AI large model enterprise knowledge base system based on cue words

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an AI large model enterprise knowledge base system based on cue words, comprising a front end interaction module; a large model module; a knowledge base module; a vector database module; a fine tuning module; a function calling module; and a monitoring and optimizing module. According to the AI large model enterprise knowledge base system based on the cue words, knowledge documents in an enterprise are stored and managed in a centralized mode, efficient knowledge retrieval and matching are conducted through the vector database module, the knowledge base module can be customized according to enterprise requirements, document storage of various formats is supported, a convenient document management function is provided, and the knowledge management efficiency is improved. And the fine tuning module integrates enterprise knowledge into the large model, so that the large model can understand and process special knowledge and implicit knowledge in the enterprise, thereby improving the specificity and knowledge application capability of the system, effectively integrating and managing the enterprise knowledge, converting the enterprise knowledge into operable information, and providing powerful support for decision and operation of the enterprise.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI large-model enterprise knowledge base system based on prompt words. Background Art

[0002] With the rapid development of artificial intelligence technology, big models have shown great potential in text generation and natural language understanding. These models can process and understand complex language information, generate high-quality text content, and extract key information from massive data. In terms of text generation, big models can automatically generate various types of texts, such as news reports, product descriptions, marketing copy, etc., greatly improving the efficiency and quality of content production. In terms of natural language understanding, big models can understand users' natural language instructions and perform corresponding operations based on users' intentions, such as question-and-answer systems, intelligent customer service, voice assistants, etc. These application scenarios have greatly improved the user experience and brought great convenience to all walks of life.

[0003] However, the existing big models still have some limitations in enterprise knowledge management. First, knowledge acquisition is difficult. Big models require a lot of data for training, and the internal knowledge of the enterprise is often scattered in different systems and documents, which is difficult to effectively integrate and utilize. Secondly, knowledge update is not timely. Enterprise knowledge is dynamically changing, and big model training requires a lot of time and resources, making it difficult to achieve real-time updating of knowledge. Finally, understanding is limited. The understanding ability of big models is limited by training data. For the proprietary knowledge and tacit knowledge within the enterprise, the understanding of big models is still limited. These limitations restrict the efficiency and ability of enterprises in knowledge management, and further technological innovation and application exploration are needed. Therefore, we propose an AI big model enterprise knowledge base system based on prompt words. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides an AI big model enterprise knowledge base system based on prompt words, which has the advantages of efficient knowledge integration, real-time knowledge updating and flexible function calling, and solves some limitations of the existing big models in enterprise knowledge management. First, knowledge acquisition is difficult. The big model requires a large amount of data for training, while the internal knowledge of the enterprise is often scattered in different systems and documents, which is difficult to effectively integrate and utilize. Secondly, knowledge updating is not timely. Enterprise knowledge is dynamically changing, and big model training requires a lot of time and resources, making it difficult to achieve real-time updating of knowledge. Finally, understanding is limited. The understanding ability of the big model is limited by the training data. For the proprietary knowledge and implicit knowledge within the enterprise, the big model's understanding is still limited. These limitations limit the efficiency and ability of the enterprise in knowledge management, and require further technological innovation and application exploration.

[0006] (II) Technical solution

[0007] In order to achieve the above-mentioned purpose of efficient knowledge integration, real-time knowledge update and flexible function calling, the present invention provides the following technical solution: an AI large model enterprise knowledge base system based on prompt words, comprising:

[0008] Front-end interaction module: used to receive user query requests and convert them into prompt words;

[0009] Large model module: used for semantic understanding and knowledge processing based on pre-trained deep learning models;

[0010] Knowledge base module: stores and manages enterprise knowledge documents and provides a retrieval interface;

[0011] Vector database module: stores vectorized representations of knowledge documents and supports fast retrieval;

[0012] Fine-tuning module: Integrate enterprise knowledge into the big model to improve professional knowledge understanding;

[0013] Function call module: realizes the interaction between large models and external systems;

[0014] Monitoring and optimization module: monitor system performance, collect feedback, and optimize models.

