Interaction method based on large language model, electronic equipment, storage medium and computer program product

Abstract

The embodiment of the invention provides an interaction method based on a large language model, electronic equipment, a storage medium and a computer program product. In the embodiment of the invention, the user can interact with the service system in a natural language mode, and the interaction process between the user and the service system is intelligently transformed by means of the powerful reasoning ability of the large language model; the big language model is combined with the tool configuration information of the multiple tools and the input information of the user to deduce tool parameters and a calling sequence of at least one tool needed for processing the input information. According to the method, the tools are sequentially called according to the calling sequence, before the tools are called, the tool parameters are automatically verified, and under the condition that verification is not successful, a user is supported to intervene in the tool parameters, so that the tool calling reliability is guaranteed, and the personalized requirements of the user are supported. In this way, the service system better meets the personalized requirements of small people, and the intelligence, reliability and operation efficiency of the service system are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an interaction method, electronic device, storage medium and computer program product based on a large language model. Background Technology

[0002] Business systems typically focus on standardized, high-volume group needs. However, personalized, niche needs are often difficult to support effectively under current software development models due to the smaller number of users and higher development costs. Using existing standardized software development models often requires multiple steps or switching between multiple pages, significantly increasing both maintenance and development costs and reducing the operational efficiency of the business system. Summary of the Invention

[0003] This application provides an interaction method, electronic device, storage medium, and computer program product based on a large language model, which enables business systems to better meet the personalized needs of niche groups.

[0004] This application provides an interaction method based on a large language model, applied to a business system. The method includes: acquiring user input information in natural language form; determining, based on the input information, tool configuration information of multiple tools, and the large language model, the tool parameters and calling order of at least one target tool required to process the input information; sequentially validating the tool parameters of the target tools according to the calling order of the at least one target tool; and outputting the tool parameters and parameter intervention guidance information of the target tool if the tool parameters fail the validation; modifying the tool parameters of the target tool in response to the user's input based on the parameter intervention guidance information, obtaining the updated tool parameters of the target tool; calling the target tool according to the updated tool parameters and obtaining the response result returned by the target tool; and generating and outputting response information corresponding to the input information based on the response results of the at least one target tool after all at least one target tool has been called.

[0005] This application also provides an interaction method based on a large language model, applied to a production and operation business system in a converged media platform. The method includes: responding to natural language input information sent by a user from a user terminal; determining the tool parameters and calling order of at least one target tool to be called to process the input information based on the input information, tool configuration information of multiple tools, and a large language model; sequentially verifying the tool parameters of the target tools according to the calling order of at least one target tool, and outputting the tool parameters and parameter intervention guidance information of the target tool if the tool parameters fail verification; modifying the tool parameters of the target tool in response to the user's input based on the parameter intervention guidance information, obtaining the updated tool parameters of the target tool; calling the target tool according to the updated tool parameters and obtaining the response result returned by the target tool; generating reply information corresponding to the input information based on the response results of at least one target tool after all at least one target tool has been called; and sending the reply information to the user terminal.

[0006] This application also provides an electronic device, including: a memory and a processor; the memory for storing a computer program; and the processor coupled to the memory for executing the computer program to perform steps of an interactive method based on a large language model.

[0007] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the processor to implement steps in an interactive method based on a large language model.

[0008] This application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, enable the processor to implement steps in an interactive method based on a large language model.

[0009] In this embodiment, users can interact with the business system via natural language. Leveraging the powerful reasoning capabilities of a large language model, the interaction process between users and the business system is intelligently transformed. The large language model combines the tool configuration information of multiple tools with the user's input information to infer the tool parameters and invocation order of at least one tool required to process the input information. Tools are invoked sequentially according to the invocation order. Before invoking a tool, its parameters are automatically validated. If validation fails, users can intervene in the tool parameters, thereby ensuring the reliability of tool invocation and supporting personalized user needs. This improves the user's interaction experience with the business system, enabling the system to more easily adapt to flexible and changing user needs, better meet personalized niche demands, and enhance the intelligence, reliability, and operational efficiency of the business system. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0011] Figure 1 This is an example application scenario diagram;

[0012] Figure 2 A flowchart illustrating an interaction method based on a large language model provided in an embodiment of this application;

[0013] Figure 3 A flowchart illustrating another interaction method based on a large language model provided in an embodiment of this application;

[0014] Figure 4 A flowchart illustrating the interaction process between users and the production and operation business systems within the converged media platform;

[0015] Figure 5 A flowchart illustrating another interaction method based on a large language model provided in an embodiment of this application;

[0016] Figure 6 A schematic diagram of the structure of an interactive device based on a large language model provided in an embodiment of this application;

[0017] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the access relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone, where A and B can be singular or plural. In the textual description of this application, the character " / " generally indicates that the preceding and following associated objects have an "or" relationship. Furthermore, in the embodiments of this application, "first," "second," "third," etc., are only used to distinguish the content of different objects and have no other special meaning.

[0020] First, some terms used in the embodiments of this application will be introduced:

[0021] A converged media platform is a comprehensive content publishing and interaction platform that integrates multiple media formats (such as text, images, audio, and video). By applying the "large language model," converged media platforms can not only improve content quality and production efficiency, but also enhance user experience and interactivity, thereby improving the overall competitiveness of the converged media platform.

[0022] AI Agent: A software or hardware entity that has the ability to autonomously perceive its environment, make decisions, and execute actions, with the aim of completing specific tasks or achieving goals in the simulated or real world.

[0023] Tools: These are computer programs or sets of programs that provide a specific and effective means of accomplishing particular software development, maintenance, management, or other tasks related to the software lifecycle. These tools are typically designed to improve the efficiency, accuracy, and quality of the software engineering process.

[0024] API (Application Programming Interface): A standard way for software to interact. It defines a set of rules and conventions that allow different software applications to communicate, exchange data, or perform specific functions.

