Intelligent agent generation method and device, storage medium and program product

Abstract

The invention provides an agent generation method and device, a storage medium and a program product, relates to the technical field of computers, and solves the technical problem of difficulty in agent construction in related technologies. The method comprises the steps of generating an agent generation instruction in response to agent generation information input by a user; performing intention analysis on the agent generation instruction to obtain an atomic task of the agent; the atomic task of the agent is used for indicating functions expected by a user to be possessed by the agent; and generating the agent based on the atomic task of the agent and a pre-training template library.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, apparatus, storage medium, and program product for generating an agent. Background Art

[0002] In the current booming development of artificial intelligence, the technology of agent construction, as a key area, is constantly innovating.

[0003] However, there are many bottlenecks in the existing agent construction methods, which seriously restrict their wide application and development. Currently, the agent construction technology based on large models mainly relies on programming capabilities or predefined frameworks, which makes it extremely difficult for ordinary users to construct agents. Summary of the Invention

[0004] This application provides a method, apparatus, storage medium, and program product for generating an agent, which solves the technical problem of difficult agent construction in related technologies.

[0005] In a first aspect, this application provides a method for generating an agent, including: generating an agent generation instruction in response to agent generation information input by a user; performing intent parsing on the agent generation instruction to obtain atomic tasks of the agent; the atomic tasks of the agent are used to indicate the functions that the user expects the agent to have; and generating an agent based on the atomic tasks of the agent and a pre-trained template library.

[0006] In one implementation, performing intent parsing on the agent generation instruction to obtain atomic tasks of the agent includes: performing semantic parsing on the agent generation instruction to obtain clear semantic information, and generating atomic tasks based on the clear semantic information.

[0007] In one implementation, performing semantic parsing on the agent generation instruction to obtain clear semantic information and generating atomic tasks based on the clear semantic information includes: performing semantic parsing on the agent generation instruction to obtain clear semantic information and fuzzy semantic information; performing intent conflict resolution processing on the fuzzy semantic information to obtain candidate semantic information; displaying the candidate semantic information, and determining target semantic information in response to a user selection operation; and generating atomic tasks based on the target semantic information and the clear semantic information.

[0008] In one implementation, an agent is generated based on the atomic tasks of the agent and a pre-trained template library, including: selecting a template corresponding to the atomic task from the pre-trained template library based on the atomic task; generating a module corresponding to the atomic task based on the template corresponding to the atomic task; the module corresponding to the atomic task is a set of codes corresponding to the atomic task; determining the dependency relationship of the module corresponding to the atomic task based on the dependency relationship between the atomic tasks; determining the execution order of the module corresponding to the atomic task based on the sequential constraint between the atomic tasks; generating an agent based on the dependency relationship of the module corresponding to the atomic task and the execution order of the module corresponding to the atomic task.

[0009] In one implementation, the method further includes: testing the agent in a simulated sandbox environment to obtain a test result; and updating the agent based on the abnormal situation in the case where there is an abnormal situation in the test result.

[0010] In one implementation, the method further includes: collecting feedback information in response to a user's evaluation operation on the running result of the agent during the user's use of the agent; and updating the agent based on the feedback information.

[0011] In one implementation, the agent generation information is one or more of text format, voice format, and picture format; in response to the agent generation information input by the user, an agent generation instruction is generated, including: performing a conversion operation on the agent generation information in a non-text format input by the user to obtain the agent generation information in text format; generating an agent generation instruction based on the converted agent generation information in text format; or generating an agent generation instruction based on the converted agent generation information in text format and the agent generation information in text format input by the user.

[0012] In a second aspect, the present application provides an agent generation device, including: a generation module and a processing module.

[0013] The generation module is configured to generate an agent generation instruction in response to the agent generation information input by the user; the processing module is configured to perform intent parsing on the agent generation instruction to obtain the atomic tasks of the agent; the atomic tasks of the agent are used to indicate the functions that the user expects the agent to have; the generation module is further configured to generate an agent based on the atomic tasks of the agent and a pre-trained template library.