[0015] Preferably, the front-end interaction module further includes a user interface and a prompt word generator, and the large model module further includes: a pre-trained language model based on the Transformer architecture, a PyTorch deep learning framework, and an Adam model optimizer.

[0016] Preferably, the knowledge base module further includes a document management system and a document vectorization tool, and the vector database module further includes a vector storage system and a vector search algorithm.

[0017] Preferably, the fine-tuning module further includes a data set selector and a hyperparameter adjuster, the function call module further includes a function interface definer and a function call executor, and the monitoring and optimization module further includes a performance monitor, a feedback analyzer and a model optimizer.

[0018] Preferably, the prompt word-based AI large model enterprise knowledge base system also includes a security mechanism for ensuring access control and data protection of the knowledge base, including user identity authentication, permission control and data encryption.

[0019] Preferably, the prompt word-based AI large model enterprise knowledge base system also includes an extension interface for supporting the scalability of the system and integration with other systems, including a plug-in architecture, an API gateway and a message queue.

[0020] Preferably, the big model application architecture of the AI ​​big model enterprise knowledge base system based on prompt words includes:

[0021] Application module: used to receive user input and convert it into prompt words;

[0022] Base large model: one of GPT or ERN IE, used to generate responses and perform tasks;

[0023] Prompt: used to guide the basic large model to generate a specific type of response;

[0024] APIs: used to interact with third-party services and internal services;

[0025] RAG module: used to combine retrieval and generation techniques to improve text generation effects;

[0026] Embeddings technology: used to map data into a low-dimensional vector space;

[0027] Finite tuning process: used to integrate knowledge from the knowledge base into the large model;

[0028] Function Calling mechanism: used to realize the interaction between large models and applications.

[0029] Preferably, the application method of the AI ​​large model enterprise knowledge base system based on prompt words includes the following process:

[0030] S1. The user enters a query through the user interface. The front-end module is responsible for converting the user's natural language query into a structured query statement and sending it to the large model module for processing;

[0031] S2. The front-end module further converts the structured query into prompt words. The front-end module needs to design appropriate prompt word templates based on the user's query intent and context information so that the large model module can accurately understand the user's query;

[0032] S3. After receiving the prompt word, the large model module first performs semantic understanding on the prompt word and extracts key information from it. Based on the extracted information, the large model module retrieves knowledge fragments related to the query in the vector database. The vector database stores the vectorized representation of enterprise knowledge documents. Through the vector similarity search algorithm, the knowledge fragments related to the query can be quickly found.

[0033] S4, the vector database returns knowledge fragments related to the query according to the retrieval request of the large model module, and the knowledge fragments include text, pictures and video content;

[0034] S5, the large model module conducts a comprehensive analysis of the returned knowledge fragments, combines the user's query intent and context information, and generates answers and actions;

[0035] S6. If the large model module interacts with the external system, the function calling module is responsible for implementing these interactions. The function calling module calls the interface provided by the external system according to the instructions of the large model module and returns the results to the large model module;

[0036] S7. The monitoring module continuously monitors system performance and user feedback, collects relevant data, analyzes the performance of the model based on the collected data, and identifies potential problems and room for improvement. The monitoring module feeds back the analysis results to the fine-tuning module to optimize the large model and improve the accuracy and efficiency of the model.

[0037] (III) Beneficial effects

[0038] Compared with the prior art, the present invention provides an AI large model enterprise knowledge base system based on prompt words, which has the following beneficial effects:

[0039] 1. This AI big model enterprise knowledge base system based on prompt words, by building a knowledge base module, centrally stores and manages the knowledge documents within the enterprise, and uses the vector database module for efficient knowledge retrieval and matching. The knowledge base module can be customized according to the needs of the enterprise, supports document storage in multiple formats, and provides convenient document management functions, such as document classification, tag management, version control, etc. The vector database module vectorizes the knowledge documents and uses the vector similarity search algorithm for rapid retrieval, which can quickly find the knowledge fragments related to the user query, thereby improving the efficiency of knowledge acquisition. The fine-tuning module integrates the enterprise knowledge into the big model, so that the big model can understand and process the proprietary knowledge and implicit knowledge within the enterprise, thereby enhancing the system's specificity and knowledge application capabilities. The collaborative work of these modules enables the system to effectively integrate and manage enterprise knowledge and transform it into actionable information, providing strong support for the decision-making and operation of the enterprise.

