding documents and identify possible problems, including data cleaning, data transformation, data mining implementation process, pattern evaluation and knowledge representation steps. The goal is to discover patterns and relationships hidden in the data, including statistical techniques, association rules, decision trees, neural networks, to identify possible problems and provide intelligent review and suggestion functions, using algorithms including the following: Classification, finding out the common features of a group of data objects in the database and dividing them into different classes according to the classification model, used to map the data items in the database to a given category through the classification model; applied to customer classification, customer attribute and feature analysis, customer satisfaction analysis, customer purchase trend prediction, used for data classification in the study of historical bidding documents, including procurement-related requirements for its procurement type, procurement scale, and procurement method; Regression analysis realizes the temporal characteristics of attribute values ​​in transaction databases, generates a function that maps data items to a real-valued prediction variable, and discovers the dependencies between variables or attributes, including the trend characteristics of data series, the prediction of data series, and the correlation between data; It is used to judge and predict the characteristic values ​​of the corresponding bidding documents, and is used to analyze and reason the bidding documents in the same type of procurement business; Clustering is to divide a set of data into several categories according to similarities and differences, so that the similarity between data belonging to the same category is as large as possible, and the similarity between data in different categories is as small as possible; it is applied to the classification of potential customer groups, potential customer background analysis, and potential market segmentation in bidding documents; Association rules are rules used to describe the relationship between data items in a database. The appearance of certain items in a transaction can lead to the appearance of other items in the same transaction, that is, the association or relationship hidden between data. In customer relationship management, by mining a large amount of data in the company's customer database, it is used to discover interesting associations from a large number of records and find out the key factors that affect the effectiveness of business operations. Extract features, feature analysis is combined with the vector database to extract feature formulas about a set of data in the vector database. These feature formulas express the overall characteristics of the data set. This includes the procurement staff extracting features from historical bidding activities to obtain a series of reasons and main features for bidding failures. These features can be used to effectively prevent bidding failures caused by unreasonable preparation of bidding documents; Change and deviation analysis is used to find meaningful differences between observations and reference quantities; focus on unexpected rules and mine unexpected rules for application to aspects including discovery, analysis, identification, evaluation and early warning of abnormal information.

6. The method for generating unified bidding documents as claimed in claim 2, wherein: In step S15, text structuring technology is used to convert unstructured text data into structured data so that it can be understood and processed by the computer system; the text data is converted into a computer-readable format by processing including word segmentation, part-of-speech tagging, syntactic analysis, and semantic analysis, so as to facilitate the computer to process and mine the text data; it is used for structured processing of bidding documents to improve the accuracy and efficiency of natural language processing, help enterprises better process and analyze large amounts of text data, and improve the efficiency of information acquisition and utilization of enterprises; the following processing is used: Text preprocessing: cleaning and preprocessing documents, including unified character encoding, eliminating redundant and repeated content, removing special characters and HTML tags, dealing with spelling errors, performing word segmentation, identifying and removing stop words, segmenting, sentence segmentation, and converting text to lowercase, to ensure a clean, structured, and consistent data foundation for subsequent extraction work; Entity recognition, not limited to Transformer-based models such as BERT, but also includes its various refined domain-specific versions and variants suitable for low-resource languages; the model is trained to recognize various entities in documents, including but not limited to names of people, organizations, places, dates, and other domain-specific terms such as products, device names, and device parameters; combining knowledge graphs and external dictionaries or databases to enhance the model's contextual understanding capabilities for more accurate entity recognition; Relation extraction, which combines knowledge graphs, external relational databases, and context-enhanced methods to ensure that multiple connections between entities are accurately captured in complex texts; Information normalization, which is used to unify the diverse representations into a standard format, thus ensuring the consistency and comparability of the data; Contextual understanding and long text processing, using models including Longformer and BigBird to process long text sequences that exceed the limits of traditional models, ensuring that every part of the document is evaluated in the appropriate context, which not only improves the accuracy of information extraction, but also captures subtle connections and hints in the document; Error handling and fault tolerance mechanisms. When there are errors, ambiguities or various noises in the document, these factors may lead to deviations and inaccuracies in information extraction. In order to improve robustness and reliability, error handling and fault tolerance mechanisms have become an indispensable part. It involves post-processing and rule-based error correction. An integrated method including Bagging and Boosting is used to merge the predictions of multiple models. Semi-supervised learning is used to learn from partially labeled data. Error handling and fault tolerance mechanisms are used to build a powerful, adaptable and reliable system that can continuously provide high-quality information extraction in complex and noisy data environments.

7. The method for generating unified bidding documents as claimed in claim 3, wherein: In step S23, rule engine technology is used to separate business decisions from application code and write business decisions using predefined semantic modules; the rule engine accepts data input, interprets business rules, and makes business decisions based on business rules; it is used to formulate and manage rules, and automatically review bidding documents; the rule engine can automatically identify possible problems and provide suggestions based on business rules and standards, and the rule engine includes the following processing: Fact processing is the way to represent facts in the rule engine, using the data structure of the fact table; the fact table is a relational data structure used to store attribute-value pairs of facts; fact processing includes operations such as inserting, updating, deleting, and querying facts; Rule processing is the way to represent rules in the rule engine, using the data structure of the rule table; the rule table is a relational data structure used to store the conditions and results of the rules; rule processing includes the operations of inserting, updating, deleting and querying rules; Decision processing is the core function of the rule engine, which is used to realize automated logical processing based on rules and facts. Decision processing includes the generation, evaluation and execution of decisions. Among them, decision generation is the process of building a decision tree based on facts and rules; decision evaluation is the process of evaluating the feasibility and advantages of decisions based on decision trees; decision execution is the process of realizing automated decision-making based on decision trees.

8. A device for generating a unified bidding document using the method for generating a unified bidding document according to any one of claims 1 to 7, comprising: The bidding document structured processing model is used to set the bidding document according to the model structure of modules and attributes; Intelligent learning model for bidding documents, which is used to extract feature vectors from pre-processed data by using machine learning models, establish a rule base that conforms to business logic, integrate the rules and policies of the tenderer, ensure the compliance of bidding documents, and comply with relevant laws, regulations and standards; The bidding document intelligent generation model is used to intelligently generate bidding documents based on procurement requirements, technical specifications, etc.

9. An electronic device, comprising: A memory, a processor, and a program for generating a unified bidding document stored in the memory and executable on the processor, wherein when the program for generating a unified bidding document is executed by the processor, the steps of the method for generating a unified bidding document as described in any one of claims 1 to 7 are implemented.

10. A computer storage medium, wherein: The computer storage medium stores a program for generating a unified bidding document, and when the program for generating a unified bidding document is executed by a processor, the steps of the method for generating a unified bidding document as described in any one of claims 1 to 7 are implemented.

Images