[0025] Large Language Models (LLMs), also known as large-scale language models, refer to a class of natural language processing models with an extremely large number of parameters. LLMs are typically based on deep learning architectures, especially the Transformer architecture, and learn the complex structure of language and rich contextual information through pre-training on massive amounts of text data. The Transformer architecture addresses the bottleneck problem of traditional neural network models when processing long sequences by introducing a self-attention mechanism, and its highly parallelizable nature greatly improves training efficiency. The Transformer architecture includes an encoder or a decoder.

[0026] Inference is the process of using a trained model to process new data and produce output.

[0027] Vector retrieval: A text retrieval method based on the Vector Space Model (VSM).

[0028] This application provides an interaction method, electronic device, storage medium, and computer program product based on a large language model. In this application, users can interact with a business system using natural language. The powerful reasoning capabilities of the large language model intelligently transform the interaction process between the user and the business system. The large language model combines the tool configuration information of multiple tools with the user's input information to infer the tool parameters and calling order of at least one tool required to process the input information. Tools are called sequentially according to the calling order. Before calling a tool, the tool parameters are automatically validated. If validation fails, the user can intervene in the tool parameters, thereby ensuring the reliability of tool calls and meeting the user's personalized needs. This improves the user's interaction experience with the business system, enabling the business system to more easily adapt to flexible and changing user needs, better meet personalized niche needs, and improve the intelligence, reliability, and operational efficiency of the business system.

[0029] The technical solutions of this application and how they solve the aforementioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The technical solutions provided by each embodiment of this application are described in detail below with reference to the accompanying drawings.

[0030] Figure 1 This is a diagram illustrating an exemplary application scenario. See also... Figure 1 The converged media platform includes a production and operation business system. This system is the software used within the platform for content production, editing, management, and distribution, ensuring the platform can efficiently produce, manage, and distribute high-quality content.

[0031] In practical applications, by combining multi-agent mechanisms and powerful large language models, production and operations business systems can become more intelligent, efficient, and flexible, better responding to ever-changing business needs. Based on the multi-agent mechanism, the production and operations business system provides multiple agents, each capable of performing specific tasks and collaborating with other agents. Within the production and operations business system, these agents can be responsible for different business processes or functional modules. Large language models provide agents with natural language processing capabilities, enabling them to better understand instructions, generate responses, or engage in complex dialogues.

[0032] In practical applications, when users have various needs such as content production, editing, management, and publishing, please refer to [the relevant documentation / reference]. Figure 1As shown in ① and ②, users send input information reflecting their needs to the converged media platform through user terminals (such as mobile phones, computers, etc.). This input information can be in various forms, including text, voice, and images. After receiving the input information, the production and operation business system within the converged media platform processes it using a multi-agent mechanism and a powerful large language model. During processing, user participation may be required, such as confirming tool parameters or providing additional input. Finally, the production and operation business system generates a response to the user's input and returns it to the user terminal. This response can be in various forms, such as news releases, articles, videos, and audio, depending on the user's needs and the capabilities of the converged media platform.

[0033] It should be noted that, Figure 1 The application scenario shown is merely an exemplary one, and the embodiments of this application do not limit the application scenario.

[0034] Figure 2 A flowchart illustrating an interaction method based on a large language model, provided as an embodiment of this application. See also... Figure 2 The method may include the following steps:

[0035] 201. Obtain the user's input information in natural language form.

[0036] The method provided in this application can be applied to various business systems, including but not limited to: production and operation business systems, e-commerce business systems, short video business systems, etc. For example, e-commerce business systems provide services such as product sales and service transactions; short video business systems provide services such as the creation, editing, sharing, and promotion of short video content, etc.

[0037] In practical applications, when users interact with business systems, they can input information via text or voice, but are not limited to these methods. For example, users can input text using devices such as keyboards or touchscreens; the input text is in the form of natural language. As another example, voice input allows users to input information by speaking. Typically, a microphone captures the voice information, which is then converted into text using speech recognition technology; the converted text is the input information in the form of natural language.

[0038] In practical applications, user input information represents the user's intent. For example, user input information might include "Please create a special feature on international travel for me," "Please generate an abstract for article 12345 for me," or "Please upload article 45987 to the sports channel for me," etc.

[0039] Optionally, before using a large language model for inference, the input data can be subjected to security checks to prevent malicious data injection, improve data quality, and protect the security of business systems. Security checks include, but are not limited to, sensitive word checks and permission checks. Sensitive word checks refer to checking whether the input data contains predefined sensitive words or phrases. Permission checks refer to checking whether a user has the appropriate permissions before performing an operation. This helps protect system resources from unauthorized access or modification.

[0040] 202. Based on the input information, the tool configuration information of multiple tools, and the large language model, determine the tool parameters and calling order of at least one target tool required to process the input information.

[0041] In practical applications, business systems provide multiple tools, each corresponding to an API interface. The tools provided by different business systems vary, depending on the specific circumstances. For example, the tools provided by a business system might include: intelligent summarization tools, intelligent title tools, manuscript upload tools, manuscript removal tools, manuscript search tools, and tools for adding special topics.

[0042] Among them, the intelligent summary tool is used to intelligently generate article summaries by analyzing given text descriptions; the intelligent title tool is used to intelligently generate article titles by analyzing given text; the manuscript online tool is used to set the status of a manuscript to online based on a given manuscript ID; the offline manuscript tool is used to set the status of a manuscript to offline based on a given manuscript ID; the manuscript query tool is used to retrieve manuscript content based on a given manuscript ID; and the add topic tool is used to create topics based on a given topic title.

[0043] In practical applications, tool configuration information typically includes a series of settings related to tool usage. These settings help ensure the tool operates as expected and can be adjusted according to different needs. A unified abstract business interface can be used to configure the tool configuration information, making tool parameters configurable.

[0044] The tool configuration information includes, but is not limited to: the tool's unique identifier, the tool's API access address, a status flag indicating whether the tool is enabled, the tool's name, request method, tool's functional description information, and various tool parameters such as request parameter definitions. Among them, request parameter definitions are used to define request parameters, which typically include request parameter names, parameter types, whether they are required, descriptions of parameter values, etc.