[0040] 2. The AI ​​large model enterprise knowledge base system based on prompt words, the vector database module supports rapid retrieval and updating of documents in the knowledge base, can timely add, delete and modify knowledge documents, and use the incremental update mechanism to ensure the timeliness and accuracy of knowledge. The fine-tuning module can be fine-tuned according to changes in enterprise knowledge, so that the large model can adapt to changes in knowledge in a timely manner, ensuring that the system can always process the latest knowledge information. These mechanisms ensure the dynamic update and real-time nature of the system knowledge base, enabling enterprises to keep abreast of market dynamics and industry trends in a timely manner and adjust strategies and decisions accordingly.

[0041] 3. The AI ​​large-model enterprise knowledge base system based on prompt words, the function call module defines the function interface and interacts with the external system. For example, the system can be integrated with the enterprise's ERP system, CRM system, OA system, etc. to achieve knowledge sharing and collaborative work. The extended interface includes plug-in architecture, API gateway and message queue, etc., which can easily expand the system's functions and integrate with other systems. The design of these interfaces and architectures enables the system to flexibly connect and interact with other systems to achieve information sharing and business collaboration, thereby improving the overall information level and operational efficiency of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a structural diagram of the AI ​​large model enterprise knowledge base system based on prompt words of the present invention;

[0043] Figure 2 The present invention is a flow chart of the application method of the prompt word-based AI large model enterprise knowledge base system. DETAILED DESCRIPTION

[0044] The following will be combined with the embodiments of the present invention and the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments 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.

[0045] See also Figure 1-2 , an AI large model enterprise knowledge base system based on prompt words, including:

[0046] Front-end interaction module: used to receive user query requests and convert them into prompt words;

[0047] Large model module: used for semantic understanding and knowledge processing based on pre-trained deep learning models;

[0048] Knowledge base module: stores and manages enterprise knowledge documents and provides a retrieval interface;

[0049] Vector database module: stores vectorized representations of knowledge documents and supports fast retrieval;

[0050] Fine-tuning module: Integrate enterprise knowledge into the big model to improve professional knowledge understanding;

[0051] Function call module: realizes the interaction between large models and external systems;

[0052] Monitoring and optimization module: monitor system performance, collect feedback, and optimize models.

[0053] Furthermore, the front-end interaction module further includes a user interface and a prompt word generator, and the large model module further includes: a pre-trained language model based on the Transformer architecture, a PyTorch deep learning framework, and an Adam model optimizer.

[0054] Specifically, the front-end interaction module serves as a bridge for users to interact with the system. It is responsible for receiving user query requests and converting them into prompt words so that the large model module can accurately understand user needs. The user interface provides a simple and intuitive operation interface to facilitate users to enter queries or ask questions. The prompt word generator designs appropriate prompt word templates based on user input and context information to guide the large model module to generate high-quality answers or perform corresponding actions. For example, when a user enters "How do I apply for annual leave?", the prompt word generator can generate a prompt word of "Please tell me how to apply for annual leave?" to guide the large model module to understand the user's intention and find relevant knowledge documents;

[0055] As the core of the system, the large model module is responsible for semantic understanding and knowledge processing. Its performance directly affects the intelligence level of the system. The pre-trained language model based on the Transformer architecture has powerful language understanding and generation capabilities, and can accurately understand user query intentions and generate relevant answers. The PyTorch deep learning framework provides a flexible and easy-to-use development environment for researchers to train and optimize models. The Adam model optimizer can effectively update model parameters and improve the model's learning efficiency and generalization ability.

[0056] Furthermore, the knowledge base module further includes a document management system and a document vectorization tool, and the vector database module further includes a vector storage system and a vector search algorithm.

