Intelligent bidding method and system based on multi-mode RAG

Abstract

The invention provides an intelligent bidding method and system based on a multi-mode RAG, and belongs to the technical field of artificial intelligence. Firstly, files such as pictures, tables, documents and audios in bidding materials are converted into a unified text mode, converted into vectors based on the file mode and stored in a vector database, and retrieved based on a text RAG multi-path recall strategy; the method comprises the steps that firstly, a business bidding task is divided into a plurality of agents (agents) such as qualification processing, performance processing and financial processing, technical bidding is divided into a plurality of agents such as signal technology compiling and cable technology compiling, compiling of a bidding document is completed through cooperation of the multiple agents, the business process is expanded in the mode of increasing the agents subsequently, and expansibility is improved; in addition, a chat and conversation man-machine interaction mode is adopted, compiling of business labels and technical labels is achieved, and user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent bidding method and system based on multimodal RAG. Background Art

[0002] In the existing technology, the traditional bidding document preparation system mainly relies on manual processing. This method not only leads to low overall efficiency, but more significantly, there is a high error rate in the manual operation process, which in turn leads to the risk of bid rejection and causes economic losses. Although there have been certain technical explorations dedicated to realizing the intelligent transformation of bid preparation, there are still many problems and shortcomings, mainly including the following: 1. Insufficient multimodal data retrieval capabilities: The retrieval capabilities for multimodal data such as pictures, tables, documents, and audio in bidding materials are insufficient and inefficient. 2. Lack of scalability: The process model is relatively fixed and cannot be quickly expanded based on changes in process links. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides an intelligent bidding method based on multimodal RAG.

[0004] The technical solution of the present invention is:

[0005] An intelligent bidding method based on multimodal RAG. First, the pictures, tables, documents, audio and other files in the bidding materials are converted into a unified text mode. Based on the file mode, they are converted into vectors and stored in a vector database. The method is retrieved based on the text RAG multi-way recall strategy to improve the efficiency and accuracy of the bidding system. Secondly, the commercial bidding tasks are divided into multiple agents (intelligent bodies) such as qualification processing, performance processing, and financial processing. The technical bids are divided into multiple agents such as signal technology writing and cable technology writing. Multiple agents collaborate to complete the writing of bidding documents. Subsequently, the business process is expanded by adding agents, which improves scalability. In addition, a chat dialogue human-computer interaction method is adopted to realize the writing of commercial bids and technical bids, which improves the user experience.

[0006] Furthermore, business documents are divided into chat sessions and databases. Chat sessions achieve multi-round optimization and generation of business documents through multi-round dialogues. At the same time, qualification generation agents, performance generation agents, and financial generation agents collaborate to complete the generation of the entire business document. The database achieves unified management of qualification certificates, performance, and financial status.

[0007] The specific implementation steps of the business mark are as follows:

[0008] Prepare industry-related materials including but not limited to qualification certificates, performance, and financial status documents;

[0009] Upload files to the database for unified operation and maintenance management;

[0010] The LLM big model implements intelligent screening based on the bidding document qualification requirements, personnel requirements, performance requirements, and financial requirements;

[0011] The LLM large model intelligently calls qualification generation, performance generation, and financial generation to achieve business document generation.

[0012] Furthermore, technical documents are divided into chat sessions and AI knowledge bases. Chat sessions enable multi-round optimization and writing of technical documents through multi-round dialogues. AI knowledge bases support multi-modal data such as images, tables, documents, and audio, and support multi-channel retrieval functions.

[0013] The specific implementation steps of the technical standard are as follows:

[0014] Prepare industry multimodal data including but not limited to pictures, tables, documents, and audio;

[0015] Convert industry multimodal data into a unified text mode based on a large model and implement vectorized storage;

[0016] Adopt different retrieval strategies for multi-way recall and sort based on the rerank model score;

[0017] The search results are passed to the LLM model to realize content generation.

[0018] in,

[0019] The LLM large model intelligently calls signal technology writing and cable technology writing to generate technical documents.

[0020] In addition, the present invention also provides an intelligent bidding system based on multimodal RAG, comprising:

[0021] The multimodal data processing module converts multiple modal data such as images, tables, documents, and audio into a unified text modality and performs vectorized storage in the same data dimension.

[0022] The text multi-way recall retrieval module is based on the text RAG multi-way recall strategy and re-ranks the retrieval results based on the rerank model.

[0023] In the agent collaborative processing module, the commercial bidding task is divided into several intelligent agents, and the technical bidding task is divided into several agents. Multiple agents collaborate to complete the writing of bidding documents, and the business process is subsequently expanded by adding agents.

[0024] The dialogue interaction module realizes the gradual optimization of business and technical indicators through human-computer interaction through chat.

[0025] Further,

[0026] Business documents are divided into chat sessions and databases. Chat sessions achieve multi-round optimization and generation of business documents through multi-round dialogues. At the same time, qualification generation agents, performance generation agents, and financial generation agents collaborate to complete the generation of the entire business document. The database achieves unified management of qualification certificates, performance, and financial status.

[0027] The technical standards are divided into several agents written for signal technology and cable technology.

[0028] Among them, the search strategies include Milvus vector search and Elecsearch keyword search.