[0045] For example, the tool configuration information for the offline manuscript tool is as follows:

[0046] {"id": "4", / / Unique identifier for the tool, here it is 4}

[0047] "url": "http: / / 192.168.0.8:18092 / op / api / content / v1 / offline", / / API access address of the tool;

[0048] "enable": true, / / Whether the tool is enabled. Setting it to true here means enabled; setting it to false means disabled.

[0049] "name": "Offline Manuscript", / / The name of the tool, set to "Offline Manuscript" here.

[0050] "is_confirm": false, / / Whether a secondary confirmation operation is required. Setting it to false here means no confirmation is needed; setting it to true means confirmation is required.

[0051] "description": "The tool provides a way to set the status of a manuscript to offline based on a given manuscript ID." / / Tool function description information

[0052] "method": "post", / / Request method

[0053] "content_type": "application / json", / / The type of request content, here it is application / json, indicating that the request body is in JSON format.

[0054] "request_params": [ / / Request parameter definition]

[0055] {"key": "itemId", / / key name of the request parameter}

[0056] "name": "Manuscript ID", / / Name of the request parameter

[0057] "option": false, / / Whether the request parameter is optional. Setting it to false here indicates that the parameter is required; setting it to true indicates that the request parameter is optional.

[0058] "description": "Manuscript ID", / / Description information of the request parameter

[0059] "model_description": "The tool provides a way to set the status of an offline article to offline based on a given article ID or item_id." / / Model description in the request parameters; the tool description is repeated here. ]

[0061] }

[0062] In practical applications, the tool's configuration information can be used to obtain some necessary tool parameters for invoking the tool, such as the tool's URL and request method. Typically, the tool's configuration information defines the request parameters, but their values ​​are not specified. Therefore, these essential tool parameters need to be determined before invoking the tool. Request parameters are information required when calling an API interface, and they are crucial for the correct execution of the tool. Specifically, an interface request is constructed based on the tool's URL, request method, and request parameters, and then sent to the tool's URL to invoke the corresponding API interface, i.e., to invoke the tool. For more information on tool invocation, please refer to the section on interface invocation in related technologies; it will not be elaborated upon here.

[0063] In this embodiment, leveraging the powerful reasoning capabilities of the large language model, the large language model, based on the user's input information and the tool configuration information of multiple tools, infers the tool parameters and calling order of at least one target tool required to be invoked based on the input information.

[0064] In some optional embodiments, the method for determining the tool parameters and calling order of at least one target tool to be called for processing the input information, based on the input information, the tool configuration information of multiple tools, and the large language model, is as follows: the input information and the tool configuration information of multiple tools are filled into the prompt word template to obtain the target prompt word; the target prompt word is input into the large language model for reasoning to determine the tool parameters and calling order of at least one target tool to be called for processing the input information.

[0065] The prompt template can be flexibly configured as needed, and includes multiple positions to be filled. For example, the prompt template could be: "Your task is to determine the tool parameters and invocation order of at least one target tool to be invoked to process the input information based on the user's input information and the tool configuration information of multiple tools. 1. Input information: [To be filled]. 2. Tool configuration information: [To be filled]", where [To be filled] is the position to be filled.

[0066] In some optional embodiments, the method for determining the tool parameters and calling order of at least one target tool to be called to process the input information, based on the input information, the tool configuration information of multiple tools, and the large language model, is as follows: from multiple tools, select at least one candidate tool whose tool configuration information matches the input information; fill the input information and the tool configuration information of at least one candidate tool into the prompt word template to obtain the target prompt word; input the target prompt word into the large language model for reasoning to determine the tool parameters and calling order of at least one target tool to be called to process the input information.

[0067] Understandably, before inference in a large language model, candidate tools are selected from multiple tools, and inference is then performed based on the input information and the tool configuration information of the candidate tools. This reduces the number of candidate tools selected, and the selected candidate tools are more in line with the user's intent. In this way, when the large language model performs inference on some initially selected candidate tools, it can improve inference efficiency and accuracy.

[0068] In practical applications, when selecting candidate tools from multiple tools, the similarity between the input information and the tool configuration information of each tool can be calculated. The top N tools with the highest similarity are then selected as candidate tools, where N is a positive integer. Optionally, since the functional description information of a tool can effectively characterize it, the similarity between the input information and the functional description information of each tool can be calculated instead of the tool configuration information, thereby improving the efficiency and accuracy of similarity calculation.

[0069] Further optionally, selecting at least one candidate tool from multiple tools whose tool configuration information matches the input information can be achieved by: performing a vector retrieval in a vector database based on the vector representation of the input information, the vector database including vector representations of tool configuration information of multiple tools; and determining at least one candidate tool whose tool configuration information matches the input information based on the vector retrieval results.

[0070] Understandably, vector retrieval can more efficiently and accurately filter candidate tools from multiple tools, thereby improving overall reasoning efficiency and accuracy.

[0071] Specifically, the tool configuration information of multiple tools is pre-vectorized to obtain vector representations of the tool configuration information. A vector database is then constructed based on these vector representations. Preferably, the vector representation of the tool configuration information can be a vector representation of the tool's functional description information. Thus, when determining candidate tools, the input information is first vectorized to obtain its vector representation. Then, the vector database is searched for vector representations of tool configuration information similar to the input information's vector representation; the tools corresponding to these similar vector representations are considered candidate tools.

[0072] In practical applications, a business system can provide one or more intelligent agents, without limitation. Further optionally, the method for determining the tool parameters and invocation order of at least one target tool required to process the input information, based on the input information, the tool configuration information of multiple tools, and the large language model, is as follows: Intent recognition is performed on the input information to obtain its intent information; a target intelligent agent matching the intent information is determined from among multiple intelligent agents; if the target intelligent agent belongs to the business domain, then the target intelligent agent determines the tool parameters and invocation order of at least one target tool required to process the input information based on the input information, the tool configuration information of the multiple tools corresponding to the target intelligent agent, and the large language model.

