Bidding document generation method and system based on feature vector matching

Abstract

The invention provides a bidding document generation method and system based on feature vector matching. The method comprises the steps of obtaining a bidding document of a bidding project; using a preset document splitting technology to split the bid invitation document into a plurality of document topics; inputting the document topics into a preset text processing model, so that the text processing model generates feature vectors corresponding to the document topics respectively; according to the feature vectors, similarity matching is carried out in a preset document material database, text content corresponding to the feature vectors is determined, and the document material database is obtained after a plurality of historical bidding document files are dismantled based on the document dismantling technology and classified and stored; and generating a bidding document file according to each document topic and each text content, thereby realizing automatic generation of the bidding document and improving bidding document making efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of text content generation, and in particular to a method and system for generating a bid based on feature vector matching. Background Art

[0002] A bid document is a legally binding and enforceable standard document for bidders, providing substantive responses to various requirements from the tender issuer. The bid document can be divided into two parts: questions and answers. The questions are formulated by the tender issuer based on the objectives to be achieved, while the answers are the bidder's substantive responses to the questions based on their own circumstances. Bids from different companies in the same field (such as short video production and pension investment and wealth management) often show high correlation between the questions. Furthermore, the responses must be rigorous and concise, without any errors.

[0003] The traditional method of preparing bid documents relies mainly on manual work. The specific process is that after the bidder obtains the bidding documents, a bid preparation team is established. Based on the questions in the bidding documents, the team searches the company-level knowledge base to find the corresponding answers and responds. For questions that are not found in the search, the team will then discuss and answer them based on their own actual situation. Currently, the traditional method of preparing bid documents has the following three major problems:

[0004] 1. Low production efficiency: The process and operations of maintaining, searching, screening, reviewing, and using materials in the preparation of bid documents are cumbersome, resulting in low production efficiency and an inability to efficiently respond to tender document responses;

[0005] 2. Extensive duplication of work: For highly similar issues in the bidding documents of different companies in the same field, the bidding process often requires repeated searches and use of materials, wasting a lot of manpower and posing risks due to repeated manual operations;

[0006] 3. Low degree of automation: The process of preparing bid documents is mainly manual, which requires a lot of manpower and time and has a low degree of automation. Summary of the Invention

[0007] In response to the above technical problems, the present application provides a method and system for generating bid documents based on feature vector matching, which is used to realize the automatic generation of bid documents and improve the efficiency of bid document production.

[0008] In a first aspect, an embodiment of the present application provides a method for generating a bid based on feature vector matching, comprising:

[0009] Obtain bidding documents for bidding projects;

[0010] Using a preset document decomposition technology to decompose the bidding document into a number of document titles;

[0011] Inputting each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles;

[0012] Based on each of the feature vectors, similarity matching is performed in a preset document material database to determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology;

[0013] Generate a tender document based on the titles of the documents and the content of the text.

[0014] The embodiment of the present application provides a method for generating bid documents based on feature vector matching, which converts the bidding documents into several feature vectors through document decomposition technology and text processing models, and converts large-scale bidding documents into multiple feature vectors that are more conducive to computer understanding, which is conducive to subsequent vector similarity matching in the document material library, and improves the matching efficiency of similarity matching. Furthermore, in the embodiment of the present application, the document material database is obtained by decomposing and classifying and storing several historical bidding documents based on the document decomposition technology, and converting the bidding documents and document materials into vector dimensions for matching, which greatly reduces the computational complexity of document matching. Compared with the existing method of manually producing bid documents, the embodiment of the present application realizes the automatic generation of bid documents through document decomposition, conversion and vector similarity matching, thereby improving the efficiency of bid document production.

[0015] Furthermore, the text processing model generates feature vectors corresponding to the respective document titles, including:

[0016] Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles;

[0017] Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

[0018] An embodiment of the present application provides a method for converting a document title into a feature vector. The embodiment of the present application avoids directly converting the document title into a vector, but instead summarizes and abstracts the document title through a large language model. After generating a shorter title summary, the natural language model is used to convert the title summary into a feature vector to generate a corresponding feature vector, thereby improving the efficiency and accuracy of feature vector extraction.

[0019] In one possible implementation, performing similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors includes:

[0020] performing similarity matching in a vector database of the document material database based on each of the feature vectors, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors;

[0021] For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector;

[0022] According to the ID of each first material vector, ID matching is performed in the text database of the document material database, thereby determining the text content corresponding to each of the feature vectors.

