Tool selection method and tool selection apparatus and artificial intelligence system

Abstract

The present disclosure provides a tool selection method and a tool selection device and an artificial intelligence system, and relates to the field of computers, in particular to the field of artificial intelligence. In response to indication information provided by a user to an artificial intelligence system, the similarity between the indication information and tool information in a tool set of the artificial intelligence system is determined, and based on the similarity, a candidate tool set related to the indication information is determined. According to the indication information and the tool information in the candidate tool set, a large language model is used to select a target tool from the candidate tool set to process the indication information. Thus, the number of tools that need to be identified by the large language model is greatly reduced, and even in the case of a large number of tools in the tool set, the accuracy and latency level of the identified tools can be guaranteed.

Description

Technical Field

[0001] This disclosure relates to the field of computers, particularly to the field of artificial intelligence, and especially to tool selection methods, tool selection devices, and artificial intelligence systems. Background Technology

[0002] In artificial intelligence applications, the ability of intelligent agents to invoke tools is a crucial element in enhancing user experience and automating tasks.

[0003] In some related technologies, intelligent agents select and invoke tools based on the function call capabilities of Large Language Models (LLMs) or on prompt words. The prompt word-based technology aims to write appropriate prompts so that the Large Language Model can provide more accurate results.

[0004] These related technologies rely on the capabilities of large language models to identify tools. When the number of tools is large, the accuracy of large language models in identifying tools will drop sharply and the latency will increase. Summary of the Invention

[0005] In this embodiment, in response to a user's instruction to an artificial intelligence system, the system determines the similarity between the instruction and tool information in the system's toolset. Based on the similarity, it determines a candidate toolset related to the instruction. Then, using a large language model, it selects a target tool from the candidate toolset to process the instruction, based on the instruction and the tool information in the candidate toolset. This significantly reduces the number of tools requiring large language model recognition, ensuring accuracy and low latency even when the toolset is large.

[0006] In addition, candidate tools in the candidate tool set are called in advance to obtain the output information of the candidate tools in response to the instruction information. In addition to the user instruction information and the tool information of the candidate tools in the candidate tool set, the output information of the candidate tools in response to the instruction information is also used as information that the large language model needs to analyze, so that the large language model can more accurately determine the target tool suitable for processing the user instruction information.

[0007] This disclosure provides a tool selection method in some embodiments, including:

[0008] In response to instructions from users to the artificial intelligence system, determine the similarity between the instructions and tool information in the toolset of the artificial intelligence system;

[0009] Based on the similarity, a candidate toolset related to the indication information is determined;

[0010] Based on the instruction information and the tool information in the candidate toolset, a target tool for processing the instruction information is selected from the candidate toolset using a large language model.

[0011] In some embodiments, after determining the candidate toolset, candidate tools in the candidate toolset are invoked in advance to obtain the result information output by the candidate tools in response to the indication information;

[0012] The process of selecting a target tool for processing the instruction information from the candidate toolset includes: selecting a target tool for processing the instruction information from the candidate toolset using a large language model, based on the instruction information and the tool information and result information of the candidate tools in the candidate toolset.

[0013] In some embodiments, the tool information in the toolset includes the tool's corresponding description information and one or more typical instruction information for configuration.

[0014] In some embodiments, determining the similarity between the instruction information and tool information in the toolset of the artificial intelligence system includes: determining the similarity between the instruction information and the description information corresponding to the tools in the toolset, and the similarity between the instruction information and each typical instruction information.

[0015] In some embodiments, determining a candidate toolset related to the indication information based on the similarity includes: if any similarity of any tool in the toolset meets the candidate tool requirements, determining the tool as a candidate toolset related to the indication information.

[0016] In some embodiments, determining the similarity between the indication information and tool information in the toolset of the artificial intelligence system includes:

[0017] Determine the embedding vector of the instruction information and the embedding vector of the tool information in the toolset;

[0018] Calculate the similarity between the embedding vector of the indication information and the embedding vector of the tool information in the toolset.

[0019] In some embodiments, determining the similarity between the indication information and tool information in the toolset of the artificial intelligence system includes: using a dual encoder model to determine the similarity between the indication information and tool information in the toolset of the artificial intelligence system.