[0073] In practical applications, multiple intelligent agents can include both business domain intelligent agents and non-business domain intelligent agents. Business domain intelligent agents can be understood as intelligent agents related to actual business, that is, intelligent agents designed specifically for a particular business domain, such as intelligent agents related to converged media, medical intelligent agents, and retail intelligent agents, etc. Non-business domain intelligent agents can be understood as intelligent agents not related to actual business, that is, general-purpose intelligent agents that do not possess specific domain knowledge or professional skills, such as pure question-answering intelligent agents (also known as question-answering domain intelligent agents) and customer service question-answering intelligent agents that provide customer service.

[0074] In practical applications, the target agent is determined by matching the intent information of the input information with the functional description information of the agent. The functional description information of the agent reflects its function. For example, the agent's functional description information might be "providing inventory query service" or "providing weather query service." The intent information of the input information could be "query weather forecast" or "query inventory information," etc. Therefore, based on the semantic matching degree between the intent information of the input information and the functional description information of the agent, the target agent that successfully matches the intent information of the input information can be determined. For another example, each agent's information includes functional description information and supported intent types. The intent information of the input information can be matched with the supported intent types in the agent's information. The functional description information corresponding to the successfully matched intent type also successfully matches the intent information of the input information, and the corresponding agent is the target agent. Yet another example is calculating the similarity between keywords in the intent information of the input information and keywords in the functional description information of the agent, and determining the agent with the highest similarity is the target agent.

[0075] If the target agent is a non-business domain agent, the user's input information can be directly input into the non-business domain agent. The non-business domain agent then calls the large language model to process the input information, obtain the response information, and send the response information back to the user. If the target agent is a business domain agent, the tools corresponding to different business domain agents may differ. The target agent determines the tool parameters and calling order of at least one target tool to process the input information based on the input information, the tool configuration information of multiple tools corresponding to the target agent, and the large language model. It is understandable that finding the appropriate agent to call the large language model for inference based on the intent information of the input information better distinguishes the user's execution intent and improves inference accuracy.

[0076] In practical applications, various intent recognition methods can be employed. Optionally, to improve the accuracy of intent recognition, a pre-trained intent recognition model can be used to identify the user's input intent. During the model training phase, a labeled training dataset is used to train the intent recognition model. During the model prediction phase, the user's input information is fed into the intent recognition model, which then outputs the predicted intent information.

[0077] Further optionally, in order to prevent malicious data injection, improve data quality, and protect the security of business systems, before performing intent recognition on the input information to obtain the intent information of the input information, the following steps can be performed: performing security verification on the input information; and if the input information passes the security verification, performing intent recognition on the input information to obtain the intent information of the input information.

[0078] 203. Verify the tool parameters of the target tools in the order in which at least one target tool is called.

[0079] 204. If the tool parameters of the target tool fail the validation, output the tool parameters of the target tool and parameter intervention guidance information.

[0080] 205. In response to the user's operation of inputting guidance information based on parameters, modify the tool parameters of the target tool to obtain the updated tool parameters of the target tool.

[0081] 206. Call the target tool according to the updated tool parameters and obtain the response result returned by the target tool.

[0082] In this embodiment, after the large language model infers the tool parameters and calling order of at least one target tool required to process the input information, each target tool is called in sequence according to the calling order of at least one target tool. That is, after the previous target tool is successfully called, the next target tool is called, until all target tools are called.

[0083] Specifically, during the invocation of each target tool, the tool parameters inferred by the large language model are first validated. If the validation is successful, the target tool is invoked. If the validation fails, manual intervention is required to modify or supplement the tool parameters inferred by the large language model to ensure the reliability of the target tool invocation and to ensure that the tool can be invoked correctly.

[0084] Optionally, when validating the tool parameters of the target tool inferred from the large language model, the following aspects can be validated:

[0085] 1. Parameter Completeness: Check that all necessary tool parameters are provided. For example, if a tool requires a file path parameter, ensure that this parameter has been correctly specified.

[0086] 2. Parameter Format: Verify that the parameter format meets expectations. For example, date parameters should follow a specific format (such as YYYY-MM-DD), and numeric parameters should be valid numbers, etc.

[0087] 3. Parameter value validity: Confirm that the parameter value is within a reasonable range. For example, if a tool accepts an integer parameter representing a page number, then this parameter should be a positive integer and not exceed the maximum page number.

[0088] 4. Parameter type consistency: Ensure that the parameter type matches the type required by the tool. For example, if a tool requires a string parameter, but an integer parameter is passed, a conversion or error should be generated.

[0089] In this embodiment, if the tool parameter validation of the target tool fails, the tool parameters and parameter intervention guidance information can be output. The parameter intervention guidance information guides the user to intervene in the tool parameters of the target tool; for example, the parameter intervention guidance information could be "The parameter value of the request parameter 'name' is missing or the format is incorrect." Guided by the parameter intervention guidance information, the user supplements the parameter value of the request parameter 'name' or modifies its format to obtain the updated tool parameters of the target tool. The intelligent agent of the business system calls the target tool based on the updated tool parameters and obtains the response result returned by the target tool.

[0090] Further optionally, in order to reduce errors caused by human intervention, the tool parameters of the target tool and parameter intervention guidance information can be output on the user interface. The user interface also includes a variety of interactive controls, such as, but not limited to, radio buttons, checkboxes, input boxes, and confirmation boxes.

[0091] For any tool parameter, when a user triggers a radio button, a pop-up appears displaying multiple selectable parameter values ​​for the user to choose from. The selected value is then used as the updated parameter value for the corresponding tool. Similarly, when a user triggers a checkbox, a pop-up appears displaying multiple selectable parameter values ​​for the user to choose from. The selected values ​​are then used as the updated parameter values ​​for the corresponding tool. When the user enters a new parameter value in the input box, this new value is used as the updated parameter value for the corresponding tool. After the user has finished making changes, they can trigger a confirmation box to submit the changes, and the business system will then update the tool parameter value accordingly.

[0092] In some optional embodiments, human intervention may lead to errors. To ensure the reliability of calling the tool, the updated tool parameters of the target tool are verified before calling the target tool based on the updated tool parameters. If the updated tool parameters fail the verification and the number of parameter interventions has not yet reached the intervention threshold, the steps of outputting the tool parameters and parameter intervention guidance information of the target tool and subsequent steps are executed again until the updated tool parameters pass the verification or the number of parameter interventions reaches the intervention threshold.