[0023] The embodiment of the present application provides a method for determining the text content corresponding to each of the feature vectors, matching a number of material vectors for each feature vector in a vector database, and performing saturation matching, so that the most suitable material vector can be selected from multiple material vectors later. At the same time, during the matching process, the correlation score between each of the feature vectors and the corresponding material vectors is determined as the basis for automatically selecting the material vector, thereby improving the accuracy of matching between the feature vector and the material vector. Finally, based on the ID of each first material vector, vector matching is performed in another document material database, and the matching of the vector and the text content is achieved through the intermediate field ID, providing data support for the subsequent generation of the bid document, thereby improving the efficiency of the bid document production.

[0024] Furthermore, performing similarity matching in the vector database of the document material database based on each of the feature vectors, and then matching a plurality of material vectors for each of the feature vectors, includes:

[0025] For each of the feature vectors, the Euclidean distance between the feature vector and all feature vectors in the vector database is calculated, and a preset number of material vectors having the closest Euclidean distance to the feature vector are obtained by screening.

[0026] In one possible implementation, the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology, including:

[0027] Obtain several historical tender documents;

[0028] For each of the historical bid documents, using the document decomposition technology, the historical bid document is decomposed into a plurality of source document topics and answers corresponding to the respective source document topics;

[0029] Adding document tags to the titles of the respective source documents in a keyword matching manner, and establishing corresponding relationships between the titles of the respective source documents and the respective historical bid documents;

[0030] Constructing a text database based on the titles of the respective source documents, the corresponding document tags, and the corresponding answers to the respective titles;

[0031] Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles;

[0032] Constructing a vector database according to each of the material vectors;

[0033] The document material database is obtained by combining the text database and the vector database.

[0034] An embodiment of the present application provides a method for constructing a document material database, which also uses document disassembly technology to disassemble historical bidding documents into several material document titles and corresponding answers to each title, and generates material vectors corresponding to each of the material document titles through a text processing model. Finally, text data and vector data are used to construct different types of databases, namely text databases and vector databases, wherein the vector database is used for vector matching, and the text database is used to provide corresponding text content, thereby providing necessary text data for the subsequent generation of bidding documents and improving the efficiency of bidding document production.

[0035] Furthermore, after obtaining a plurality of historical bidding documents, each of the historical bidding documents is classified into authority levels according to a preset authority level target;

[0036] When adding a document tag to each of the material document titles in the manner of keyword matching, a corresponding authority level is added to each of the material document titles.

[0037] In an embodiment of the present application, historical bid documents are further classified into authority levels, so that the authority level of each text content can be referenced in the subsequent similarity matching process, further improving the accuracy of similarity matching.

[0038] In a second aspect, accordingly, an embodiment of the present application provides a tender document generation system based on feature vector matching, including an acquisition module, a document disassembly module, a feature vector generation module, a vector matching module, and a tender document generation module;

[0039] Wherein, the acquisition module is used to obtain the bidding documents of the bidding project;

[0040] The document disassembly module is used to disassemble the bidding document into a plurality of document titles using a preset document disassembly technology;

[0041] The feature vector generation module is used to input each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles;

[0042] The vector matching module is used to perform similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and classifying and storing a number of historical bidding documents based on the document disassembly technology;

[0043] The bid document generation module is used to generate a bid document according to the titles of the documents and the text contents.

[0044] Furthermore, the feature vector generation module generates feature vectors corresponding to the respective document titles, including:

[0045] Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles;

[0046] Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

[0047] In one possible implementation, the vector matching module performs similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors, including:

[0048] performing similarity matching in a vector database of the document material database based on each of the feature vectors, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors;

[0049] For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector;

[0050] According to the ID of each first material vector, ID matching is performed in the text database of the document material database, thereby determining the text content corresponding to each of the feature vectors.

[0051] In one possible implementation, the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology, including:

[0052] Obtain several historical tender documents;

[0053] For each of the historical bid documents, using the document decomposition technology, the historical bid document is decomposed into a plurality of source document topics and answers corresponding to the respective source document topics;

[0054] Adding document tags to the titles of the respective source documents in a keyword matching manner, and establishing corresponding relationships between the titles of the respective source documents and the respective historical bid documents;

[0055] Constructing a text database based on the titles of the respective source documents, the corresponding document tags, and the corresponding answers to the respective titles;

[0056] Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles;

[0057] Constructing a vector database according to each of the material vectors;

[0058] The document material database is obtained by combining the text database and the vector database. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 : A flow chart of a method for generating a bid based on feature vector matching provided in an embodiment of the present application.