[0020] In some embodiments, the indication information is the user's activities on a specified date, and the candidate toolset includes to-do list tools and meeting tools;

[0021] The candidate tools in the pre-call candidate toolset include: a pre-call to-do list tool to obtain the user's to-do list for a specified date, and a pre-call meeting tool to obtain the meeting that the user has scheduled for a specified date;

[0022] Selecting a target tool from the candidate toolset to process the instruction information includes: if the user has a to-do list for a specified date, the target tool selected includes a to-do list tool; if the user has a meeting scheduled for a specified date, the target tool selected includes a meeting tool.

[0023] Some embodiments of this disclosure provide a tool selection apparatus, including: a memory; and a processor coupled to the memory, the processor being configured to execute a tool selection method based on instructions stored in the memory.

[0024] This disclosure provides some embodiments of a tool selection apparatus, including a module that performs a tool selection method.

[0025] Some embodiments of this disclosure propose an artificial intelligence system, including: a tool selection device configured to perform a tool selection method.

[0026] Some embodiments of this disclosure provide a computer-readable storage medium having computer instructions stored thereon that, when executed by a processor, implement a tool selection method.

[0027] Some embodiments of this disclosure provide a computer program product including computer instructions that, when executed by a processor, implement a tool selection method. Attached Figure Description

[0028] The accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. This disclosure can be more clearly understood from the following detailed description with reference to the accompanying drawings.

[0029] Obviously, the accompanying drawings described below are merely some embodiments of this disclosure. Those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0030] Figure 1 A flowchart illustrating a similarity-based tool selection method according to some embodiments of this disclosure is shown.

[0031] Figure 2 This illustration shows a schematic diagram of some embodiments of the present disclosure that determine candidate tools related to user instruction information based on the similarity between user instruction information and tool information.

[0032] Figure 3The diagram illustrates a flowchart of a pre-invocation-based tool selection method according to some embodiments of this disclosure.

[0033] Figure 4 A schematic diagram illustrating a pre-invocation-based tool selection method according to some embodiments of this disclosure is shown.

[0034] Figure 5 A schematic diagram of the structure of a tool selection device according to some embodiments of the present disclosure is shown.

[0035] Figure 6 A schematic diagram of the structure of a tool selection device according to some embodiments of the present disclosure is shown. Detailed Implementation

[0036] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0037] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of this disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0038] It should also be understood that in the embodiments disclosed herein, "a plurality of" may refer to two or more, and "at least one" may refer to one, two or more.

[0039] It should also be understood that any component, data or structure mentioned in the embodiments of this disclosure can generally be understood as one or more unless expressly defined or given to the contrary in the context.

[0040] Furthermore, the term "and / or" in this disclosure is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this disclosure generally indicates that the preceding and following related objects have an "or" relationship.

[0041] It should also be understood that the description of the various embodiments in this disclosure emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0042] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0043] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0044] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0045] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0046] Furthermore, to avoid obscuring this disclosure with unnecessary detail, only processing steps and / or apparatus structures closely related to at least the solutions according to this disclosure are shown in the accompanying drawings, while other details not closely related to this disclosure are omitted. It should also be noted that similar reference numerals and letters in the drawings indicate similar items, and therefore once an item is defined in one drawing, it need not be discussed again in subsequent drawings.

[0047] Figure 1 The diagram illustrates a similarity-based tool selection method according to some embodiments of this disclosure. The tool selection method can be executed, for example, by a tool selection device of an artificial intelligence system, which can be, for example, an intelligent agent of the artificial intelligence system.

[0048] like Figure 1 As shown, the tool selection method in this embodiment includes the following steps 110-130.

[0049] In step 110, in response to the instruction information provided by the user to the artificial intelligence system, the similarity between the instruction information and the tool information in the tool set of the artificial intelligence system is determined. The similarity is used to characterize the degree of matching between the instruction information and the tool information.

[0050] Artificial intelligence systems are computer systems that can simulate human intelligence and possess abilities such as learning, reasoning, decision-making, and perception to accomplish specific tasks or solve specific problems.

[0051] Artificial intelligence systems typically contain a variety of tools with diverse functions to meet users' varied service needs. A tool refers to an application or other medium with specific functions. Examples of tools include, but are not limited to, weather forecasting tools, to-do list tools, meeting tools, translation tools, mapping tools, and other functional tools.