[0093] Specifically, users can be allowed to perform multiple manual interventions within a set threshold. This means that after obtaining the tool parameters of the target tool for large language model inference, upon detecting a user's input based on parameter-guided intervention information, a parameter intervention is confirmed, and the intervention count is incremented. If the intervention count has not reached the threshold, the user is allowed to intervene again. If the intervention count has reached the threshold, intervention is prohibited. The intervention threshold can be set as needed. Optionally, to set an appropriate threshold, it can be based on one or more factors such as the importance of the business system, compliance requirements, operating costs, or stability. Higher business system importance, higher compliance requirements, lower operating costs, and higher stability generally warrant a larger intervention threshold.

[0094] Alternatively, to set an appropriate intervention threshold, an upper limit for the number of interventions can be determined first, ensuring that the operational costs and stability of the business system meet the requirements. With the goal of ensuring that the intervention threshold does not exceed the upper limit, an intervention threshold matching the importance and compliance requirements of the business system can be set. It is understood that the intervention threshold is less than or equal to the upper limit.

[0095] In practical applications, big data mining can be performed on historical data of business systems to determine the upper limit of intervention times (also known as the maximum number of interventions) that satisfies the requirements for both operational costs and stability. For example, expert experience analysis can be used to mine historical data to determine this upper limit. Another example is using a large language model to mine historical data, or using a trained deep learning model to output the upper limit of intervention times based on historical data of the business system.

[0096] In practical applications, a reasonable intervention threshold is set to balance the importance of the business system with compliance requirements, while not exceeding the upper limit of intervention frequency. For example, expert experience analysis can be used to set a reasonable intervention threshold based on this balance. Another example is using a trained deep learning model to output the intervention threshold based on the importance of the business system, compliance requirements, and the upper limit of intervention frequency. Yet another example is determining the intervention frequency range that matches the importance of the business system by querying a pre-established correspondence between importance and intervention frequency ranges; determining the intervention frequency range that matches the compliance requirements by querying a pre-established correspondence between compliance requirements and intervention frequency ranges; identifying the overlapping intervention frequency ranges between the ranges matching the importance and compliance requirements; and selecting the number of interventions not exceeding the upper limit from these overlapping ranges as the intervention threshold.

[0097] 207. After at least one target tool has been fully invoked, generate and output the response information corresponding to the input information based on the response results of at least one target tool.

[0098] Specifically, after all the target tools have been invoked, the system can return corresponding response information to the user based on their input. For example, if the user's input is "Please create a special feature on cross-border tourism for me," the response would be "Cross-border tourism promotes economic cooperation and development among countries along the route, revitalizing the ancient Silk Road...". Another example is "Please upload article 45987 to the sports channel," which would receive the response "Article uploaded."

[0099] The technical solution provided in this application allows users to interact with the business system via natural language. Leveraging the powerful reasoning capabilities of a large language model, the interaction process between the user and the business system is intelligently transformed. The large language model combines the tool configuration information of multiple tools with the user's input information to infer the tool parameters and calling order of at least one tool required to process the input information. Tools are called sequentially according to the calling order. Before calling a tool, the tool parameters are automatically validated, and if validation fails, the user can intervene in the tool parameters, thereby ensuring the reliability of tool calls and meeting the user's personalized needs. This improves the user's interaction experience with the business system, enabling the business system to more easily adapt to flexible and changing user needs, better meet personalized niche demands, and improve the intelligence, reliability, and operational efficiency of the business system.

[0100] Figure 3 A flowchart illustrating another interaction method based on a large language model provided in this application embodiment. See also... Figure 3 The method may include the following steps:

[0101] 301. Obtain the user's input information in natural language form.

[0102] 302. Obtain the user's input information in natural language form, and based on the input information, the tool configuration information of multiple tools, and the large language model, determine the tool parameters and calling order of at least one target tool required to process the input information.

[0103] 303. Verify the tool parameters of the target tools in the order in which at least one target tool is called.

[0104] 304. If the tool parameters of the target tool fail the validation, output the tool parameters of the target tool and parameter intervention guidance information.

[0105] 305. In response to the user's operation of inputting guidance information based on parameters, modify the tool parameters of the target tool to obtain the updated tool parameters of the target tool; and call the target tool according to the updated tool parameters and obtain the response result returned by the target tool.

[0106] For details on the implementation of steps 301 to 305, please refer to the relevant content of the foregoing embodiments, which will not be described here again.

[0107] 306. If the tool parameters of the target tool pass the verification, output the tool parameters of the target tool and the secondary confirmation guidance information.

[0108] 307. In response to the user's confirmation operation based on the secondary confirmation guidance information, call the target tool according to the tool parameters and obtain the response result returned by the target tool.

[0109] In practical applications, after the business system automatically confirms that the tool parameters of the target tool have passed the verification, the tool parameters can be manually confirmed a second time before the tool is called, thereby improving the reliability of the tool call. The secondary confirmation mechanism increases user participation and enhances the user experience.

[0110] Optionally, when configuring the tool's configuration information, it's possible to flexibly configure whether secondary confirmation of tool parameters is required before calling the tool, improving the flexibility of tool invocation. The implementation method for outputting the target tool's tool parameters and secondary confirmation guidance information is as follows: determine whether a secondary confirmation flag exists in the target tool's configuration information; if a secondary confirmation flag exists, output the target tool's tool parameters and secondary confirmation guidance information. If no secondary confirmation flag exists in the target tool's configuration information, directly call the target tool based on the tool parameters and obtain the response result returned by the target tool.

[0111] When a user enters a confirmation action based on the secondary confirmation prompt, the target tool is invoked according to the tool parameters; when a user enters a cancel action based on the secondary confirmation prompt, the target tool is not invoked.

[0112] 308. After at least one target tool has been fully invoked, generate and output the response information corresponding to the input information based on the response results of at least one target tool.

