Interaction method and device based on intelligent agent, storage medium and computer equipment

Abstract

The invention discloses an agent-based interaction method and device, a storage medium and computer equipment, and the method comprises the steps: receiving user input information under an interaction process based on a target agent; performing intention recognition on the user input information through the target agent to determine an interaction intention of the user input information; performing cue word optimization on the user input information based on the interaction intention, and generating interaction output information by using a large model based on target input information obtained through optimization; and displaying the interactive output information. According to the invention, through the intention identification and processing capability of the target agent, the intention of the user can be intelligently identified and the corresponding response can be provided, so that the intelligent level of the system is improved and the user can more conveniently obtain the required information or create.

Description

Technical Field

[0001] The present application relates to the field of large model technology, and in particular to an agent-based interaction method, device, storage medium and computer equipment. Background Art

[0002] With the rapid development of artificial intelligence technology, agents are increasingly used in the field of human-computer interaction. Traditional interaction methods often rely on fixed rules and templates to parse user input and generate corresponding responses. However, this method often seems to be unable to cope with complex and changeable user input, and it is difficult to accurately understand the user's true intentions, resulting in poor interaction effects.

[0003] In addition, existing interactive systems usually lack effective optimization and expansion of user input information, resulting in the generated interactive output information being too simple or single and unable to meet the diverse needs of users. For example, when the information input by the user is vague or incomplete, the system may not be able to give an accurate and useful response, thus reducing the user experience.

[0004] In addition, with the continuous advancement of big data and deep learning technology, large models (such as the GPT series, BERT, etc.) have demonstrated powerful capabilities in text generation and understanding. However, how to effectively apply large models to intelligent agent interactions to achieve more accurate and rich interactive outputs is still a problem that needs to be solved. Summary of the invention

[0005] In view of this, the embodiments of the present application provide an agent-based interaction method, device, storage medium and computer equipment, which can more accurately understand user needs by intelligently identifying and optimizing user input information, and use a large model to generate interactive output information that better meets user expectations, thereby improving the user experience.

[0006] According to one aspect of the present application, an agent-based interaction method is provided, the method comprising: Receive user input information in the target agent-based interaction process; Performing intention recognition on the user input information by the target agent to determine the interaction intention of the user input information; Optimizing the prompt words of the user input information based on the interaction intention, and generating interaction output information based on the optimized target input information using a large model; The interactive output information is displayed.

[0007] In the embodiment of the present application, optionally, before receiving user input information in the target agent-based interaction process, the method further includes: Constructing an intelligent agent, embedding an intent recognition algorithm in the intelligent agent, wherein the intent recognition algorithm is used to recognize at least one preset large model interaction intent and at least one preset tool use intent; The target agent is obtained by binding the agent to a knowledge base corresponding to any preset large model interaction intention, and the target agent is associated with the front-end function so that when the front-end function is triggered, an interaction process based on the target agent is entered.

[0008] In an embodiment of the present application, optionally, optimizing the prompt words of the user input information based on the interaction intention, and generating the interaction output information based on the optimized target input information using the large model, includes: If the interaction intention hits the preset big model interaction intention, the prompt word is embedded in the user input information based on the preset prompt word corresponding to the interaction intention to obtain the target input information, and the knowledge base corresponding to the interaction intention is RAG recalled according to the target input information through the big model, and the interaction output information corresponding to the target input information is determined according to the recall result.

[0009] In the embodiment of the present application, optionally, when the interaction intention belongs to the first type of preset big model interaction intention, the interaction output information includes the answer information corresponding to the interaction intention; when the interaction intention belongs to the second type of preset big model interaction intention, the interaction output information includes the creation condition of the pre-intention corresponding to the interaction intention; wherein the first type of preset big model interaction intention is an inquiry intention, and the second type of preset big model interaction intention is a creation intention based on the creation condition; The displaying of the interactive output information includes: In the case where the interaction intention is the first type of preset large model interaction intention, based on the answer points in the answer information and the information links corresponding to the answer points, a first LUI card is generated for display, wherein the first LUI card includes a key point query control corresponding to each answer point, the key point query control displays the answer points and is associated with the corresponding information link, and the key point query control enters the associated information link when it is triggered; In the case where the interaction intention is the second type of preset big model interaction intention, a second LUI card is generated for display based on the creation points in the creation conditions, wherein the second LUI card includes an interaction intention trigger control and a key point filling control corresponding to each creation point, the key point filling control is a fill-in content modifiable control and is initially filled with the corresponding creation point, and the interaction intention trigger control, when triggered, calls the big model to generate the creation content corresponding to the interaction intention based on the filled content of each creation point control in the second LUI card.

[0010] In the embodiment of the present application, optionally, after displaying the interactive output information, the method further includes: Receiving next round of user input information based on the interaction output information as current round of user input information, and performing intent recognition on the current round of input information through the target agent to recognize that the intent corresponding to the current round of input information is a new interaction intent or a supplementary interaction to the previous round of input information; If the current round of user input information is a supplementary interaction to the previous round of input information, then based on the current round of user input information, the context information corresponding to the current round of user input information, and the preset prompt words of the interaction intention, the current round of target input information is constructed, and the knowledge base RAG is recalled and the current round of interaction output information is generated according to the current round of target input information through the big model; If the current round of user input information is a new interaction intention, the original interaction intention is terminated, and the current round of user input information is used as new user input information, and the process returns to the step of optimizing the prompt words of the user input information based on the interaction intention.