[0057] Specifically, the knowledge base module is the core of the system's knowledge storage and management. Its role is to centrally store and manage the enterprise's internal knowledge documents and provide convenient document management functions. The document management system is responsible for the storage, classification, tag management, version control, etc. of knowledge documents, which facilitates users to quickly find and manage knowledge documents. The document vectorization tool converts knowledge documents into vector representations to facilitate subsequent vector retrieval and matching;

[0058] The vector database module is responsible for storing the vectorized representation of knowledge documents and supports fast retrieval and matching. The vector storage system is responsible for storing a large amount of vector data and provides efficient storage and retrieval capabilities. The vector search algorithm calculates the similarity between the user's query vector and the vectors in the knowledge base to quickly find knowledge fragments related to the user's query. For example, when a user enters the query "How do I apply for annual leave?", the vector database module will retrieve similar knowledge fragments in the knowledge base based on the user's query vector and return relevant documents, such as "Employee Handbook", "Annual Leave Application Process", etc.

[0059] The collaborative work of the knowledge base module and the vector database module realizes the centralized management and efficient retrieval of enterprise knowledge, and improves the efficiency and effectiveness of enterprise knowledge management.

[0060] Furthermore, the fine-tuning module further includes a dataset selector and a hyperparameter adjuster, the function call module further includes a function interface definer and a function call executor, and the monitoring and optimization module further includes a performance monitor, a feedback analyzer and a model optimizer.

[0061] Specifically, the fine-tuning module integrates enterprise knowledge into the big model to improve the model's ability to understand and apply professional knowledge. The dataset selector is responsible for selecting high-quality datasets related to enterprise knowledge to provide training data for model fine-tuning. The hyperparameter adjuster is responsible for adjusting model parameters, such as learning rate and batch size, to improve the performance and generalization ability of the model. Through the fine-tuning module, the big model can better understand the professional knowledge and implicit knowledge within the enterprise, thereby improving the accuracy and practicality of the knowledge base system.

[0062] The function call module is responsible for realizing the interaction between the big model and external systems, such as integration with ERP systems, CRM systems, etc., to achieve knowledge sharing and collaborative work. The function interface definer is responsible for defining the function interface and providing interface documents to the external system to facilitate external system calls. The function call executor is responsible for executing the function call instructions issued by the big model module and returning the results to the big model module. For example, when the big model module needs to query the annual leave information of employees in the ERP system, the function call module will call the API of the ERP system according to the defined interface and return the query results to the big model module.

[0063] The monitoring and optimization module is responsible for monitoring system performance, collecting user feedback, and optimizing the large model based on the analysis results to improve the accuracy and efficiency of the model. The performance monitor is responsible for monitoring the system operation status, such as response time, resource utilization, etc., to promptly identify system performance bottlenecks and optimize them. The feedback analyzer collects user feedback, such as satisfaction with answers, system usage experience, etc., and identifies potential problems and room for improvement based on the analysis results. The model optimizer adjusts and optimizes the large model based on the results of the performance monitor and feedback analyzer, such as adjusting model parameters, updating training data, etc., to improve the model performance and user experience.

[0064] Furthermore, the prompt word-based AI large-model enterprise knowledge base system also includes a security mechanism for ensuring access control and data protection of the knowledge base, including user authentication, permission control, and data encryption.

[0065] Specifically, the security mechanism is an important part of the knowledge base system. Its role is to ensure access control and data security of the knowledge base, and to prevent unauthorized access and data leakage. The user authentication mechanism is responsible for verifying the identity of the user, ensuring that only authorized users can access the knowledge base. The permission control mechanism restricts the user's operation on the knowledge base according to the user's role and permission. For example, only some users are allowed to access sensitive information or modify knowledge documents. The data encryption mechanism encrypts the data in the knowledge base to prevent the data from being stolen or tampered with during transmission and storage.

[0066] The process of the security mechanism is as follows:

[0067] Step 1: When a user accesses the knowledge base system, he needs to enter his username and password for identity authentication;

[0068] Step 2: The system verifies the user's identity based on the user name and password, and obtains the user's permission information;

[0069] Step 3: The user performs operations based on the permission information, such as querying and modifying knowledge documents;

[0070] Step 4: The system controls the permissions of user operations to prevent unauthorized access and data leakage;

[0071] Step 5: The system encrypts the data in the knowledge base to prevent the data from being stolen or tampered with during transmission and storage;

[0072] The effective implementation of security mechanisms can ensure the security of the enterprise knowledge base, prevent knowledge leakage and abuse, and safeguard the interests and competitiveness of the enterprise.