[0060] Figure 2 : A detailed flow chart of a bid generation method based on feature vector matching provided in an embodiment of the present application.

[0061] Figure 3 : A structural diagram of a bid generation system based on feature vector matching provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0063] It should be noted that the step numbers herein are for convenience of explanation of the specific embodiments and do not serve to define the order in which the steps are to be performed. In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature designated "first" or "second" may explicitly or implicitly include one or more of such features.

[0064] Example 1:

[0065] like Figure 1 As shown, the first embodiment provides a method for generating a bid based on feature vector matching, including steps S1-S5:

[0066] Step S1: Obtain the bidding documents of the bidding project;

[0067] Step S2: using a preset document decomposition technology to decompose the bidding document into a plurality of document titles;

[0068] Step S3: inputting each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles;

[0069] Step S4: performing similarity matching on each of the feature vectors in a preset document material database to determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology;

[0070] Step S5: Generate a bid document based on the titles of the documents and the contents of the text.

[0071] In a preferred embodiment, in step S1, the bidding document is obtained by a user uploading it via a webpage. In step S2, the bidding document is decomposed into a plurality of document titles using a preset document decomposition technology, specifically by automatically identifying the smallest title in the document using a preset text recognition program, and processing the title into titles to obtain a plurality of document titles corresponding to the bidding document.

[0072] Furthermore, in step S3, the text processing model generates feature vectors corresponding to the respective document titles, including:

[0073] Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles;

[0074] Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

[0075] An embodiment of the present application provides a method for converting a document title into a feature vector. The embodiment of the present application avoids directly converting the document title into a vector, but instead summarizes and abstracts the document title through a large language model. After generating a shorter title summary, the natural language model is used to convert the title summary into a feature vector to generate a corresponding feature vector, thereby improving the efficiency and accuracy of feature vector extraction.

[0076] In a preferred embodiment, the large language model is OPENAI's LLM GPT-4, and the natural language model is Google's BERT model.

[0077] In one possible implementation, in step S4, performing similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors includes:

[0078] performing similarity matching in a vector database of the document material database based on each of the feature vectors, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors;

[0079] For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector;

[0080] According to the ID of each first material vector, ID matching is performed in the text database of the document material database, thereby determining the text content corresponding to each of the feature vectors.

[0081] The embodiment of the present application provides a method for determining the text content corresponding to each of the feature vectors, matching a number of material vectors for each feature vector in a vector database, and performing saturation matching, so that the most suitable material vector can be selected from multiple material vectors later. At the same time, during the matching process, the correlation score between each of the feature vectors and the corresponding material vectors is determined as the basis for automatically selecting the material vector, thereby improving the accuracy of matching between the feature vector and the material vector. Finally, based on the ID of each first material vector, vector matching is performed in another document material database, and the matching of the vector and the text content is achieved through the intermediate field ID, providing data support for the subsequent generation of the bid document, thereby improving the efficiency of the bid document production.

[0082] Furthermore, performing similarity matching in the vector database of the document material database based on each of the feature vectors, and then matching a plurality of material vectors for each of the feature vectors, includes:

[0083] For each of the feature vectors, the Euclidean distance between the feature vector and all feature vectors in the vector database is calculated, and a preset number of material vectors having the closest Euclidean distance to the feature vector are obtained by screening.

[0084] The bid generation method provided in the embodiment of the present application is essentially a process of understanding, converting, and matching the meaning of each topic in the new bid with the meaning of topics in the existing bid material library. Therefore, a natural language algorithm is required to achieve the closest match between two semantically similar topics. In a preferred embodiment, the specific implementation process of step S4 is as follows:

[0085] In tender document preparation, the matching questions need to be converted into an abstract understanding of the meaning of the questions using a large language model, and then a feature vector is generated through a natural language algorithm. Then, in the vector database, the feature vector of the new question can be easily searched for the closest feature vector, such as the Euclidean distance, with the existing feature vectors, and the final 20 closest material vectors can be obtained. At this time, the text ID of the material vector is used to return to the text database, and the corresponding closest question and answer can be found and automatically matched. In addition, in the automatic tender document generation system, users can also use the manual selection mode to generate 20 corresponding questions and answers for each question, and then manually select the most appropriate question answer from these 20 questions as the text content.