[0011] In the embodiment of the present application, optionally, after displaying the interactive output information, the method further includes: When any interaction intention ends, a first model fine-tuning sample is constructed based on the interaction output information corresponding to any interaction intention, the triggering operation and / or modification information of the interaction output information, and any interaction intention, and the large model is fine-tuned based on the first model fine-tuning sample.

[0012] In the embodiment of the present application, optionally, after the target agent performs intent recognition on the user input information to determine the interaction intent of the user input information, the method further includes: If the interaction intention hits the preset tool usage intention, the user input information is optimized based on the interaction intention through the big model, and the target tool corresponding to the interaction intention is called based on the optimized input optimization information.

[0013] In an embodiment of the present application, optionally, optimizing the user input information based on the interaction intent through the large model, and calling the target tool corresponding to the interaction intent based on the optimized input optimization information, includes: Obtaining a target input parameter type of a target tool corresponding to the interaction intention, and optimizing the user input information based on the target input parameter type through a large model to obtain a user input parameter that matches the target input parameter type and corresponds to the user input information; Obtaining a call link of the target tool, and generating a third LUI card for display based on the call link, the user input parameter, and the target input parameter type, wherein the third LUI card includes a use trigger control for the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control, wherein the parameter filling control is a control whose filling content can be modified; In response to a trigger operation on the use trigger control, the target tool is called based on the filled content of each parameter fill control in the third LUI card.

[0014] In the embodiment of the present application, optionally, after calling the target tool, the method further includes: Determine modification information corresponding to the user input parameter according to the filling content of each parameter filling control in the third LUI card and the user input parameter; According to the modification information, the user input information and the interaction intention, a second model fine-tuning sample is constructed, and the large model is fine-tuned based on the second model fine-tuning sample.

[0015] According to another aspect of the present application, there is provided an agent-based interaction device, the device comprising: An information receiving module, used to receive user input information in an interactive process based on a target agent; An intention recognition module, used to perform intention recognition on the user input information through the target agent to determine the interaction intention of the user input information; An information generation module, used to optimize the prompt words of the user input information based on the interaction intention, and generate interaction output information based on the optimized target input information using a large model; The information display module is used to display the interactive output information.

[0016] In the embodiment of the present application, optionally, the device further includes: an agent creation module, which is used to: Constructing an intelligent agent, embedding an intent recognition algorithm in the intelligent agent, wherein the intent recognition algorithm is used to recognize at least one preset large model interaction intent and at least one preset tool use intent; The target agent is obtained by binding the agent to a knowledge base corresponding to any preset large model interaction intention, and the target agent is associated with the front-end function so that when the front-end function is triggered, an interaction process based on the target agent is entered.

[0017] In the embodiment of the present application, optionally, the information generating module is further used to: If the interaction intention hits the preset big model interaction intention, the prompt word is embedded in the user input information based on the preset prompt word corresponding to the interaction intention to obtain the target input information, and the knowledge base corresponding to the interaction intention is RAG recalled according to the target input information through the big model, and the interaction output information corresponding to the target input information is determined according to the recall result.

[0018] In the embodiment of the present application, optionally, when the interaction intention belongs to the first type of preset big model interaction intention, the interaction output information includes the answer information corresponding to the interaction intention; when the interaction intention belongs to the second type of preset big model interaction intention, the interaction output information includes the creation condition of the pre-intention corresponding to the interaction intention; wherein the first type of preset big model interaction intention is an inquiry intention, and the second type of preset big model interaction intention is a creation intention based on the creation condition; The information display module is also used for: In the case where the interaction intention is the first type of preset large model interaction intention, based on the answer points in the answer information and the information links corresponding to the answer points, a first LUI card is generated for display, wherein the first LUI card includes a key point query control corresponding to each answer point, the key point query control displays the answer points and is associated with the corresponding information link, and the key point query control enters the associated information link when it is triggered; In the case where the interaction intention is the second type of preset big model interaction intention, a second LUI card is generated for display based on the creation points in the creation conditions, wherein the second LUI card includes an interaction intention trigger control and a key point filling control corresponding to each creation point, the key point filling control is a fill-in content modifiable control and is initially filled with the corresponding creation point, and the interaction intention trigger control, when triggered, calls the big model to generate the creation content corresponding to the interaction intention based on the filled content of each creation point control in the second LUI card.

[0019] In the embodiment of the present application, optionally, after displaying the interactive output information, the method further includes: The information receiving module is further used to receive the next round of user input information based on the interactive output information as the current round of user input information; The intention recognition module is further used to perform intention recognition on the current round of input information through the target agent to recognize that the intention corresponding to the current round of input information is a new interaction intention or a supplementary interaction to the previous round of input information; The information generation module is further configured to construct the current round target input information based on the current round user input information, the context information corresponding to the current round user input information, and the preset prompt word of the interaction intention if the current round user input information is a supplementary interaction to the previous round input information, and perform knowledge base RAG recall and current round interaction output information generation according to the current round target input information through the large model; If the current round of user input information is a new interaction intention, the original interaction intention is terminated, and the current round of user input information is used as new user input information, and the process returns to the step of optimizing the prompt words of the user input information based on the interaction intention.