[0052] The tool information in a toolset includes, for example, the tool's name, parameters, description, and one or more typical instructions. The tool name is its identifier, uniquely identifying the tool within the toolset for easy identification and retrieval. Tool parameters are the input values ​​or configuration options the tool needs to receive when running or performing specific operations, controlling its behavior and functionality. Tool parameters make the tool more flexible and versatile, adapting to different application scenarios and user needs. The tool description explains its functions, features, applicable scenarios, or working principles, helping users understand its basic information and purpose. Typical instructions refer to examples of requests or operation commands that users typically send to the tool, demonstrating common usage methods. To more accurately calculate the similarity between instructions and tool information, one or more typical instructions can be configured for each tool.

[0053] User instructions may include requests for information or assistance from the AI ​​system. Examples include, but are not limited to, requests for weather, flight, or meeting information from the AI ​​system.

[0054] Determining the similarity between instruction information and tool information in the toolset of the artificial intelligence system includes, for example, determining the similarity between the instruction information and the corresponding descriptive information of the tool in the toolset, as well as the similarity between the instruction information and each typical instruction information. Therefore, based on different tool information items of an instruction information and a tool, multiple similarities between the instruction information and the tool can be calculated.

[0055] An exemplary method for determining similarity includes: determining the embedding vector of the indicator information and the embedding vector of the tool information in the toolset; and calculating the similarity between the embedding vector of the indicator information and the embedding vector of the tool information in the toolset. This calculation method allows for flexible selection of different algorithms and tools according to specific needs in the steps of determining the embedding vector and calculating the similarity, offering high flexibility.

[0056] This involves using word embedding algorithms or pre-trained language models to determine the embedding vectors of information. Examples of word embedding algorithms include Word2Vec and Global Vectors for Word Representation (GloVe). Examples of pre-trained language models include Bidirectional Encoder Representations from Transformer (BERT) and Generative Pretrained Transformer (GPT) models.

[0057] Another exemplary method for determining similarity involves using a bi-encoder model to determine the similarity between instruction information and tool information in the toolset of an AI system. Bi-encoder models are typically built on deep learning architectures such as Transformer, Convolutional Neural Networks (CNNs), or Recurrent Neural Networks (RNNs). In a bi-encoder model, two encoders encode the instruction information and tool information respectively, and then the similarity between the encoded information of the instruction information and the encoded information of the tool information is calculated. This allows for the automatic learning of complex semantic features and patterns, better capturing the semantic relationships between the instruction information and the tool information, and thus more accurately calculating the similarity.

[0058] In step 120, a set of candidate tools related to the indication information is determined based on similarity.

[0059] As mentioned before, multiple similarities between a certain instruction and a certain tool may be calculated. In this case, if any similarity score of a tool in the tool set meets the requirements for a candidate tool, that tool is identified as a candidate tool related to the instruction. Thus, by evaluating the similarity between the same tool and the user's instruction from multiple information dimensions, candidate tools related to the instruction can be identified more accurately.

[0060] One approach is to set certain thresholds to select candidate tools. For example, a threshold can be set for the number of candidate tools to select a certain number of them. Another example is to set a similarity threshold to identify tools with a similarity score higher than the threshold as candidate tools.

[0061] Figure 2 This illustration shows a schematic diagram illustrating how, according to some embodiments of the present disclosure, candidate tools related to user instruction information are determined based on the similarity between user instruction information and tool information. For example... Figure 2 As shown, user instruction information and tool information are input into the embedding vector model, respectively, to calculate the embedding vectors for user instruction information and tool information. If a tool has multiple tool information items, multiple embedding vectors for that tool are output. For example, if tool 1's tool information items include tool description, typical instruction information 1, typical instruction information 2, and typical instruction information 3, then four embedding vectors for tool 1 are output: the embedding vector for tool description, the embedding vector for typical instruction information 1, the embedding vector for typical instruction information 2, and the embedding vector for typical instruction information 3. Then, the similarity between the embedding vectors of user instruction information and tool information is calculated. If user instruction information and tool 1 have multiple similarities, and if any similarity of tool 1 meets the requirements for a candidate tool, tool 1 is identified as a candidate tool related to the user instruction information. All similarities are sorted to select a candidate tool set.

[0062] In step 130, based on the instruction information and the tool information in the candidate tool set, a target tool for processing the instruction information is selected from the candidate tool set using the Large Language Model (LLM).

[0063] The instruction information and tool information from the candidate tool set are input into the trained large language model. The large language model analyzes and processes the input information and outputs a target tool suitable for processing the instruction information.