[0073] Furthermore, the prompt word-based AI large model enterprise knowledge base system also includes an extension interface to support the scalability of the system and integration with other systems, including a plug-in architecture, an API gateway, and a message queue.

[0074] Specifically, the extension interface is an important part of the knowledge base system. Its role is to support the scalability of the system and its integration with other systems, so that the system can adapt to the ever-changing needs and business scenarios of the enterprise. The plug-in architecture allows developers to develop plug-ins to extend system functions, such as developing new query functions, data analysis functions, etc. The API gateway provides a unified interface specification and protocol to facilitate other systems to call the API of the knowledge base system to achieve data sharing and business collaboration. The message queue is responsible for asynchronously processing messages to improve the response speed and reliability of the system.

[0075] The process of extending the interface is as follows:

[0076] Step 1: Developers develop new functional modules based on the plug-in architecture provided by the system, such as new query functions, data analysis functions, etc.

[0077] Step 2: The developer registers the function module into the system and provides an interface for other systems to call through the API gateway;

[0078] Step 3: Other systems call the knowledge base system's API through the API gateway, such as querying knowledge documents, obtaining data reports, etc.

[0079] Step 4: The system executes the corresponding function module according to the API call request and returns the result to the caller;

[0080] Step 5: The system processes messages asynchronously through the message queue. For example, a user query request is sent to the message queue, which is processed by the background processing program and the result is returned to the user.

[0081] The effective implementation of the extended interface can improve the flexibility and scalability of the knowledge base system, facilitate enterprises to customize and expand it according to their own needs, and integrate with other systems to achieve data sharing and business collaboration, thereby improving the overall information level and operational efficiency of the enterprise.

[0082] Furthermore, the big model application architecture of the AI ​​big model enterprise knowledge base system based on prompt words includes:

[0083] Application module: used to receive user input and convert it into prompt words;

[0084] Base large model: one of GPT or ERN IE, used to generate responses and perform tasks;

[0085] Prompt: used to guide the basic large model to generate a specific type of response;

[0086] APIs: used to interact with third-party services and internal services;

[0087] RAG module: used to combine retrieval and generation techniques to improve text generation effects;

[0088] Embeddings technology: used to map data into a low-dimensional vector space;

[0089] Finite tuning process: used to integrate knowledge from the knowledge base into the large model;

[0090] Function Calling mechanism: used to realize the interaction between large models and applications.

[0091] Specifically, the big model application architecture is the core of the knowledge base system. Its function is to receive user input, use the big model for semantic understanding and knowledge processing, and finally generate answers or perform corresponding tasks. This architecture has the following advantages:

[0092] Efficient knowledge processing: Based on the pre-trained language model, it can quickly understand user input and generate relevant answers, improving the efficiency of knowledge acquisition;

[0093] Flexible function calling: supports integration with third-party services and internal systems to achieve knowledge sharing and business collaboration;

[0094] Accurate text generation: Combining retrieval and generation technologies to generate more accurate and fluent text content;

[0095] Powerful data expression capability: Embeddings technology is used to map data into a low-dimensional vector space, facilitating subsequent vector retrieval and matching;

[0096] The process of the large model application architecture is as follows:

[0097] Step 1: The application module receives user input: The user enters a query or asks a question through the user interface;

[0098] Step 2: The application module converts the user input into prompt words: Based on the user input and context information, a suitable prompt word template is designed to guide the large model module to generate a specific type of response;

[0099] Step 3: The basic large model performs semantic understanding and knowledge processing: Based on the pre-trained language model, it understands the user's intention and generates relevant answers or performs corresponding tasks;

[0100] Step 4: APIs interact with third-party services and internal systems: for example, integrate with ERP systems, CRM systems, etc., to obtain relevant data or perform related operations;

[0101] Step 5: The RAG module combines retrieval and generation technologies to improve text generation effects: Retrieval technology is used to obtain relevant knowledge fragments, and generation technology is combined to generate more accurate and fluent text content;