[0020] In an embodiment of the present application, optionally, the device also includes: a model fine-tuning module, which is used to construct a first model fine-tuning sample based on the interaction output information corresponding to any interaction intention, the triggering operation and / or modification information of the interaction output information and any interaction intention when ending any interaction intention, and perform large model fine-tuning based on the first model fine-tuning sample.

[0021] In the embodiment of the present application, optionally, the information generation module is further configured to optimize the user input information based on the interaction intention through a large model if the interaction intention hits a preset tool usage intention; The information display module is also used to call the target tool corresponding to the interaction intention based on the optimized input optimization information.

[0022] In an embodiment of the present application, optionally, the information generation module is further used to obtain a target input parameter type of a target tool corresponding to the interaction intention, and optimize the user input information based on the target input parameter type through a large model to obtain a user input parameter that matches the target input parameter type corresponding to the user input information; The information display module is also used to obtain a calling link of the target tool, and generate a third LUI card for display based on the calling link, the user input parameters and the target input parameter type, wherein the third LUI card includes a usage trigger control for the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control, and the parameter filling control is a control whose filling content can be modified; in response to a trigger operation on the usage trigger control, the target tool is called based on the filling content of each parameter filling control in the third LUI card.

[0023] In the embodiment of the present application, optionally, the model fine-tuning module is further used to: Determine modification information corresponding to the user input parameter according to the filling content of each parameter filling control in the third LUI card and the user input parameter; According to the modification information, the user input information and the interaction intention, a second model fine-tuning sample is constructed, and the large model is fine-tuned based on the second model fine-tuning sample.

[0024] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned agent-based interaction method is implemented.

[0025] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned agent-based interaction method when executing the program.

[0026] By means of the above technical solution, an agent-based interaction method, device, storage medium and computer equipment provided by the embodiments of the present application, in an interaction process based on a target agent, after receiving user input information, the target agent first performs intent recognition on the user input information, and then optimizes the prompt words of the user input information based on the recognized interaction intent, thereby using a large model to generate interactive output information for the optimized target input information. The embodiments of the present application can more accurately understand user needs by intelligently identifying and optimizing user input information, and use a large model to generate interactive output information that better meets user expectations, thereby improving user experience.

[0027] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A schematic diagram of a flow chart of an agent-based interaction method provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of a flow chart of another agent-based interaction method provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of a flow chart of another agent-based interaction method provided in an embodiment of the present application is shown; Figure 4A schematic structural diagram of an agent-based interaction device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0029] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0030] In this embodiment, an agent-based interaction method is provided, such as Figure 1 As shown, the method includes: Step 101: Receive user input information in an interactive process based on a target agent.

[0031] Step 102: Use the target agent to perform intent recognition on the user input information to determine the interaction intent of the user input information.

[0032] Step 103: optimize the prompt words of the user input information based on the interaction intention, and generate interaction output information based on the optimized target input information using a large model.

[0033] Step 104: display the interactive output information.

[0034] The embodiment of the present application provides an agent-based interaction method, which aims to optimize and enhance the interaction experience between users and agents through intelligent technology. First, the user can enter the interaction process based on the target agent in a specific way, such as selecting the customer service function provided by the client front end to enter the interaction process of the target agent pre-bound to the customer service function, and specifically can provide a visual interface that allows users to input information in the form of text or voice. Receive information provided by the user through the interface (such as a text input box, a voice input device, etc.). The target agent is an agent used to process user input information. It uses natural language processing (NLP) technology to parse the user input to identify the user's interaction intention, so as to understand the user's needs and subsequently generate appropriate responses. Then, after identifying the user's interaction intention, the original user input information is optimized, specifically, by adding or modifying prompt words to make the input information clearer or conform to a specific format. The optimized target input information is then sent to a large model, and the large model generates interactive output information matching the user's intention based on the target input information. The large model used in the embodiment of the present application can select a generalized large model, or a large model for fine-tuning training based on data samples corresponding to different interaction intentions. This output information can be in the form of text, voice or other forms, and is intended to meet user needs or answer questions. Finally, the generated interactive output information is displayed through a user interface (such as screen display, voice playback, etc.), thereby completing the entire interactive process. The embodiment of the present application achieves accurate understanding and rich output of user input information by introducing the intention recognition capability of the intelligent agent and the text generation capability of the large model, which not only improves the accuracy and diversity of the interaction, but also improves the user experience, and provides new ideas and solutions for the application of intelligent agents in the field of human-computer interaction.

[0035] By applying the technical solution of this embodiment, in the interactive process based on the target agent, after receiving the user input information, the target agent first performs intent recognition on the user input information, and then optimizes the prompt words of the user input information based on the recognized interactive intent, so as to generate interactive output information for the optimized target input information using the large model. The embodiment of the present application can more accurately understand user needs by intelligently identifying and optimizing user input information, and use the large model to generate interactive output information that better meets user expectations, thereby improving user experience.

[0036] In an embodiment of the present application, optionally, before step 101, it also includes: constructing an intelligent agent, embedding an intent recognition algorithm in the intelligent agent, wherein the intent recognition algorithm is used to recognize at least one preset large model interaction intention and at least one preset tool usage intention; binding the intelligent agent with a knowledge base corresponding to any preset large model interaction intention to obtain the target intelligent agent, and associating the target intelligent agent to the front-end function so that when the front-end function is triggered, it enters an interaction process based on the target intelligent agent.