[0064] Large language models possess powerful language understanding and generation capabilities, and can be guided to generate results that meet specific needs through appropriate prompts. For example, in artificial intelligence tasks such as text generation, question-answering systems, and dialogue interaction, carefully designed prompts can be input to help large language models understand the task's objectives and requirements, thereby generating high-quality text content.

[0065] First, the user request (i.e., user instruction information) is parsed. A preliminary analysis of the user's input is performed to extract key information, such as the intent of the request, the relevant field, and the required function. For example, if a user requests "find the weather for a certain place tomorrow," the key information is "a certain place," "tomorrow," "weather," and the implicit intent to "check the weather."

[0066] Next, the extracted key information is integrated with the functions and applicable scope of each tool in the toolset to construct prompt words. These prompt words need to convey the user's needs to the large language model and the available tools. For example, if the toolset includes weather query tools, map tools, and translation tools, the prompt word could be designed as: "The user's need is to query the weather for a certain location tomorrow. Existing tools include a weather query tool for querying weather information for various locations; a map tool for geographic information queries; and a translation tool for language translation. Please determine which tool is suitable for handling this user request."

[0067] Next, the constructed prompt words are input into the large language model. Based on its pre-trained knowledge and language understanding capabilities, the large language model processes and analyzes the prompt words, evaluating the relevance of the user's request to each tool. By understanding the semantic and contextual information in the prompt words, it calculates the degree of matching between each tool and the user's request. For example, for a request to "check the weather for a certain location tomorrow," the model will analyze that the weather query tool is highly relevant to the request, while the map tool and translation tool are less relevant.

[0068] Finally, based on the relevance calculation results, the large language model generates a list of recommended tools, usually arranged in descending order of relevance. In simple cases, it may directly recommend the most relevant single tool; in complex scenarios, it may provide multiple relevant tools and explain the applicability or advantages of each tool.

[0069] The above approach significantly reduces the number of tools required for large language model recognition. Even with a large number of tools in the toolkit, it can ensure the accuracy and latency of the recognition tools and avoid exceeding the maximum input length of the model.

[0070] Figure 3 The diagram illustrates a flowchart of a pre-invocation-based tool selection method according to some embodiments of this disclosure. The tool selection method can be executed, for example, by a tool selection device of an artificial intelligence system, which can be, for example, an intelligent agent of the artificial intelligence system.

[0071] like Figure 3 As shown, the tool selection method in this embodiment includes the following steps 310-340.

[0072] In step 310, in response to the instruction information provided by the user to the artificial intelligence system, the similarity between the instruction information and the tool information in the tool set of the artificial intelligence system is determined. The specific details of this step are described in step 110 and will not be repeated here.

[0073] In step 320, a candidate toolset related to the indication information is determined based on similarity. The details of this step are described in step 120 and will not be repeated here.

[0074] Normally, you select a tool first, then invoke it. However, in some cases, even based on tool descriptions and other tool information, it can still be difficult to accurately determine the target tool suitable for processing user instructions. In such cases, steps 330 and 340 can be performed to more accurately determine the target tool suitable for processing user instructions.

[0075] In step 330, after determining the candidate toolset, the candidate tools in the candidate toolset are invoked in advance to obtain the result information output by the candidate tools in response to the instruction information.

[0076] If there are multiple candidate tools, each candidate tool in the candidate tool set can be called in parallel in advance to obtain the output information of each candidate tool for the instruction information more quickly, and thus select the target tool for processing the instruction information more quickly.

[0077] In step 340, based on the instruction information and the tool information and result information of the candidate tools in the candidate tool set, the target tool for processing the instruction information is selected from the candidate tool set using a large language model.

[0078] In addition to the instruction information and the tool information of the candidate tools in the candidate tool set, the output information of the candidate tools in response to the instruction information is also included as information that the large language model needs to analyze. That is, the instruction information, the tool information and the output information in the candidate tool set are all input into the trained large language model. The large language model analyzes and processes the input information and outputs the target tool suitable for processing the instruction information. Thus, the target tool suitable for processing the user instruction information is determined more accurately.

[0079] This step is similar to step 130 above, and can also guide the large language model to generate results that meet specific needs through appropriate prompt words. Examples of prompt words are listed below.

[0080] Role: Tool access assistant, selects the most suitable tool and its parameters based on user instructions.

[0081] Currently available tools: A list of currently available tools.

[0082] Results information: Results information output by the candidate tool in response to user instructions.

[0083] Limitations: Output the tools to be called and their corresponding parameters, in the following format: tool name; tool input parameters.