[0102] Step 6: Embeddings technology maps the data into a low-dimensional vector space to facilitate subsequent vector retrieval and matching;

[0103] Step 7: The Finning process integrates the knowledge base into the large model: improving the model's ability to understand and apply professional knowledge;

[0104] Step 8: The Function Calling mechanism realizes the interaction between the large model and the application: the output results of the large model are returned to the application module and displayed to the user;

[0105] Through the large model application architecture, the knowledge base system can efficiently process user queries and generate high-quality answers or perform corresponding tasks, thereby improving enterprise knowledge management efficiency and user experience.

[0106] The application method of the AI ​​large model enterprise knowledge base system based on prompt words includes the following processes:

[0107] S1. The user enters a query through the user interface. The front-end module is responsible for converting the user's natural language query into a structured query statement and sending it to the large model module for processing;

[0108] S2. The front-end module further converts the structured query into prompt words. The front-end module needs to design appropriate prompt word templates based on the user's query intent and context information so that the large model module can accurately understand the user's query;

[0109] S3. After receiving the prompt word, the large model module first performs semantic understanding on the prompt word and extracts key information from it. Based on the extracted information, the large model module retrieves knowledge fragments related to the query in the vector database. The vector database stores the vectorized representation of enterprise knowledge documents. Through the vector similarity search algorithm, the knowledge fragments related to the query can be quickly found.

[0110] S4, the vector database returns knowledge fragments related to the query according to the retrieval request of the large model module, and the knowledge fragments include text, pictures and video content;

[0111] S5, the large model module conducts a comprehensive analysis of the returned knowledge fragments, combines the user's query intent and context information, and generates answers and actions;

[0112] S6. If the large model module interacts with the external system, the function calling module is responsible for implementing these interactions. The function calling module calls the interface provided by the external system according to the instructions of the large model module and returns the results to the large model module;

[0113] S7. The monitoring module continuously monitors system performance and user feedback, collects relevant data, analyzes the performance of the model based on the collected data, and identifies potential problems and room for improvement. The monitoring module feeds back the analysis results to the fine-tuning module to optimize the large model and improve the accuracy and efficiency of the model.

[0114] In summary, the AI ​​large model enterprise knowledge base system based on prompt words, by building a knowledge base module, centrally stores and manages the knowledge documents within the enterprise, and uses the vector database module for efficient knowledge retrieval and matching. The knowledge base module can be customized according to the needs of the enterprise, supports document storage in multiple formats, and provides convenient document management functions, such as document classification, tag management, version control, etc. The vector database module vectorizes the knowledge documents and uses the vector similarity search algorithm for rapid retrieval, which can quickly find the knowledge fragments related to the user query, thereby improving the efficiency of knowledge acquisition. The fine-tuning module integrates the enterprise knowledge into the large model, so that the large model can understand and process the proprietary knowledge and implicit knowledge within the enterprise, thereby improving the system's specificity and knowledge application capabilities. The collaborative work of these modules enables the system to effectively integrate and manage enterprise knowledge and transform it into actionable information, providing strong support for the decision-making and operation of the enterprise.

[0115] In addition, the AI ​​large model enterprise knowledge base system based on prompt words, the vector database module supports rapid retrieval and updating of documents in the knowledge base, can timely add, delete and modify knowledge documents, and use the incremental update mechanism to ensure the timeliness and accuracy of knowledge. The fine-tuning module can be fine-tuned according to changes in enterprise knowledge, so that the large model can adapt to changes in knowledge in a timely manner, ensuring that the system can always process the latest knowledge information. These mechanisms ensure the dynamic update and real-time nature of the system knowledge base, enabling enterprises to keep abreast of market dynamics and industry trends in a timely manner and adjust strategies and decisions accordingly.

[0116] In addition, the AI ​​large-model enterprise knowledge base system based on prompt words has a function call module that defines function interfaces and interacts with external systems. For example, the system can be integrated with the enterprise's ERP system, CRM system, OA system, etc. to achieve knowledge sharing and collaborative work. The extended interface includes a plug-in architecture, API gateway, and message queue, which can easily expand the system's functions and integrate with other systems. The design of these interfaces and architectures enables the system to flexibly connect and interact with other systems to achieve information sharing and business collaboration, thereby improving the overall information level and operational efficiency of the enterprise.