[0037] In the above embodiment, first, an agent framework capable of performing specific tasks is created. This framework is the carrier of agent behavior, which defines how the agent receives information, processes information, and responds. Then, a specially designed intent recognition algorithm is embedded in the agent framework. The function of this algorithm is to parse the information input by the user to identify the user's interaction intention. The intent recognition algorithm here is designed to be able to recognize at least one preset large model interaction intention (such as question and answer, recommendation, etc.) and at least one preset tool usage intention (such as calling API, executing specific functions, etc.). This design enables the agent to flexibly respond to different types of user requests. Next, the agent is bound to the knowledge base corresponding to any preset large model interaction intention. The knowledge base contains data and information related to specific interaction intentions, such as question and answer pairs, recommendation algorithms, etc. By binding the knowledge base, the agent can obtain the key information required to process user requests, thereby generating more accurate and useful replies. Further, the agent is associated with the preset prompt words corresponding to the knowledge base. Finally, the target agent is associated with the front-end function, so that when the user triggers a function in the front-end interface, the corresponding target agent is automatically called and enters the interactive process based on the agent. This design makes the application of the agent more convenient and efficient. In a specific application scenario, the above-mentioned target agent creation process can be configured to be executed through a visual interface, so that technicians can easily create agents. By applying the technical solution of this embodiment, by building a unified agent framework, and embedding the intention recognition algorithm and binding knowledge base therein, the construction process of the agent is simplified. The technician does not need to design the agent from scratch, but can customize and expand on the basis of the existing framework, thereby shortening the development cycle and reducing development costs. In addition, since the embodiment of the present application provides a standardized agent construction process, the technician can create a new agent more quickly. They only need to select the appropriate intention recognition algorithm and knowledge base according to specific needs, and perform necessary configuration and debugging. This design enables the technician to focus more on the core functions and performance optimization of the agent, rather than spending a lot of time on cumbersome construction and integration work.

[0038] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another agent-based interaction method is provided, such as Figure 2 As shown, the method includes: Step 201: Receive user input information in the target agent-based interaction process.

[0039] In this embodiment, in the interactive process based on the target intelligent agent, the information provided by the user through the front-end interface or other input methods is first received. The user can input in the form of text or voice. If it is in the form of voice, the voice needs to be converted into text.

[0040] Step 202: Use the target agent to perform intent recognition on the user input information to determine the interaction intent of the user input information.

[0041] In this embodiment, the target agent uses the embedded intent recognition algorithm to parse the user input information to determine the user's interaction intention. This intention can be one of the preset large model interaction intentions or other types of intentions. In this step, the user's true intention is determined based on the content and context of the user input and the preset intent recognition model.

[0042] Step 203: If the interaction intention hits the preset big model interaction intention, based on the preset prompt word corresponding to the interaction intention, the prompt word is embedded in the user input information to obtain the target input information, and the knowledge base corresponding to the interaction intention is RAG recalled according to the target input information through the big model, and the interaction output information corresponding to the target input information is determined according to the recall result.

[0043] In this embodiment, if the user's interaction intention matches the preset large model interaction intention, the user input information can be embedded with prompt words according to the pre-configured preset prompt words corresponding to the user's interaction intention to obtain the target input information. The preset prompt words are used to guide the large model to generate content according to the knowledge base that matches the user's interaction intention. After obtaining the target input information, the target input information is input into the large model so that the large model performs RAG recall in the knowledge base corresponding to the interaction intention according to the target input information, and generates the interaction output information corresponding to the target input information according to the recall result.

[0044] Among them, when the interaction intention belongs to the first category of preset big model interaction intention, the interaction output information includes the answer information corresponding to the interaction intention; when the interaction intention belongs to the second category of preset big model interaction intention, the interaction output information includes the creation condition of the preceding intention corresponding to the interaction intention; among them, the first category of preset big model interaction intention is the inquiry intention, and the second category of preset big model interaction intention is the creation intention based on the creation condition.

[0045] Different processing methods can be adopted according to the different types of interaction intentions. The first type of preset large model interaction intention (inquiry intention): the user input information is embedded with prompt words according to the preset prompt words corresponding to the interaction intention, and then the large model is used to perform RAG (Retrieval Augmented Generation) recall on the embedded target input information to retrieve relevant information from the knowledge base. The recall result will be used as the interactive output information, including the key points of the answer and the corresponding information link. The second type of preset large model interaction intention (creative intention based on creative conditions): prompt word embedding and RAG recall will also be performed, but the recall result will be used as the creative condition, including the creative points. These creative points will be used to guide users to create or to guide subsequent large model creation. For example, if the creative intention is to make a logo, the corresponding pre-intention is the creative condition for generating a logo, and the creative points generated by the large model are the design points of the logo.

[0046] Step 204: When the interaction intention is the first type of preset large model interaction intention, a first LUI card is generated for display based on the answer points in the answer information and the information links corresponding to the answer points, wherein the first LUI card includes a key point query control corresponding to each answer point, and the key point query control displays the answer points and the associated corresponding information links, and enters the associated information link when the key point query control is triggered.

[0047] In this embodiment, when interactive output information is displayed, it can be displayed in the form of LUI (Lightweight User Interface) cards. Specifically, for the first type of preset large model interaction intention (inquiry intention), a first LUI card is generated based on the answer points and the corresponding information links for display. The card includes a key query control corresponding to each answer point, and the user can view detailed information or enter the associated information link by clicking the control.