[0084] Query example: The user instructs "Query the weather for a certain location", the output is: tool name "get_weather", and the tool input parameter is {location: a certain location}.

[0085] Figure 4 Schematic diagrams illustrating a pre-invocation-based tool selection method according to some embodiments of this disclosure are shown. For example... Figure 4 As shown, based on the similarity between user instruction information and tool information in the tool set, a preliminary selection of candidate tools related to user instruction information is made. Candidate tools in the candidate tool set are called in advance to obtain the output information of the candidate tools in response to the instruction information. In addition to user instruction information and tool information of candidate tools in the candidate tool set, the output information of candidate tools in response to the instruction information is also used as information that the large language model needs to analyze, so that the large language model can more accurately determine the target tool suitable for processing user instruction information based on the input triple information.

[0086] The following is an example of an application of the pre-invocation-based tool selection method, but the scheme disclosed herein is not limited to this example.

[0087] User instructions refer to user activities scheduled for a specified date. The candidate toolset includes, for example, to-do list tools and meeting tools. Pre-calling candidate tools from the toolet set includes: pre-calling the to-do list tool to retrieve the user's to-do list for the specified date, and pre-calling the meeting tool to retrieve the meetings scheduled by the user for the specified date. Selecting the target tool from the toolet set to process the instructions includes: if the user's to-do list for the specified date is retrieved, the target tool selected is the to-do list tool; if the user's meetings scheduled for the specified date are retrieved, the target tool selected is the meeting tool.

[0088] In this embodiment, candidate tools in the candidate tool set are called in advance to obtain the output information of the candidate tools in response to the instruction information. In addition to the user instruction information and the tool information of the candidate tools in the candidate tool set, the output information of the candidate tools in response to the instruction information is also used as information that the large language model needs to analyze, so that the large language model can more accurately determine the target tool suitable for processing the user instruction information.

[0089] Figure 5 A schematic diagram of the tool selection device according to some embodiments of this disclosure is shown. For example... Figure 5 As shown, the tool selection device 500 of this embodiment includes the following modules.

[0090] The similarity determination module 510 is configured to respond to indication information provided by a user to the artificial intelligence system and determine the similarity between the indication information and tool information in the tool set of the artificial intelligence system.

[0091] The candidate tool determination module 520 is configured to determine a set of candidate tools related to the indication information based on the similarity.

[0092] The target tool determination module 530 is configured to select a target tool for processing the indication information from the candidate tool set using a large language model, based on the indication information and the tool information in the candidate tool set.

[0093] The tool selection device 500 further includes a pre-call module 540, configured to pre-call candidate tools in the candidate toolset after determining the candidate toolset, to obtain the result information output by the candidate tools in response to the instruction information. Correspondingly, the target tool determination module 530 is configured to select a target tool for processing the instruction information from the candidate toolset using a large language model, based on the instruction information and the tool information and result information of the candidate tools in the candidate toolset.

[0094] The tool information in the toolset includes the tool's description information and one or more typical instruction information for configuration.

[0095] The similarity determination module 510 is configured to: determine the similarity between the indication information and the description information corresponding to the tools in the toolset, and the similarity between the indication information and each typical indication information.

[0096] Accordingly, the candidate tool determination module 520 is configured to determine the tool as a candidate tool set related to the indication information if any similarity of any tool in the tool set meets the candidate tool requirements.

[0097] The similarity determination module 510 is configured to: determine the embedding vector of the indication information and the embedding vector of the tool information in the toolset; and calculate the similarity between the embedding vector of the indication information and the embedding vector of the tool information in the toolset.

[0098] The similarity determination module 510 is configured to: use a dual encoder model to determine the similarity between the indication information and the tool information in the toolset of the artificial intelligence system.

[0099] In some embodiments, the indication information is the user's activities on a specified date, and the candidate toolset includes a to-do list tool and a meeting tool. The pre-call module 540 is configured to pre-call the to-do list tool to obtain the user's to-do list for the specified date, and to pre-call the meeting tool to obtain the meetings the user has scheduled for the specified date. The target tool determination module 530 is configured to, when the user's to-do list for the specified date is obtained, select a to-do list tool as the target tool; and when the user's meetings scheduled for the specified date are obtained, select a meeting tool as the target tool.

[0100] Figure 6A schematic diagram of the tool selection device according to some embodiments of this disclosure is shown. For example... Figure 6 As shown, the tool selection device 600 of this embodiment includes a memory 610 and a processor 620 coupled to the memory 610. The processor 620 is configured to execute the tool selection method of each embodiment based on instructions stored in the memory 610.