[0113] The technical solution provided in this application allows users to interact with the business system via natural language. Leveraging the powerful reasoning capabilities of a large language model, the interaction process between the user and the business system is intelligently transformed. The large language model combines the tool configuration information of multiple tools with the user's input information to infer the tool parameters and calling order of at least one tool required to process the input information. Tools are called sequentially according to the calling order. Before calling a tool, the tool parameters are automatically verified, and if verification fails, the user can intervene in the tool parameters, thereby ensuring the reliability of tool calls. Furthermore, if verification is successful, the tool parameters can be manually confirmed a second time before calling the tool, further improving the reliability of tool calls. This secondary confirmation mechanism increases user participation and enhances the user experience. In this way, the interaction experience between the user and the business system is improved, enabling the business system to more easily adapt to flexible and changing user needs, better meet personalized niche needs, and improve the intelligence, reliability, and operational efficiency of the business system.

[0114] The following is combined with Figure 4 This describes the interaction process between users and the production and operation business systems within the converged media platform. (See also...) Figure 4The production and operation business system provides large-scale model application services, a security detection module (shown in the diagram), an intent service (not shown in the diagram), and a data loading module. The large-scale model application service manages and connects the entire business process; the security detection module is responsible for detecting the legality of interface calls to ensure the security and controllability of actual business operations; the intent service identifies the user's input questions to ensure that the input questions can be routed to the corresponding intelligent agent for execution; and the data loading module is used to load business domain knowledge and tool configurations.

[0115] First, the data loading module loads the business domain knowledge corresponding to each business in the production and operation business system; then, the data loading module loads the tool configuration information of each tool; and finally, the tool loader loads the tool configuration information of each tool into the large model application service.

[0116] Next, when a user interacts with the production and operation business system, the system obtains the user's input question information and performs a security check on it. If the security check fails, the entire process ends. If the security check passes, the system calls the intent service to perform intent recognition on the question information and determines the matching agent from multiple agents based on the intent recognition result. If the matched agent is a question-answering domain agent, the user's question information is directly input into the large language model for inference, and the inference result is obtained, ending the entire process. If the matched agent is a business domain agent, vector retrieval is performed by combining business domain knowledge, the vector representation of the question, and the vector representation of the tool's configuration information to determine the tool configuration information of candidate tools that match the question information. The large language model is then used to perform inference by combining business domain knowledge, the tool configuration information of candidate tools, and the question information to obtain the inference result. The inference result includes the tool parameters and calling order of at least one tool required to process the question information. The tool parameters of the inferred tool are then verified, i.e., it is determined whether the tool parameters are standardized. If the parameters do not meet the specifications, intervention is required on the inferred tool's parameters. In this case, with parameter backfilling capabilities, the user is allowed to modify the tool's parameters, such as filling in missing or incorrect values. After modification, the user can use the updated parameters to invoke the tool. The entire process ends after all tools have been invoked. If the parameters meet the specifications, it is further determined whether secondary confirmation is needed. If not, the tool is invoked directly. If secondary confirmation is required, the process waits for user confirmation before invoking the tool.

[0117] For a specific example, the production and operation business system, supported by a multi-agent mechanism, provides tool invocation agents, question-and-answer domain agents, and customer service question-and-answer agents. The tool invocation agent provides multiple tools, including "intelligent summary tool," "intelligent title tool," "article upload tool," "article removal tool," and "topic addition tool," etc. The intelligent summary tool intelligently generates article summaries by analyzing given text descriptions; the intelligent title tool intelligently generates article titles by analyzing given text; the article upload tool sets the status of an article to upload based on a given article ID; and the article removal tool sets the status of an article to offline based on a given article ID.

[0118] After receiving a user's input question, if the production and operation business system determines that a tool needs to be invoked by an intelligent agent, it first uses vector retrieval to identify at least one candidate tool from among several tools provided by the intelligent agent. Then, it fills the prompt word template with the user's input question and the tool configuration information of at least one candidate tool to obtain the target prompt word. For example, the target prompt word might be: "Your task is to perform reasoning based on the user's input question and the tool configuration information of at least one tool, returning the tool parameters and invocation order, where the tool information is returned in the format [{"tool name":;"tool parameters":}]. The target prompt word is then input into a large language model for reasoning to obtain the inference tool parameters. These parameters can be corrected with manual intervention, and the tool is invoked based on the corrected parameters.

[0119] For example, a user inputs the question "Please help me upload article 45987 to the sports channel." The large language model returns the inference result "[{"Tool Name": Article Upload, "Tool Parameters": "item_id": "45987", "channel_id": "Sports"}}]", which provides complete tool parameters and allows the article upload tool to upload article 45987 to the sports channel. However, if the large language model returns the inference result "[{"Tool Name": Article Upload, "Tool Parameters": "item_id": "45987", "channel_id": ""}]", meaning the large language model does not provide a value for the "channel_id" parameter, manual intervention is needed to complete the "channel_id" parameter to "sports". After completing the parameter, the article upload tool can then be used to upload article 45987 to the sports channel.

[0120] Figure 5 A flowchart illustrating another interaction method based on a large language model provided in this application embodiment. This method is applied to the production and operation business system in a converged media platform; see [link to relevant documentation]. Figure 5The method includes the following steps:

[0121] 501. In response to natural language input information sent by the user from the user terminal, determine the tool parameters and invocation order of at least one target tool to be invoked to process the input information based on the input information, tool configuration information of multiple tools and the large language model.

[0122] 502. Verify the tool parameters of the target tools in the order in which at least one target tool is invoked.

[0123] 503. If the tool parameters of the target tool fail the validation, output the tool parameters and parameter intervention guidance information of the target tool.

[0124] 504. In response to the user's operation of inputting guidance information based on parameters, modify the tool parameters of the target tool to obtain the updated tool parameters of the target tool.

[0125] 505. Call the target tool according to the updated tool parameters and obtain the response result returned by the target tool.

[0126] 506. After at least one target tool has been fully invoked, generate response information corresponding to the input information based on the response results of at least one target tool.

[0127] 507. Send a reply message to the user.

[0128] about Figure 5 The implementation methods of each step in the illustrated embodiment can be found in the relevant content of the foregoing embodiments, and will not be repeated here.