[0086] In one possible implementation, the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology, including:

[0087] Obtain several historical tender documents;

[0088] For each of the historical bid documents, using the document decomposition technology, the historical bid document is decomposed into a plurality of source document topics and answers corresponding to the respective source document topics;

[0089] Adding document tags to the titles of the respective source documents in a keyword matching manner, and establishing corresponding relationships between the titles of the respective source documents and the respective historical bid documents;

[0090] Constructing a text database based on the titles of the respective source documents, the corresponding document tags, and the corresponding answers to the respective titles;

[0091] Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles;

[0092] Constructing a vector database according to each of the material vectors;

[0093] The document material database is obtained by combining the text database and the vector database.

[0094] An embodiment of the present application provides a method for constructing a document material database, which also uses document disassembly technology to disassemble historical bidding documents into several material document titles and corresponding answers to each title, and generates material vectors corresponding to each of the material document titles through a text processing model. Finally, text data and vector data are used to construct different types of databases, namely text databases and vector databases, wherein the vector database is used for vector matching, and the text database is used to provide corresponding text content, thereby providing necessary text data for the subsequent generation of bidding documents and improving the efficiency of bidding document production.

[0095] Furthermore, after obtaining a plurality of historical bidding documents, each of the historical bidding documents is classified into authority levels according to a preset authority level target;

[0096] When adding a document tag to each of the material document titles in the manner of keyword matching, a corresponding authority level is added to each of the material document titles.

[0097] In a preferred embodiment, the authority level is divided into three levels, including: company-level bidding materials (shareholder information, director information, company compliance system, etc.), department-level bidding materials (personnel structure, department regulations, etc.), and historical winning bid documents (XXX enterprise annuity bidding documents, etc.).

[0098] In an embodiment of the present application, historical bid documents are further classified into authority levels, so that the authority level of each text content can be referenced in the subsequent similarity matching process, further improving the accuracy of similarity matching.

[0099] In a preferred embodiment, the document material database can be continuously supplemented with new document materials, specifically:

[0100] Manually upload document materials to a specific file directory, and divide them into multiple bid materials of different authority levels based on comprehensive considerations such as the compliance level and business matching level of the bid materials, and set corresponding bid material directories of different authority levels on the system; the program automatically identifies the minimum title and its content in the document; the program extracts the minimum title and content into a whole paragraph, and processes the title into a question, and the content into an answer; the question is automatically labeled in the tag library based on keyword matching; finally, the corresponding question, answer and label are stored in the text database, the question is stored as a dictionary key, and the corresponding text ID attribute is generated, the answer is stored as a dictionary value, and the label is stored as an attribute, thus completing the disassembly and storage of the disassembled bid materials.

[0101] Furthermore, a large language model is used to abstract the corresponding question summary. After the abstraction is completed, the natural language model converts the question into a corresponding feature vector and stores it in the vector database. Furthermore, the system's front-end bid preparation software can view the disassembled questions and answers by specific file directory, and can perform high-precision and fuzzy matching searches on the questions and answers.

[0102] In a preferred embodiment, the bid generation method based on feature vector matching mainly consists of five steps: Figure 2 As shown, including:

[0103] Material library maintenance: In daily work, materials that can be used to create bid documents are uploaded to sub-material libraries with different authority scores based on their authority. The materials include candidate bid documents and the corresponding topics. Based on the uploaded materials, the system automatically uses the large language model and natural language model to automatically extract the abstract and feature vector of the topic in the materials, store the feature vector in the vector database, and store the documents in the document database.

[0104] Automatic splitting of bidding documents: After the bidding documents are uploaded via the web page, the system will split the bidding documents into independent questions according to the questions through document processing technology, so as to automatically match the questions later;

[0105] Feature vector generation: A pre-trained large language model is used to extract the summary of the decomposed question and a corresponding feature vector is generated using a natural language model.

[0106] Similarity matching: The generated feature vector is matched with the feature vectors of pre-generated candidate content titles for tender documents using a vector database, and the candidate content title that is closest to the feature vector corresponding to each single question summary in the tender document is selected;

[0107] Bid document generation: Search the document database for the corresponding candidate content based on the ID of the closest candidate content question, and generate a response bid document based on the candidate answer content that matches each question.