[0117] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements of this system. The improvements of this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI large model enterprise knowledge base system based on prompt words, characterized in that: include: Front-end interaction module: used to receive user query requests and convert them into prompt words; Large model module: used for semantic understanding and knowledge processing based on pre-trained deep learning models; Knowledge base module: stores and manages enterprise knowledge documents and provides a retrieval interface; Vector database module: stores vectorized representations of knowledge documents and supports fast retrieval; Fine-tuning module: Integrate enterprise knowledge into the big model to improve professional knowledge understanding; Function call module: realizes the interaction between large models and external systems; Monitoring and optimization module: monitor system performance, collect feedback, and optimize models.

2. According to the AI ​​large model enterprise knowledge base system based on prompt words according to claim 1, it is characterized in that: The front-end interaction module further includes a user interface and a prompt word generator, and the large model module further includes: a pre-trained language model based on the Transformer architecture, a PyTorch deep learning framework, and an Adam model optimizer.

3. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The knowledge base module further includes a document management system and a document vectorization tool, and the vector database module further includes a vector storage system and a vector search algorithm.

4. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The fine-tuning module further includes a data set selector and a hyperparameter adjuster, the function call module further includes a function interface definer and a function call executor, and the monitoring and optimization module further includes a performance monitor, a feedback analyzer and a model optimizer.

5. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The prompt word-based AI large-model enterprise knowledge base system also includes a security mechanism for ensuring access control and data protection of the knowledge base, including user identity authentication, permission control and data encryption.

6. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The prompt word-based AI large-model enterprise knowledge base system also includes an extension interface for supporting the scalability of the system and integration with other systems, including a plug-in architecture, an API gateway, and a message queue.

7. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The big model application architecture of the AI ​​big model enterprise knowledge base system based on prompt words includes: Application module: used to receive user input and convert it into prompt words; The base model: one of GPT or ERNIE, used to generate responses and perform tasks; Prompt: used to guide the basic large model to generate a specific type of response; APIs: used to interact with third-party services and internal services; RAG module: used to combine retrieval and generation techniques to improve text generation effects; Embeddings technology: used to map data into a low-dimensional vector space; Fine-tuning process: used to integrate knowledge from the knowledge base into the large model; Function Calling mechanism: used to implement interaction between large models and applications.

8. The AI ​​large model enterprise knowledge base system based on prompt words according to claim 1 is characterized in that: The application method of the AI ​​large model enterprise knowledge base system based on prompt words includes the following process: S1. The user enters a query through the user interface. The front-end module is responsible for converting the user's natural language query into a structured query statement and sending it to the large model module for processing; S2. The front-end module further converts the structured query into prompt words. The front-end module needs to design appropriate prompt word templates based on the user's query intent and context information so that the large model module can accurately understand the user's query; S3. After receiving the prompt word, the large model module first performs semantic understanding on the prompt word and extracts key information from it. Based on the extracted information, the large model module retrieves knowledge fragments related to the query in the vector database. The vector database stores the vectorized representation of enterprise knowledge documents. Through the vector similarity search algorithm, the knowledge fragments related to the query can be quickly found. S4, the vector database returns knowledge fragments related to the query according to the retrieval request of the large model module, and the knowledge fragments include text, pictures and video content; S5, the large model module conducts a comprehensive analysis of the returned knowledge fragments, combines the user's query intent and context information, and generates answers and actions; S6. If the large model module interacts with the external system, the function calling module is responsible for implementing these interactions. The function calling module calls the interface provided by the external system according to the instructions of the large model module and returns the results to the large model module; S7. The monitoring module continuously monitors system performance and user feedback, collects relevant data, analyzes the performance of the model based on the collected data, and identifies potential problems and room for improvement. The monitoring module feeds back the analysis results to the fine-tuning module to optimize the large model and improve the accuracy and efficiency of the model.

Images