[0048] Step 205: When the interaction intention is the second type of preset big model interaction intention, a second LUI card is generated for display based on the creation points in the creation conditions, wherein the second LUI card includes an interaction intention trigger control and a key point filling control corresponding to each creation point, the key point filling control is a control for modifying the filling content and is initially filled with the corresponding creation point, and the interaction intention trigger control, when triggered, calls the big model to generate the creation content corresponding to the interaction intention based on the filling content of each creation point control in the second LUI card.

[0049] In this embodiment, for the second type of preset macromodel interaction intent (creation intent), a second LUI card is generated based on the creation points in the creation conditions for display. The card includes an interaction intent trigger control and a key point fill control corresponding to each creation point. The user can modify the initial creation points in the fill control, and then call the macromodel to generate the creation content by triggering the interaction intent control.

[0050] By applying the technical solution of this embodiment, through the intention recognition and processing capabilities of the target intelligent body, it is possible to intelligently identify the user's intention and provide a corresponding response, thereby improving the intelligence level of the system, and enabling users to more conveniently obtain the required information or create. Through the RAG recall technology, information related to user input can be efficiently retrieved from the knowledge base, improving the accuracy and efficiency of information retrieval, and enabling users to obtain the required information faster. For the second type of preset large model interaction intention (creation intention), a modifiable key point filling control is provided, allowing users to perform personalized creation according to their own needs, which can better meet the personalized needs of users and improve user experience. In addition, through the display method of LUI cards, an intuitive and easy-to-use user interface is provided. Users can view detailed information, modify creation points or trigger interaction intentions by clicking on the controls on the cards. This visual interaction method reduces the user's learning cost and improves the ease of use of the system.

[0051] In an embodiment of the present application, optionally, after displaying the interactive output information, it also includes: receiving the next round of user input information based on the interactive output information as the current round of user input information, and performing intent recognition on the current round of input information through the target intelligent agent to identify that the intention corresponding to the current round of input information is a new interactive intention or a supplementary interaction to the previous round of input information; if the current round of user input information is a supplementary interaction to the previous round of input information, then based on the current round of user input information, the context information corresponding to the current round of user input information, and the preset prompt word of the interactive intention, construct the current round of target input information, and perform knowledge base RAG recall and current round interactive output information generation according to the current round of target input information through a large model; if the current round of user input information is a new interactive intention, then end the original interactive intention, and use the current round of user input information as the new user input information, and return to the step of optimizing the prompt word of the user input information based on the interactive intention.

[0052] In this embodiment, after the interactive output information is displayed, the next round of user input information based on the current interactive output information can continue to be received, and this round of input information is regarded as the current round of user input information for subsequent explanation. The current round of user input information is recognized by the target agent. The result of the recognition may be a new interactive intention or a supplementary interaction to the previous round of input information. If the user input information of the current round is a supplement or correction to the input information of the previous round, the current round of target input information is constructed according to the current round of user input information, its corresponding context information (including the user input information related to the current interactive intention in the previous round and the interactive output information of the large model), and the preset prompt words of the interactive intention. Then, the knowledge base RAG is recalled for the current round of target input information using the large model, and the current round of interactive output information is generated, and the display in the form of LUI cards and the reception of the next round of user input information are continued until the current interactive intention is terminated. Among them, the end of a certain interactive intention can be to exit the interactive process or to start a new interactive intention. If the user input information of the current round represents a new interaction intention, the original interaction intention process will be ended, and the user input information of the current round will be used as the new user input information. If the new interaction intention belongs to the preset large model interaction intention, the prompt word optimization step will be re-entered. If the new interaction intention belongs to the preset tool use intention, the subsequent process based on the identification of the tool use intention will be executed to start a new interaction process. By implementing the above technical solution, intent recognition and information processing are performed according to the continuous input of the user, which improves the interactivity and flexibility of the system and enables users to communicate with the intelligent agent more naturally. When processing supplementary interactions, context information can be used to construct the target input information of the current round, better understand the user's intention, and generate more accurate interaction output information. When the user inputs a new interaction intention, the original interaction intention can be smoothly ended and a new interaction process can be started, ensuring the smoothness of the user experience.

[0053] In an embodiment of the present application, optionally, when any interaction intention is ended, a first model fine-tuning sample is constructed based on the interaction output information corresponding to any interaction intention, the triggering operation and / or modification information of the interaction output information, and any interaction intention, and the large model is fine-tuned based on the first model fine-tuning sample.

[0054] In this embodiment, when any interaction intention is ended, the data of the interaction process can also be used to fine-tune the large model. Specifically, when any interaction intention is ended, a first model fine-tuning sample is constructed according to the interaction output information corresponding to the interaction intention, the user's triggering operation on the interaction output information (such as clicking a link, selecting an option, etc.), the modification information (such as the creation points modified by the user on the LUI card), and the interaction intention itself. The sample contains interaction data and user feedback, which can be used to optimize the performance of the large model. The large model is fine-tuned using the constructed first model fine-tuning sample. The fine-tuning process aims to adjust the parameters of the large model based on the user's actual interaction data and feedback, so that it can more accurately understand the user's intention and generate more interactive output information that meets the user's needs in future interactions. The embodiment of the present application can continuously optimize the performance of the model by fine-tuning the large model using user interaction data and feedback. The fine-tuning process not only takes into account the user's interaction intention and output information, but also takes into account the user's triggering operation and modification information on the interactive output information, which helps to better capture the user's interest in the interactive output information given by the large model, so as to optimize the model performance.