[0129] Figure 6 This is a schematic diagram illustrating the structure of an interactive device based on a large language model, provided as an embodiment of this application. The device can consist of hardware and / or software and can be applied to business systems. See also... Figure 6 The interactive device 60 based on a large language model may include:

[0130] Module 61 is used to acquire the user's input information in natural language form;

[0131] Inference module 62 is used to determine the tool parameters and calling order of at least one target tool to be called to process the input information based on the input information, the tool configuration information of multiple tools and the large language model;

[0132] The calling module 63 is used to verify the tool parameters of the target tools in the order of calling at least one target tool, and output the tool parameters and parameter intervention guidance information of the target tool if the tool parameters of the target tool fail the verification; in response to the user's operation based on the parameter intervention guidance information, modify the tool parameters of the target tool to obtain the updated tool parameters of the target tool; and call the target tool according to the updated tool parameters and obtain the response result returned by the target tool.

[0133] The output module 64 is used to generate and output the response information corresponding to the input information based on the response results of at least one target tool after all target tools have been invoked.

[0134] Optionally, module 63 is invoked to verify the updated tool parameters of the target tool. If the updated tool parameters fail the verification and the number of parameter interventions has not yet reached the intervention threshold, the steps of outputting the tool parameters and parameter intervention guidance information of the target tool and subsequent steps are executed again until the updated tool parameters pass the verification or the number of parameter interventions reaches the intervention threshold.

[0135] Optionally, the device may further include a setting module for setting an intervention threshold based on one or more of the following: the importance of the business system, compliance requirements, operating costs, or stability.

[0136] Further optionally, the setting module is specifically used to: determine the upper limit of the number of interventions that ensures both the operating cost and stability of the business system meet the requirements; and set an intervention threshold that matches the importance and compliance requirements of the business system, with the goal of ensuring that the intervention threshold does not exceed the upper limit of the intervention threshold.

[0137] Further optionally, the inference module 62 is specifically used to: select at least one candidate tool from multiple tools whose tool configuration information matches the input information; fill the input information and the tool configuration information of at least one candidate tool into the prompt word template to obtain the target prompt word; input the target prompt word into the large language model for inference to determine the tool parameters and invocation order of at least one target tool required to process the input information.

[0138] Further optionally, when the inference module 62 selects at least one candidate tool from multiple tools whose tool configuration information matches the input information, it is specifically used to: perform vector retrieval in a vector database based on the vector representation of the input information, the vector database including vector representations of tool configuration information of multiple tools; and determine at least one candidate tool whose tool configuration information matches the input information based on the vector retrieval results.

[0139] Optionally, the calling module 63 is further configured to: output the tool parameters of the target tool and secondary confirmation guidance information if the tool parameters of the target tool pass the verification; and in response to the user's confirmation operation based on the secondary confirmation guidance information, call the target tool according to the tool parameters and obtain the response result returned by the target tool.

[0140] Optionally, when calling module 63 to output the tool parameters and secondary confirmation guidance information of the target tool, it is specifically used to: determine whether there is a secondary confirmation mark in the tool configuration information of the target tool; if there is a secondary confirmation mark in the tool configuration information of the target tool, then output the tool parameters and secondary confirmation guidance information of the target tool.

[0141] Further optionally, the inference module 62 is specifically used for: performing intent recognition on the input information to obtain the intent information of the input information; determining the target intelligent agent that matches the intent information from multiple intelligent agents; if the target intelligent agent belongs to the business domain intelligent agent, then determining the tool parameters and calling order of at least one target tool to be called to process the input information based on the input information, the tool configuration information of multiple tools corresponding to the target intelligent agent, and the large language model.

[0142] Further optionally, before the inference module 62 performs intent recognition on the input information to obtain the intent information of the input information, it is also used to: perform security verification on the input information; and if the input information passes the security verification, perform the step of performing intent recognition on the input information to obtain the intent information of the input information.

[0143] Figure 6 The apparatus shown can perform the methods described in the foregoing embodiments; its implementation principles and technical effects will not be elaborated further. Regarding the above embodiments... Figure 6 The specific ways in which each module and unit of the device performs operations have been described in detail in the embodiments of the method, and will not be elaborated here.

[0144] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 201 to 207 can be device A; or the execution subject of steps 201 and 202 can be device A, and the execution subject of steps 203 to 207 can be device B; and so on.

[0145] Furthermore, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 201, 202, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0146] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0147] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device includes: a memory 71 and a processor 72;

[0148] Memory 71 is used to store computer programs and can be configured to store various other data to support operation on the computing platform. Examples of this data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, pictures, videos, etc.

[0149] The memory 71 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0150] Processor 72, coupled to memory 71, is used to execute computer programs in memory 71 for: executing steps in an interactive method based on a large language model.

[0151] Optional, such as Figure 7 As shown, the electronic device also includes other components such as a communication component 73, a display 74, a power supply component 75, and an audio component 76. Figure 7 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 7 The components shown. Additionally... Figure 7 The components within the dashed box are optional, not mandatory, and their specific requirements depend on the product form of the electronic device. The electronic device in this embodiment can be a desktop computer, laptop computer, smartphone, or IoT (Internet of Things) device, or a server-side device such as a conventional server, cloud server, or server array. If the electronic device in this embodiment is a desktop computer, laptop computer, or smartphone, it may include... Figure 7 The components within the dashed box; if the electronic device in this embodiment is implemented as a conventional server, cloud server, or server array, etc., it may be omitted. Figure 7 The component within the dashed box.

[0152] For a detailed description of the implementation process of each action by the processor, please refer to the relevant descriptions in the foregoing method embodiments or device embodiments, which will not be repeated here.

[0153] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be performed by an electronic device in the above method embodiments.

[0154] Accordingly, this application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, enables the processor to perform the steps that can be executed by an electronic device in the above method embodiments.