[0029] The beneficial effects of the present invention are

[0030] First, it has the ability to generate business and technical intelligence, which greatly improves the efficiency of bidding document generation and saves a lot of manpower; second, it supports multi-modal data and multi-channel retrieval, which improves the accuracy of retrieval; then, it uses multiple agents to collaborate, which improves generation efficiency and enhances scalability; and finally, it uses chat dialogue human-computer interaction to enhance user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the working structure of the present invention;

[0032] Figure 2 It is a system workflow diagram of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0034] The working structure of the present invention is as follows Figure 1 As shown, the business and technical specifications are divided into chat sessions and a database. The chat session uses multiple rounds of dialogue to optimize the generation of business documents. It also collaborates with qualification generation agents, performance generation agents, and financial generation agents to complete the generation of the entire business document. The database enables unified management of qualification certificates, performance, and financial status. The technical specifications are divided into chat sessions and an AI knowledge base. The chat session uses multiple rounds of dialogue to optimize the compilation of technical documents. The AI ​​knowledge base supports multi-modal data such as images, tables, documents, and audio, and supports multi-path retrieval.

[0035] The specific implementation steps of the business mark are as follows:

[0036] 1. Prepare industry-related materials including but not limited to qualification certificates, performance, financial status and other documents.

[0037] 2. Upload the files to the database for unified operation and maintenance management.

[0038] 3. The LLM model implements intelligent screening based on bidding document qualification requirements, personnel requirements, performance requirements, financial requirements, etc.

[0039] 4. The LLM large model intelligently calls qualification generation, performance generation, financial generation and other tools to generate business documents.

[0040] The specific implementation steps of the technical standard are as follows:

[0041] 1. Prepare industry multimodal data including but not limited to pictures, tables, documents, audio and other files.

[0042] 2. Convert industry multimodal data into a unified text mode based on a large model and implement vectorized storage.

[0043] 3. Adopt different retrieval strategies such as Milvus vector retrieval, Elecsearch keyword retrieval, and other multi-way recall, and rank based on the rerank model score.

[0044] 4. Pass the search results to the LLM model to realize content generation.

[0045] 5. The LLM large model intelligently calls signal technology writing, cable technology writing and other tools to generate technical documents.

[0046] The above description is only a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. An intelligent bidding method based on multimodal RAG, characterized in that: First, the images, tables, documents, and audio files in the bidding materials are converted into a unified text modality. Based on the file modality, they are converted into vectors and stored in a vector database. They are then retrieved based on the text RAG multi-way recall strategy. Secondly, the commercial bidding task is divided into several intelligent agents, and the technical bidding task is divided into several intelligent agents. The intelligent agents collaborate to complete the preparation of bidding documents. The business process can be expanded later by adding agents. In addition, chat dialogue human-computer interaction is used to realize the writing of commercial and technical bids.

2. The method according to claim 1, characterized in that Business documents are divided into chat sessions and databases. Chat sessions achieve multi-round optimization and generation of business documents through multi-round dialogues. At the same time, qualification generation agents, performance generation agents, and financial generation agents collaborate to complete the generation of the entire business document. The database achieves unified management of qualification certificates, performance, and financial status.

3. The method according to claim 2, characterized in that The specific implementation steps of the business mark are as follows: Prepare industry-related materials including qualification certificates, performance, and financial status documents; Upload files to the database for unified operation and maintenance management; The LLM big model implements intelligent screening based on the bidding document qualification requirements, personnel requirements, performance requirements, and financial requirements; The LLM large model intelligently calls qualification generation, performance generation, and financial generation to achieve business document generation.

4. The method according to claim 1, wherein Technical documents are divided into chat sessions and AI knowledge bases. Chat sessions enable multi-round optimization and writing of technical documents through multiple rounds of dialogue. The AI ​​knowledge base supports multiple modal data such as pictures, tables, documents, and audio, and supports multi-channel retrieval functions.

5. The method according to claim 4, characterized in that The specific implementation steps of the technical standard are as follows: Prepare industry multimodal data including pictures, tables, documents, and audio; Convert industry multimodal data into a unified text mode based on a large model and implement vectorized storage; Adopt different retrieval strategies for multi-way recall and sort based on the rerank model score; The search results are passed to the LLM model to realize content generation.

6. The method according to claim 5, characterized in that The LLM large model intelligently calls signal technology writing and cable technology writing to generate technical documents.

7. An intelligent bidding system based on multimodal RAG, characterized in that: include: The multimodal data processing module converts multiple modal data such as images, tables, documents, and audio into a unified text modality and performs vectorized storage in the same data dimension. The text multi-way recall retrieval module is based on the text RAG multi-way recall strategy and re-ranks the retrieval results based on the rerank model; Agent collaborative processing module: Business bidding tasks are divided into several intelligent agents, and technical bidding tasks are divided into several agents. Multiple agents collaborate to complete the writing of bidding documents, and the business process can be expanded later by adding agents. The dialogue interaction module realizes the gradual optimization of business and technical indicators through human-computer interaction through chat.

8. The system according to claim 7, characterized in that Business documents are divided into chat sessions and databases. Chat sessions achieve multi-round optimization and generation of business documents through multi-round dialogues. At the same time, qualification generation agents, performance generation agents, and financial generation agents collaborate to complete the generation of the entire business document. The database achieves unified management of qualification certificates, performance, and financial status.

9. The system according to claim 7, wherein: The technical standards are divided into several agents written for signal technology and cable technology.

10. The system according to claim 7, wherein: The search strategies include Milvus vector search and Elecsearch keyword search.

Images