Web interaction method based on LLM and natural language

Abstract

The invention relates to the technical field of artificial intelligence interaction, in particular to a Web interaction method based on LLM and a natural language. According to the technical scheme, the Web interaction method based on the LLM and the natural language comprises a construction method of an LLM Agent; compared with a traditional Web interaction method based on the LLM and the natural language, the method has the advantages that the user instruction is generally analyzed through relatively fixed rules and templates, although the LLM is used for performing natural language processing, the flexibility and expansibility are limited when complex tasks are understood and executed, and the user experience is greatly improved. In the Web interaction method based on the LLM and the natural language, a user still needs to interact with the system to a certain degree through clicking or inputting, the Web interaction method based on the LLM and the natural language deeply understands all functions of the Web system by training the LLM model, complete natural language interaction is achieved based on the LLM agent, the user only needs to use the natural language to send out an instruction, and the user experience is improved. The LLM agent can analyze the instruction and simulate the user operation to complete the complex task, and the user does not need to carry out additional click or input operation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence interaction technologies, and in particular, to a Web interaction method based on LLM and natural language. Background Art

[0002] The Web interaction method based on LLM and natural language is an innovative way of operating a Web interface. By integrating large language model technology and natural language processing technology, it allows users to issue instructions in the form of natural language, and the LLM understands and processes these instructions, thereby simulating the user's operation of the Web interface.

[0003] In the existing Web interaction method based on LLM and natural language, user instructions are generally parsed through relatively fixed rules and templates. Although the LLM is used for natural language processing, its flexibility and scalability are limited when understanding and executing complex tasks, and users still need to interact with the system to a certain extent by clicking or inputting.

[0004] To address the problem that in the existing Web interaction method based on LLM and natural language, users still need to interact with the system to a certain extent by clicking or inputting, this Web interaction method based on LLM and natural language trains the LLM model to deeply understand all the functions of the Web system, and realizes full natural language interaction based on the LLM agent. Users only need to issue instructions in natural language, and the LLM agent can parse the instructions and simulate the user's operation to complete complex tasks without the need for users to perform additional click or input operations. Summary of the Invention

[0005] To overcome the problem that in the existing Web interaction method based on LLM and natural language, user instructions are generally parsed through relatively fixed rules and templates. Although the LLM is used for natural language processing, its flexibility and scalability are limited when understanding and executing complex tasks, and users still need to interact with the system to a certain extent by clicking or inputting.

[0006] The technical solution of the present invention is: a Web interaction method based on LLM and natural language, including:

[0007] S11: First, collect and preprocess the data required for the interaction to ensure the accuracy of the data;

[0008] S12: Train the LLM model and optimize and verify the LLM model;

[0009] S13: Design and implement the LLM Agent;

[0010] S14: Develop Agent Tools;

[0011] S15: When the LLM Agent receives a user instruction, it parses the user instruction;

[0012] S16: Execute the operation and call the function;

[0013] S17: Provide feedback and confirmation to the user.

[0014] Preferably, when performing data collection and preprocessing, the following steps are included:

[0015] S21: Collect the function descriptions, operation manuals, and user guide documents of the Web system, clean, organize, and annotate them to construct a training dataset;

[0016] S22: Clean, organize, and annotate the collected data to form a structured training dataset.

[0017] Preferably, when training the LLM model, the following steps are included:

[0018] S31: Select a suitable LLM as the base model;

[0019] S32: Use the preprocessed dataset to train the LLM so that it understands all the functions and operation steps of the Web system;

[0020] S33: Optimize the performance and accuracy of the model through fine-tuning and verification.

[0021] Preferably, when constructing the LLM Agent, the following steps are included:

[0022] S41: Develop a series of Agent Tools, including text input, click, and built-in function calls, for operating the Web system;

[0023] S42: Integrate the trained LLM model into the LLM Agent so that it has the ability to understand natural language instructions.

[0024] Preferably, by introducing the LLM agent as a bridge between the user and the system, it can not only understand the user's natural language instructions but also convert these instructions into specific operation processes, simulate the user's operation of the Web interface, greatly simplify the user's operation process, improve the interaction efficiency, and at the same time reduce the user's learning cost.

[0025] Preferably, when receiving the user instruction, the following steps are included:

[0026] S51: Design a user instruction receiving interface, including a text input box and a voice input interface;

[0027] S52: Receive the instructions issued by the user in natural language.