[0101] The tool selection device 600 may also include an input / output interface 630, a network interface 640, a storage interface 650, etc. These interfaces 630, 640, 650, as well as the memory 610 and the processor 620, can be connected, for example, via a bus 660.

[0102] The memory 610 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, the operating system, application programs, boot loader, and other programs.

[0103] The processor 620 can be implemented using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates, or transistors, or other discrete hardware components.

[0104] The input / output interface 630 provides a connection interface for input / output devices such as monitors, mice, keyboards, and touchscreens. The network interface 640 provides a connection interface for various networked devices. The storage interface 650 provides a connection interface for external storage devices such as SD cards and USB flash drives. The bus 660 can use any bus architecture from a variety of bus structures. For example, bus architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, and Peripheral Component Interconnect (PCI) bus.

[0105] This disclosure also proposes an artificial intelligence system, including: a tool selection device configured to execute the tool selection method of various embodiments.

[0106] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more (non-transitory) computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, cloud storage, etc.) containing computer program code. A computer program product should be understood as a software product that primarily implements its solution through a computer program.

[0107] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should 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, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] 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.

[0109] 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.

Claims

1. A tool selection method, comprising: In response to instructions from users to the artificial intelligence system, determine the similarity between the instructions and tool information in the toolset of the artificial intelligence system; Based on the similarity, a candidate toolset related to the indication information is determined; Based on the instruction information and the tool information in the candidate toolset, a target tool for processing the instruction information is selected from the candidate toolset using a large language model.

2. The tool selection method according to claim 1 further includes: After determining the candidate toolset, the candidate tools in the candidate toolset are invoked in advance to obtain the result information output by the candidate tools in response to the indication information; The process of selecting a target tool for processing the instruction information from the candidate toolset includes: selecting a target tool for processing the instruction information from the candidate toolset using a large language model, based on the instruction information and the tool information and result information of the candidate tools in the candidate toolset.

3. The tool selection method according to claim 1 or 2, wherein, The tool information in the toolset includes the tool's corresponding description information and one or more typical instruction information for configuration.

4. The tool selection method according to claim 3, wherein, Determining the similarity between the instruction information and the tool information in the toolset of the artificial intelligence system includes: determining the similarity between the instruction information and the description information corresponding to the tools in the toolset, and the similarity between the instruction information and each typical instruction information.

5. The tool selection method according to claim 4, wherein, Based on the similarity, the candidate toolset related to the indication information is determined as follows: If any similarity of any tool in the toolset meets the candidate tool requirements, the tool is identified as a candidate toolset related to the indication information.

6. The tool selection method according to claim 1 or 2, wherein, Determining the similarity between the instruction information and the tool information in the toolset of the artificial intelligence system includes: Determine the embedding vector of the instruction information and the embedding vector of the tool information in the toolset; Calculate the similarity between the embedding vector of the indication information and the embedding vector of the tool information in the toolset.

7. The tool selection method according to claim 1 or 2, wherein, Determining the similarity between the instruction information and the tool information in the toolset of the artificial intelligence system includes: Using a dual encoder model, the similarity between the indication information and the tool information in the toolset of the artificial intelligence system is determined.

8. The tool selection method according to claim 2, wherein, The instruction information refers to the user's activities on a specified date, and the candidate toolset includes to-do list tools and meeting tools; The candidate tools in the pre-call candidate toolset include: a pre-call to-do list tool to obtain the user's to-do list for a specified date, and a pre-call meeting tool to obtain the meeting that the user has scheduled for a specified date; Selecting a target tool from the candidate toolset to process the instruction information includes: if the user has a to-do list for a specified date, the target tool selected includes a to-do list tool; if the user has a meeting scheduled for a specified date, the target tool selected includes a meeting tool.

9. A tool selection device, comprising: Memory; And a processor coupled to the memory, the processor being configured to execute the tool selection method of any one of claims 1-8 based on instructions stored in the memory.

10. A tool selection device, comprising: A module that performs the tool selection method according to any one of claims 1-8.

11. An artificial intelligence system, comprising: The tool selection device is configured to perform the tool selection method according to any one of claims 1-8.

12. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by a processor, the computer instructions implement the tool selection method as described in any one of claims 1-8.

13. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the tool selection method as described in any one of claims 1-8.

Images