[0055] In an embodiment of the present application, optionally, the interaction intent of the user input information also includes: if the interaction intent hits a preset tool usage intention, the user input information is optimized based on the interaction intent through a large model, and the target tool corresponding to the interaction intent is called based on the optimized input optimization information.

[0056] In this embodiment, tool usage intention refers to the intention of the user to use a specific tool or function to complete a task. After identifying the interaction intention of the user input information, if the interaction intention hits the preset tool usage intention, the user input information can be optimized through the large model. The optimization process includes analyzing the information that needs to be provided when using the tool corresponding to the tool usage intention from the user input information. For example, if the tool usage intention is to schedule an online meeting, the corresponding tool is a meeting reservation tool. Optimizing the user input information includes analyzing the meeting reservation time, meeting reservation type, etc. from the user input information. Based on the input optimization information obtained by optimization, the target tool corresponding to the interaction intention is called. When calling the tool, the optimized input information can be passed to the tool as the input parameter of the tool, and the tool can be waited for to return the result. By implementing the above functions, the user's tool usage intention is automatically identified, and the input information is optimized, thereby simplifying the process of the user calling the tool, reducing the time for the user to manually input and correct information, and improving the efficiency of task execution.

[0057] In the embodiment of the present application, optionally, Figure 3As shown, the user input information is optimized based on the interaction intent by the large model, and the target tool corresponding to the interaction intent is called based on the optimized input optimization information, including: Step 301: Obtain a target input parameter type of a target tool corresponding to the interaction intention, and optimize the user input information based on the target input parameter type through a large model to obtain a user input parameter that matches the target input parameter type and corresponds to the user input information.

[0058] Step 302: obtain the call link of the target tool, and generate a third LUI card for display based on the call link, the user input parameters and the target input parameter type, wherein the third LUI card includes a use trigger control for the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control, and the parameter filling control is a control whose filling content can be modified.

[0059] Step 303: In response to the trigger operation of the use trigger control, the target tool is called based on the filling content of each parameter filling control in the third LUI card.

[0060] In the above embodiment, the target tool corresponding to the user interaction intention is first determined, and the input parameter type required by the tool is obtained. These input parameter types define the data format and structure required for the tool execution. Then, the user input information is optimized using the large model. The optimization process can specifically include syntax correction, information completion, format conversion, etc. to ensure that the user input information can match the target input parameter type. Through this process, the user input parameters that meet the input requirements of the target tool are obtained. Further, the call link of the target tool is obtained to trigger the tool execution. Then, based on the call link, the user input parameters and the target input parameter type, a third LUI card is generated for display. The third LUI card includes a use trigger control of the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control. These parameter filling controls are fillable, allowing the user to modify or adjust the input parameters. When the user clicks or selects the trigger control, the target tool is called according to the fill content of each parameter filling control in the third LUI card. These fill contents represent the input parameters finally confirmed by the user, and they will be passed to the target tool to perform the corresponding task or operation. Through the technical solution of this embodiment, it is possible to automatically identify the user's interaction intention, optimize the user's input information, generate and display a third LUI card containing a trigger control for using the target tool and a parameter filling control, and finally respond to the user's trigger operation to call the target tool. This process improves the interaction efficiency between the user and the system, reduces the difficulty of user operation, and improves the user experience.

[0061] Step 304: determine the modification information corresponding to the user input parameter according to the filling content of each parameter filling control in the third LUI card and the user input parameter; construct a second model fine-tuning sample according to the modification information, the user input information and the interaction intention, and perform large model fine-tuning based on the second model fine-tuning sample.

[0062] In the above embodiment, the user interaction data in the third LUI card can be further used to optimize the performance of the large model. The specific process is as follows: According to the filling content of each parameter filling control in the third LUI card and the user input parameters generated previously, determine the user's modification of the input parameters during actual use. These modifications may include correcting errors, supplementing missing information, adjusting the format, etc. These modification information reflects the user's understanding and expectations of the input parameters and is an important reference for optimizing the large model. Then, based on these modification information, the original user input information (i.e., unoptimized input), and the user's interaction intention, a second model fine-tuning sample is constructed. This sample contains the user's full process data from input to modification to final confirmation, providing rich training materials for the model. Finally, the large model is fine-tuned using the constructed second model fine-tuning sample, so that the model can better understand the user's input intention, predict the user's possible modification behavior, and generate initial input parameters that are more in line with the user's expectations. The embodiment of the present application can more accurately understand the user's input intention and expectations by utilizing the user's modification information during actual use. The constructed second model fine-tuning sample provides rich training data for the model, which helps to improve the model's prediction ability and adaptability.

[0063] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides an interactive device based on an intelligent agent, such as Figure 4 As shown, the device comprises: An information receiving module, used to receive user input information in an interactive process based on a target agent; An intention recognition module, used to perform intention recognition on the user input information through the target agent to determine the interaction intention of the user input information; An information generation module, used to optimize the prompt words of the user input information based on the interaction intention, and generate interaction output information based on the optimized target input information using a large model; The information display module is used to display the interactive output information.