[0108] The embodiment of the present application provides a method for generating bid documents based on feature vector matching, which converts the bidding documents into several feature vectors through document decomposition technology and text processing models, and converts large-scale bidding documents into multiple feature vectors that are more conducive to computer understanding, which is conducive to subsequent vector similarity matching in the document material library, and improves the matching efficiency of similarity matching. Furthermore, in the embodiment of the present application, the document material database is obtained by decomposing and classifying and storing several historical bidding documents based on the document decomposition technology, and converting the bidding documents and document materials into vector dimensions for matching, which greatly reduces the computational complexity of document matching. Compared with the existing method of manually producing bid documents, the embodiment of the present application realizes the automatic generation of bid documents through document decomposition, conversion and vector similarity matching, thereby improving the efficiency of bid document production.

[0109] Example 2:

[0110] like Figure 3 As shown, accordingly, embodiment 2 provides a tender document generation system based on feature vector matching, including an acquisition module 10, a document decomposition module 20, a feature vector generation module 30, a vector matching module 40 and a tender document generation module 50;

[0111] Wherein, the acquisition module 10 is used to obtain the bidding documents of the bidding project;

[0112] The document disassembly module 20 is used to disassemble the bidding document into a plurality of document titles using a preset document disassembly technology;

[0113] The feature vector generating module 30 is used to input each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles;

[0114] The vector matching module 40 is used to perform similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and categorizing a number of historical bidding documents based on the document disassembly technology;

[0115] The bid document generating module 50 is used to generate a bid document according to the titles of the documents and the text contents.

[0116] Furthermore, the feature vector generation module 30 generates feature vectors corresponding to the respective document titles, including:

[0117] Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles;

[0118] Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

[0119] In one possible implementation, the vector matching module 40 performs similarity matching in a preset document material database based on each of the feature vectors to determine the text content corresponding to each of the feature vectors, including:

[0120] performing similarity matching in a vector database of the document material database based on each of the feature vectors, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors;

[0121] For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector;

[0122] According to the ID of each first material vector, ID matching is performed in the text database of the document material database, thereby determining the text content corresponding to each of the feature vectors.

[0123] In one possible implementation, the document material database is obtained by disassembling and categorizing and storing a number of historical bidding documents based on the document disassembly technology, including:

[0124] Obtain several historical tender documents;

[0125] For each of the historical bid documents, using the document decomposition technology, the historical bid document is decomposed into a plurality of source document topics and answers corresponding to the respective source document topics;

[0126] Adding document tags to the titles of the respective source documents in a keyword matching manner, and establishing corresponding relationships between the titles of the respective source documents and the respective historical bid documents;

[0127] Constructing a text database based on the titles of the respective source documents, the corresponding document tags, and the corresponding answers to the respective titles;

[0128] Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles;

[0129] Constructing a vector database according to each of the material vectors;

[0130] The document material database is obtained by combining the text database and the vector database.

[0131] The embodiment of the present application provides a tender document generation system based on feature vector matching, which converts the tender document into a number of feature vectors through document decomposition technology and text processing models, and converts a large tender document into a plurality of feature vectors that are more conducive to computer understanding, which is conducive to the subsequent vector similarity matching in the document material library, and improves the matching efficiency of similarity matching. Furthermore, in the embodiment of the present application, the document material database is obtained by decomposing and classifying a number of historical tender documents based on the document decomposition technology, and converting the tender document and the document material into vector dimensions for matching, which greatly reduces the computational complexity of document matching. Compared with the existing method of manually making tender documents, the embodiment of the present application realizes the automatic generation of tender documents through the decomposition, conversion and vector similarity matching of documents, thereby improving the efficiency of tender document production.

[0132] The more detailed working principle and process flow of this embodiment can be referred to, but not limited to, the relevant records of the first embodiment.

[0133] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application by those skilled in the art should be included within the scope of protection of this application.

Claims

1. A method for generating a bid based on feature vector matching, characterized in that: include: Obtain the bidding documents for bidding projects; Using a preset document decomposition technology to decompose the bidding document into a plurality of document titles; Inputting each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles; According to each of the feature vectors, similarity matching is performed in a preset document material database to determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and classifying and storing a number of historical bidding documents based on the document disassembly technology; Generate a tender document based on the titles of the various documents and the contents of the various texts.

2. A method for generating a bid based on feature vector matching as claimed in claim 1, characterized in that: The text processing model generates feature vectors corresponding to the respective document titles, including: Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles; Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

3. A method for generating a bid based on feature vector matching as claimed in claim 1, characterized in that: The similarity matching is performed in a preset document material database according to each of the feature vectors to determine the text content corresponding to each of the feature vectors, including: According to each of the feature vectors, similarity matching is performed in the vector database of the document material database, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors; For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector; According to the ID of each first material vector, ID matching is performed in the text database of the document material database, so as to determine the text content corresponding to each of the feature vectors.