[0028] Preferably, when parsing the user instructions, the following steps are included:

[0029] S61: After receiving the user instructions, the LLM Agent parses them to understand the user's operation intention;

[0030] S62: Match the parsed user requirements with the functions of the Web system and integrate them into specific operation requirements.

[0031] Preferably, when forming the workflow, the following steps are included:

[0032] S71: According to the user requirements, the LLM Agent plans a series of operation steps to form a workflow;

[0033] S72: Optimize the workflow to ensure the rationality and efficiency of the operations.

[0034] Preferably, when calling and executing the functions, the following steps are included:

[0035] S81: Based on the workflow, the LLM Agent calls the corresponding Agent Tools to execute the operations;

[0036] S82: Simulate the user's operation of the Web interface through the Agent Tools to complete complex tasks.

[0037] Preferably, when providing feedback and confirmation on the user's operations, the following steps are included:

[0038] S91: The Web system provides operation feedback to the user, including reminders of whether the operation is successful or failed;

[0039] S92: Request the user to confirm the operation result when necessary to ensure the accuracy of the operation.

[0040] Preferably, when continuously optimizing and iterating the interaction method, the following aspects are included:

[0041] S1001: Collect the feedback opinions of the user during the use process;

[0042] S1002: Continuously optimize the performance of the LLM model and Agent Tools according to the user feedback;

[0043] S1003: Expand the functions and interaction methods of the Web system according to the user requirements and market changes.

[0044] Advantages of the present invention:

[0045] 1. Compared with traditional Web interaction methods based on LLM and natural language, which generally parse user instructions through relatively fixed rules and templates, although LLM is used for natural language processing, its flexibility and scalability are limited when understanding and executing complex tasks. Users still need to interact with the system to a certain extent by clicking or inputting. This Web interaction method based on LLM and natural language trains the LLM model to deeply understand all the functions of the Web system and realizes full natural language interaction based on the LLM agent. Users only need to issue instructions in natural language, and the LLM agent can parse the instructions and simulate user operations to complete complex tasks without the need for users to perform additional clicking or input operations.

[0046] 2. By introducing the LLM agent as a bridge between users and the system, it can not only understand users' natural language instructions but also convert these instructions into specific operation processes, simulate user operations on the Web interface, greatly simplify the user operation process, improve the interaction efficiency, and reduce the user's learning cost at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It shows a schematic flowchart of a Web interaction method based on LLM and natural language according to the present invention;

[0048] Figure 2 It shows a schematic flowchart of the LLM model training process of a Web interaction method based on LLM and natural language according to the present invention;

[0049] Figure 3 It shows a schematic flowchart of the LLM Agent construction process of a Web interaction method based on LLM and natural language according to the present invention;

[0050] Figure 4 It shows a schematic flowchart of the process from natural language instructions to the system executing corresponding operations of a Web interaction method based on LLM and natural language according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] The present invention will be further described below with reference to the drawings and embodiments.

[0052] Please refer to Figures 1-4 , the present invention provides an embodiment: a Web interaction method based on LLM and natural language, including:

[0053] S11: First, collect and preprocess the data required for interaction to ensure the accuracy of the data;

[0054] S12: Train the LLM model and optimize and verify the LLM model;

[0055] S13: Design and implement the LLM Agent;

[0056] S14: Develop Agent Tools;

[0057] S15: When the LLM Agent receives a user instruction, parse the user instruction;

[0058] S16: Execute operations and call functions;

[0059] S17: Provide feedback and confirmation to the user.

[0060] Preferably, when collecting and preprocessing data, the following steps are included:

[0061] S21: Collect function descriptions, operation manuals, and user guide documents of the Web system, clean, organize, and annotate them to construct a training dataset;

[0062] S22: Clean, organize, and annotate the collected data to form a structured training dataset.

[0063] Preferably, when training the LLM model, the following steps are included:

[0064] S31: Select a suitable LLM as the base model;

[0065] S32: Use the preprocessed dataset to train the LLM so that it understands all the functions and operation steps of the Web system;

[0066] S33: Optimize the performance and accuracy of the model through fine-tuning and verification.

[0067] Preferably, when constructing the LLM Agent, the following steps are included:

[0068] S41: Develop a series of Agent Tools, including text input, click, and built-in function calls, for operating the Web system;

[0069] S42: Integrate the trained LLM model into the LLM Agent so that it has the ability to understand natural language instructions.

[0070] Preferably, by introducing the LLM agent as a bridge between the user and the system, it can not only understand the user's natural language instructions but also convert these instructions into specific operation processes, simulate the user's operation of the Web interface, greatly simplify the user's operation process, improve the interaction efficiency, and at the same time reduce the user's learning cost.