[0064] In the embodiment of the present application, optionally, the device further includes: an agent creation module, which is used to: Constructing an intelligent agent, embedding an intent recognition algorithm in the intelligent agent, wherein the intent recognition algorithm is used to recognize at least one preset large model interaction intent and at least one preset tool use intent; The target agent is obtained by binding the agent to a knowledge base corresponding to any preset large model interaction intention, and the target agent is associated with the front-end function so that when the front-end function is triggered, an interaction process based on the target agent is entered.

[0065] In the embodiment of the present application, optionally, the information generating module is further used to: If the interaction intention hits the preset big model interaction intention, the prompt word is embedded in the user input information based on the preset prompt word corresponding to the interaction intention to obtain the target input information, and the knowledge base corresponding to the interaction intention is RAG recalled according to the target input information through the big model, and the interaction output information corresponding to the target input information is determined according to the recall result.

[0066] In the embodiment of the present application, optionally, when the interaction intention belongs to the first type of preset big model interaction intention, the interaction output information includes the answer information corresponding to the interaction intention; when the interaction intention belongs to the second type of preset big model interaction intention, the interaction output information includes the creation condition of the pre-intention corresponding to the interaction intention; wherein the first type of preset big model interaction intention is an inquiry intention, and the second type of preset big model interaction intention is a creation intention based on the creation condition; The information display module is also used for: In the case where the interaction intention is the first type of preset large model interaction intention, based on the answer points in the answer information and the information links corresponding to the answer points, a first LUI card is generated for display, wherein the first LUI card includes a key point query control corresponding to each answer point, the key point query control displays the answer points and is associated with the corresponding information link, and the key point query control enters the associated information link when it is triggered; In the case where the interaction intention is the second type of preset big model interaction intention, a second LUI card is generated for display based on the creation points in the creation conditions, wherein the second LUI card includes an interaction intention trigger control and a key point filling control corresponding to each creation point, the key point filling control is a fill-in content modifiable control and is initially filled with the corresponding creation point, and the interaction intention trigger control, when triggered, calls the big model to generate the creation content corresponding to the interaction intention based on the filled content of each creation point control in the second LUI card.

[0067] In the embodiment of the present application, optionally, after displaying the interactive output information, the method further includes: The information receiving module is further used to receive the next round of user input information based on the interactive output information as the current round of user input information; The intention recognition module is further used to perform intention recognition on the current round of input information through the target agent to recognize that the intention corresponding to the current round of input information is a new interaction intention or a supplementary interaction to the previous round of input information; The information generation module is further configured to construct the current round target input information based on the current round user input information, the context information corresponding to the current round user input information, and the preset prompt word of the interaction intention if the current round user input information is a supplementary interaction to the previous round input information, and perform knowledge base RAG recall and current round interaction output information generation according to the current round target input information through the large model; If the current round of user input information is a new interaction intention, the original interaction intention is terminated, and the current round of user input information is used as new user input information, and the process returns to the step of optimizing the prompt words of the user input information based on the interaction intention.

[0068] In an embodiment of the present application, optionally, the device also includes: a model fine-tuning module, which is used to construct a first model fine-tuning sample based on the interaction output information corresponding to any interaction intention, the triggering operation and / or modification information of the interaction output information and any interaction intention when ending any interaction intention, and perform large model fine-tuning based on the first model fine-tuning sample.

[0069] In the embodiment of the present application, optionally, the information generation module is further configured to optimize the user input information based on the interaction intention through a large model if the interaction intention hits a preset tool usage intention; The information display module is also used to call the target tool corresponding to the interaction intention based on the optimized input optimization information.

[0070] In an embodiment of the present application, optionally, the information generation module is further used to obtain a target input parameter type of a target tool corresponding to the interaction intention, and optimize the user input information based on the target input parameter type through a large model to obtain a user input parameter that matches the target input parameter type corresponding to the user input information; The information display module is also used to obtain a calling link of the target tool, and generate a third LUI card for display based on the calling link, the user input parameters and the target input parameter type, wherein the third LUI card includes a usage trigger control for the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control, and the parameter filling control is a control whose filling content can be modified; in response to a trigger operation on the usage trigger control, the target tool is called based on the filling content of each parameter filling control in the third LUI card.

[0071] In the embodiment of the present application, optionally, the model fine-tuning module is further used to: Determine modification information corresponding to the user input parameter according to the filling content of each parameter filling control in the third LUI card and the user input parameter; According to the modification information, the user input information and the interaction intention, a second model fine-tuning sample is constructed, and the large model is fine-tuned based on the second model fine-tuning sample.

[0072] It should be noted that for other corresponding descriptions of the functional units involved in the agent-based interactive device provided in the embodiment of the present application, reference can be made to Figures 1 to 3 The corresponding description in the method will not be repeated here.

[0073] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0074] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0075] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0076] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0078] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0079] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An agent-based interaction method, characterized in that: The method comprises: Receive user input information in the target agent-based interaction process; Performing intention recognition on the user input information by the target agent to determine the interaction intention of the user input information; The user input information is optimized for prompt words based on the interaction intention, and the interaction output information is generated based on the optimized target input information using the big model; wherein, when the interaction intention belongs to the inquiry intention, the interaction output information includes the answer information corresponding to the interaction intention; when the interaction intention belongs to the creation intention based on the creation condition, the interaction output information includes the creation condition of the pre-intention corresponding to the interaction intention; The interactive output information is displayed.