4. A method for generating a bid based on feature vector matching as claimed in claim 3, characterized in that: The similarity matching is performed in a vector database of the document material database according to each of the feature vectors, and then a plurality of material vectors are matched for each of the feature vectors, including: For each of the feature vectors, the Euclidean distance between the feature vector and all feature vectors in the vector database is calculated, and a preset number of material vectors having the closest Euclidean distance to the feature vector are obtained by screening.

5. A method for generating a bid based on feature vector matching as claimed in claim 1, characterized in that: The document material database is obtained by disassembling and classifying and storing a number of historical bidding documents based on the document disassembly technology, including: Obtain certain historical tender documents; For each of the historical tender documents, using the document decomposition technology, decompose the historical tender document into a plurality of material document titles and answers to the titles corresponding to the respective material document titles; Adding document tags to the titles of the respective material documents in a keyword matching manner, and establishing a corresponding relationship between the titles of the respective material documents and the respective historical bidding documents; Constructing a text database according to each of the material document titles, the corresponding document tags, and the corresponding answers to the titles; Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles; Constructing a vector database according to each of the material vectors; The document material database is obtained by combining the text database with the vector database.

6. A method for generating a bid based on feature vector matching as claimed in claim 5, characterized in that: After obtaining a plurality of historical bidding documents, classifying the authority level of each of the historical bidding documents according to a preset authority level target; When adding document tags to each of the material document titles in the manner of keyword matching, a corresponding authority level is added to each of the material document titles.

7. A tender document generation system based on feature vector matching, characterized in that: It includes acquisition module, document disassembly module, feature vector generation module, vector matching module and bid generation module; Wherein, the acquisition module is used to acquire the bidding documents of the bidding project; The document disassembly module is used to disassemble the bidding document into a plurality of document titles using a preset document disassembly technology; The feature vector generation module is used to input each of the document titles into a preset text processing model, so that the text processing model generates a feature vector corresponding to each of the document titles; The vector matching module is used to perform similarity matching in a preset document material database according to each of the feature vectors, and determine the text content corresponding to each of the feature vectors, wherein the document material database is obtained by disassembling and classifying and storing a number of historical bidding documents based on the document disassembly technology; The bid document generation module is used to generate a bid document file according to the titles of the various documents and the contents of the various texts.

8. A bidding document generation system based on feature vector matching as claimed in claim 7, characterized in that: The feature vector generation module generates feature vectors corresponding to the respective document titles, including: Inputting each of the document titles into the large language model in the text processing model, so that the large language model abstractly understands and converts the meaning of each of the document titles, and generates a title summary corresponding to each of the document titles; Each of the topic summaries is input into the natural language model in the text processing model, so that the natural language model performs feature vector conversion on each of the topic summaries to generate feature vectors corresponding to each of the topic summaries.

9. A bidding document generation system based on feature vector matching as claimed in claim 7, characterized in that: The vector matching module performs similarity matching in a preset document material database according to each of the feature vectors to determine the text content corresponding to each of the feature vectors, including: According to each of the feature vectors, similarity matching is performed in the vector database of the document material database, thereby matching a plurality of material vectors for each of the feature vectors, and determining a correlation score between each of the feature vectors and the corresponding material vectors; For each of the feature vectors, determining a first material vector uniquely corresponding to each of the feature vectors according to a correlation score between the feature vector and each corresponding material vector; According to the ID of each first material vector, ID matching is performed in the text database of the document material database, so as to determine the text content corresponding to each of the feature vectors.

10. A bidding document generation system based on feature vector matching as claimed in claim 7, characterized in that: The document material database is obtained by disassembling and classifying and storing a number of historical bidding documents based on the document disassembly technology, including: Obtain certain historical tender documents; For each of the historical tender documents, using the document decomposition technology, decompose the historical tender document into a plurality of material document titles and answers to the titles corresponding to the respective material document titles; Adding document tags to the titles of the respective material documents in a keyword matching manner, and establishing a corresponding relationship between the titles of the respective material documents and the respective historical bidding documents; Constructing a text database according to each of the material document titles, the corresponding document tags, and the corresponding answers to the titles; Inputting each of the material document titles into the text processing model so that the text processing model generates a material vector corresponding to each of the material document titles; Constructing a vector database according to each of the material vectors; The document material database is obtained by combining the text database with the vector database.

Images