[0071] Preferably, when receiving a user instruction, the following steps are included:

[0072] S51: Design a user instruction receiving interface, including a text input box and a voice input interface;

[0073] S52: Receive the instruction issued by the user in natural language.

[0074] Preferably, when parsing a user instruction, the following steps are included:

[0075] S61: After receiving the user instruction, the LLM Agent parses it to understand the user's operation intention;

[0076] S62: Match the parsed user requirements with the functions of the Web system and integrate them into specific operation requirements.

[0077] Preferably, when forming a workflow, the following steps are included:

[0078] S71: According to the user requirements, the LLM Agent plans a series of operation steps to form a workflow;

[0079] S72: Optimize the workflow to ensure the rationality and efficiency of the operation.

[0080] Preferably, when calling and executing functions, the following steps are included:

[0081] S81: Based on the workflow, the LLM Agent calls the corresponding Agent Tools to execute the operation;

[0082] S82: Simulate the user's operation of the Web interface through Agent Tools to complete complex tasks.

[0083] Preferably, when providing feedback and confirmation on the user's operation, the following steps are included:

[0084] S91: The Web system provides operation feedback to the user, including reminders of whether the operation is successful or failed;

[0085] S92: Request the user to confirm the operation result when necessary to ensure the accuracy of the operation.

[0086] Preferably, when continuously optimizing and iterating the interaction method, the following aspects are included:

[0087] S1001: Collect the feedback from users during the use process;

[0088] S1002: Continuously optimize the performance of the LLM model and Agent Tools according to the user feedback;

[0089] S1003: Expand the functions and interaction methods of the Web system according to user requirements and market changes.

[0090] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A Web interaction method based on LLM and natural language; characterized in that: It includes: S11: First, collect and preprocess the data required for interaction to ensure data accuracy; S12: Train the LLM model and optimize and validate the LLM model; S13: Design and implement the LLM Agent; S14: Develop Agent Tools; S15: When the LLM Agent receives a user instruction, parse the user instruction; S16: Execute operations and call functions; S17: Provide feedback and confirmation to the user.

2. The Web interaction method based on LLM and natural language according to claim 1, wherein: When collecting and preprocessing data, it includes the following steps: S21: Collect the function descriptions, operation manuals, and user guide documents of the Web system, clean, organize, and annotate them to build a training dataset; S22: Clean, organize, and annotate the collected data to form a structured training dataset.

3. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When training the LLM model, it includes the following steps: S31: Select a suitable LLM as the base model; S32: Use the preprocessed dataset to train the LLM so that it understands all the functions and operation steps of the Web system; S33: Optimize the performance and accuracy of the model through fine-tuning and validation.

4. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When building the LLM Agent, it includes the following steps: S41: Develop a series of Agent Tools, including text input, click, and built-in function calls, for operating the Web system; S42: Integrate the trained LLM model into the LLM Agent to enable it to understand natural language instructions.

5. The Web interaction method based on LLM and natural language according to claim 1, characterized in that: When receiving user instructions, it includes the following steps: S51: Design a user instruction receiving interface, including a text input box and a voice input interface; S52: Receive the instructions sent by the user in natural language.

6. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When parsing user instructions, it includes the following steps: S61: After the LLM Agent receives the user instruction, parse it to understand the user's operation intention; S62: Match the parsed user requirements with the functions of the Web system and integrate them into specific operation requirements.

7. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When forming a workflow, it includes the following steps: S71: According to the user requirements, the LLM Agent plans a series of operation steps to form a workflow; S72: Optimize the workflow to ensure the rationality and efficiency of the operations.

8. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When calling and executing functions, it includes the following steps: S81: The LLM Agent calls the corresponding Agent Tools to execute operations based on the workflow; S82: Simulate the user's operation of the Web interface through Agent Tools to complete complex tasks.

9. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When providing feedback and confirmation on user operations, it includes the following steps: S91: The Web system provides operation feedback to the user, including reminders of whether the operation is successful or failed; S92: Request the user to confirm the operation result when necessary to ensure the accuracy of the operation.

10. A Web interaction method based on LLM and natural language according to claim 1, characterized in that: When continuously optimizing and iterating the interaction method, it includes the following aspects: S1001: Collect the feedback from users during the use process; S1002: Continuously optimize the performance of the LLM model and Agent Tools based on user feedback; S1003: Expand the functions and interaction methods of the Web system according to user needs and market changes.

Images