2. The method according to claim 1, characterized in that In the target agent-based interaction process, before receiving user input information, the method further includes: Constructing an intelligent agent, embedding an intent recognition algorithm in the intelligent agent, wherein the intent recognition algorithm is used to recognize at least one preset large model interaction intent and at least one preset tool use intent; The target agent is obtained by binding the agent to a knowledge base corresponding to any preset large model interaction intention, and the target agent is associated with the front-end function so that when the front-end function is triggered, an interaction process based on the target agent is entered.

3. The method according to claim 2, characterized in that The step of optimizing the prompt words of the user input information based on the interaction intention, and generating the interaction output information based on the optimized target input information using the large model, includes: If the interaction intention hits the preset big model interaction intention, the prompt word is embedded in the user input information based on the preset prompt word corresponding to the interaction intention to obtain the target input information, and the knowledge base corresponding to the interaction intention is RAG recalled according to the target input information through the big model, and the interaction output information corresponding to the target input information is determined according to the recall result.

4. The method according to claim 2, characterized in that: in, The first type of preset large model interaction intention is inquiry intention, and the second type of preset large model interaction intention is creation intention based on creation conditions; The displaying of the interactive output information includes: In the case where the interaction intention is the first type of preset large model interaction intention, based on the answer points in the answer information and the information links corresponding to the answer points, a first LUI card is generated for display, wherein the first LUI card includes a key point query control corresponding to each answer point, the key point query control displays the answer points and is associated with the corresponding information link, and the key point query control enters the associated information link when it is triggered; In the case where the interaction intention is the second type of preset big model interaction intention, a second LUI card is generated for display based on the creation points in the creation conditions, wherein the second LUI card includes an interaction intention trigger control and a key point filling control corresponding to each creation point, the key point filling control is a fill-in content modifiable control and is initially filled with the corresponding creation point, and the interaction intention trigger control, when triggered, calls the big model to generate the creation content corresponding to the interaction intention based on the filled content of each creation point control in the second LUI card.

5. The method according to claim 4, characterized in that After displaying the interactive output information, the method further includes: Receiving next round of user input information based on the interaction output information as current round of user input information, and performing intent recognition on the current round of input information through the target agent to recognize that the intent corresponding to the current round of input information is a new interaction intent or a supplementary interaction to the previous round of input information; If the current round of user input information is a supplementary interaction to the previous round of input information, then based on the current round of user input information, the context information corresponding to the current round of user input information, and the preset prompt words of the interaction intention, the current round of target input information is constructed, and the knowledge base RAG is recalled and the current round of interaction output information is generated according to the current round of target input information through the big model; If the current round of user input information is a new interaction intention, the original interaction intention is terminated, and the current round of user input information is used as new user input information, and the process returns to the step of optimizing the prompt words of the user input information based on the interaction intention.

6. The method according to claim 4, characterized in that After displaying the interactive output information, the method further includes: When any interaction intention ends, a first model fine-tuning sample is constructed based on the interaction output information corresponding to any interaction intention, the triggering operation and / or modification information of the interaction output information, and any interaction intention, and the large model is fine-tuned based on the first model fine-tuning sample.

7. The method according to any one of claims 1 to 6, characterized in that After the target agent performs intent recognition on the user input information to determine the interaction intent of the user input information, the method further includes: If the interaction intention hits the preset tool usage intention, the user input information is optimized based on the interaction intention through the big model, and the target tool corresponding to the interaction intention is called based on the optimized input optimization information.

8. The method according to claim 7, characterized in that The optimizing the user input information based on the interaction intent by using the large model, and calling the target tool corresponding to the interaction intent based on the optimized input optimization information, includes: Obtaining a target input parameter type of a target tool corresponding to the interaction intention, and optimizing the user input information based on the target input parameter type through a large model to obtain a user input parameter that matches the target input parameter type and corresponds to the user input information; Obtaining a call link of the target tool, and generating a third LUI card for display based on the call link, the user input parameter, and the target input parameter type, wherein the third LUI card includes a use trigger control for the target tool, a parameter filling control corresponding to each target input parameter type, and a user input parameter filled in the corresponding parameter filling control, wherein the parameter filling control is a control whose filling content can be modified; In response to a trigger operation on the use trigger control, the target tool is called based on the filled content of each parameter fill control in the third LUI card.

9. The method according to claim 8, characterized in that After calling the target tool, the method further includes: Determine modification information corresponding to the user input parameter according to the filling content of each parameter filling control in the third LUI card and the user input parameter; According to the modification information, the user input information and the interaction intention, a second model fine-tuning sample is constructed, and the large model is fine-tuned based on the second model fine-tuning sample.

10. An agent-based interactive device, characterized in that: The device comprises: An information receiving module, used to receive user input information in an interactive process based on a target agent; An intention recognition module, used to perform intention recognition on the user input information through the target agent to determine the interaction intention of the user input information; An information generation module, configured to optimize the prompt words of the user input information based on the interaction intention, and generate interaction output information based on the optimized target input information using a large model; wherein, when the interaction intention belongs to an inquiry intention, the interaction output information includes answer information corresponding to the interaction intention; and when the interaction intention belongs to a creation intention based on a creation condition, the interaction output information includes the creation condition of the pre-intention corresponding to the interaction intention; The information display module is used to display the interactive output information.

11. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

12. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images