[0155] The aforementioned communication components are configured to facilitate wired or wireless communication between the device containing the communication components and other devices. The device containing the communication components can access wireless networks based on communication standards, such as WiFi (Wireless Fidelity), 2G (2nd Generation), 3G (3rd Generation), 4G (4th Generation) / LTE (long Term Evolution), 5G (5th Generation), or combinations thereof. In one exemplary embodiment, the communication components receive broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication components also include a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wide Band (UWB), Bluetooth, and other technologies.

[0156] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0157] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0158] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0159] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0160] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0161] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0162] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0163] In a typical configuration, an electronic device includes one or more processors (Central Processing Unit, CPU), input / output interfaces, network interfaces, and memory.

[0164] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0165] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by electronic devices. As defined in this article, computer-readable media do not include transient media, such as modulated data signals and carrier waves.

[0166] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0167] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A large language model-based interaction method, characterized in that, The method is applied to a business system and comprises: obtaining input information in a natural language form of a user, and determining tool parameters and a calling sequence of at least one target tool required to be called for processing the input information according to the input information, tool configuration information of a plurality of tools and a large language model; checking the tool parameters of the target tool in sequence according to the calling sequence of the at least one target tool, and outputting the tool parameters of the target tool and parameter intervention guide information in a case where the tool parameters of the target tool fail to pass the checking; in response to an operation input by the user based on the parameter intervention guide information, modifying the tool parameters of the target tool to obtain updated tool parameters of the target tool, calling the target tool according to the updated tool parameters, and obtaining a response result returned by the target tool; generating and outputting reply information corresponding to the input information according to the response result of the at least one target tool after the at least one target tool is completely called. Before calling the target tool according to the updated tool parameters, the method further comprises:

2. The method of claim 1, wherein, checking the updated tool parameters of the target tool; if the updated tool parameters fail to pass the checking and a parameter intervention frequency has not reached a threshold value of the intervention frequency, the step of outputting the tool parameters of the target tool and the parameter intervention guide information and subsequent steps are executed again until the updated tool parameters pass the checking or the parameter intervention frequency reaches the threshold value of the intervention frequency. The method further comprises:

3. The method of claim 2, wherein, setting the threshold value of the intervention frequency according to one or more of an importance of the business system, a compliance requirement, an operation cost or stability. Setting the threshold value of the intervention frequency according to one or more of the importance of the business system, the compliance requirement, the operation cost or the stability comprises:

4. The method of claim 3, wherein, determining an upper limit value of the intervention frequency such that the operation cost and the stability of the business system both meet requirements; setting the threshold value of the intervention frequency to match the importance and the compliance requirement of the business system, with the threshold value of the intervention frequency not exceeding the upper limit value of the intervention frequency as a target. Determining the tool parameters and the calling sequence of the at least one target tool required to be called for processing the input information according to the input information, the tool configuration information of the plurality of tools and the large language model comprises:

5. The method of claim 1, wherein, selecting at least one candidate tool whose tool configuration information matches the input information from the plurality of tools; filling the input information and the tool configuration information of the at least one candidate tool into a prompt word template to obtain a target prompt word; inputting the target prompt word into the large language model for reasoning to determine the tool parameters and the calling sequence of the at least one target tool required to be called for processing the input information. Selecting at least one candidate tool whose tool configuration information matches the input information from the plurality of tools comprises:

6. The method of claim 5, wherein, performing vector retrieval according to a vector representation of the input information in a vector database, the vector database comprising vector representations of tool configuration information of a plurality of tools; determining at least one candidate tool whose tool configuration information matches the input information according to a vector retrieval result. The method further comprises:

7. The method of claim 1, wherein, ​ In a case where the tool parameter of the target tool is verified, output the tool parameter of the target tool and secondary confirmation guide information; In response to a confirmation operation input by the user based on the secondary confirmation guide information, invoke the target tool according to the tool parameter, and obtain a response result returned by the target tool.

8. The method of claim 7, wherein, Outputting the tool parameter of the target tool and the secondary confirmation guide information includes: Determining whether there is a secondary confirmation mark in the tool configuration information of the target tool; If there is a secondary confirmation mark in the tool configuration information of the target tool, output the tool parameter of the target tool and the secondary confirmation guide information.

9. The method of claim 1, wherein, According to the input information, the tool configuration information of a plurality of tools, and the large language model, determining the tool parameter and the calling sequence of at least one target tool required to process the input information includes: Performing intent recognition on the input information to obtain intent information of the input information; Determining a target agent matching the intent information from a plurality of agents; If the target agent belongs to a business domain agent, according to the input information, the tool configuration information of a plurality of tools corresponding to the target agent, and the large language model, determining the tool parameter and the calling sequence of at least one target tool required to process the input information.

10. The method of claim 9, wherein, Before performing intent recognition on the input information to obtain intent information of the input information, it further includes: Performing security verification on the input information; In a case where the input information passes the security verification, performing intent recognition on the input information to obtain intent information of the input information.

11. A large language model-based interaction method, characterized in that, Applied to a production and operation business system in a fusion media platform, the method includes: In response to input information in a natural language form of a user sent by a user end, according to the input information, tool configuration information of a plurality of tools, and a large language model, determining the tool parameter and the calling sequence of at least one target tool required to process the input information; According to the calling sequence of the at least one target tool, verifying the tool parameter of the target tool in sequence, and in a case where the tool parameter of the target tool does not pass the verification, outputting the tool parameter of the target tool and parameter intervention guide information; In response to an operation input by the user based on the parameter intervention guide information, modifying the tool parameter of the target tool to obtain an updated tool parameter of the target tool; and invoking the target tool according to the updated tool parameter, and obtaining a response result returned by the target tool; After the at least one target tool is completely invoked, generating reply information corresponding to the input information according to the response result of the at least one target tool; Sending the reply information to the user end.

12. An electronic device, comprising: Including: A memory and a processor; The memory is used to store a computer program; The processor is coupled to the memory and is used to execute the computer program to perform the steps in the method of any one of claims 1-11.

13. A computer readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the processor can implement the steps in the method of any one of claims 1-11.

14. A computer program product, characterised in that, Computer program / instructions, which when executed by a processor, enable the processor to implement the steps of the method of any of claims 1-11.

Images