[0014] In one implementation, the processing module is specifically configured to perform semantic parsing on the agent generation instruction to obtain clear semantic information, and generate atomic tasks based on the clear semantic information.

[0015] In one implementation, the processing module is further specifically configured to perform semantic parsing on the instructions generated by the agent to obtain clear semantic information and fuzzy semantic information; the processing module is further specifically configured to perform intention conflict resolution processing on the fuzzy semantic information to obtain candidate semantic information; the processing module is further specifically configured to display the candidate semantic information and determine the target semantic information in response to a user selection operation; the generation module is specifically configured to generate atomic tasks based on the target semantic information and the clear semantic information.

[0016] In one implementation, the processing module is further configured to select a template corresponding to the atomic task from a pre-trained template library based on the atomic task; the generation module is further configured to generate a module corresponding to the atomic task based on the template corresponding to the atomic task; the module corresponding to the atomic task is a code set corresponding to the atomic task; the processing module is further configured to determine the dependency relationship of the module corresponding to the atomic task based on the dependency relationship between the atomic tasks; the processing module is further configured to determine the execution order of the module corresponding to the atomic task based on the sequential constraint between the atomic tasks; the generation module is further configured to generate an agent based on the dependency relationship of the module corresponding to the atomic task and the execution order of the module corresponding to the atomic task.

[0017] In one implementation, the processing module is further configured to test the agent in a simulated sandbox environment to obtain a test result; the processing module is further configured to update the agent based on the abnormal situation in the case where an abnormal situation exists in the test result.

[0018] In one implementation, the processing module is further configured to collect feedback information in response to a user's evaluation operation on the running result of the agent during the user's use of the agent; the processing module is further configured to update the agent based on the feedback information.

[0019] In one implementation, the agent generation information is one or more of text format, voice format, and picture format; the processing module is further configured to perform a conversion operation on the agent generation information in a non-text format input by the user to obtain agent generation information in text format; the generation module is further configured to generate an agent generation instruction based on the converted agent generation information in text format; or, the generation module is further configured to generate an agent generation instruction based on the converted agent generation information in text format and the agent generation information in text format input by the user.

[0020] [[ID=I5]]In a third aspect, the present application provides an agent generation device, including: a processor and a memory configured to store processor-executable instructions; wherein, the processor is configured to execute the instructions to implement any one of the optional methods in the first aspect above.

[0021] Fourth aspect, the present application provides a computer-readable storage medium, on which instructions are stored. When the instructions in the computer-readable storage medium are executed by the agent generation device, the agent generation device can execute any of the optional methods in the first aspect above.

[0022] Fifth aspect, the present application provides a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, any of the optional methods in the first aspect above is implemented.

[0023] The agent generation method, device, storage medium and program product provided by the present application generate agent generation instructions by responding to the agent generation information input by the user, perform intention parsing on the agent generation instructions to obtain the atomic tasks of the agent; and generate an agent based on the atomic tasks of the agent and the pre-trained template library. Compared with the prior art, the requirements for operators in programming and other fields are lower during the process of generating an agent, and it enables users to generate agents that meet their needs more simply and conveniently. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art.

[0025] Figure 1 It is a schematic diagram of the architecture of the agent generation system provided by the embodiment of the present application;

[0026] Figure 2 It is a schematic flowchart of an agent generation method provided by the embodiment of the present application;

[0027] Figure 3 It is a schematic structural diagram of an agent generation device provided by the embodiment of the present application;

[0028] Figure 4 It is a schematic structural diagram of another agent generation device provided by the embodiment of the present application. Detailed Embodiments

[0029] The following will describe in detail the agent generation method, device, storage medium and program product provided by the embodiments of the present application with reference to the drawings.

[0030] As used in the description of the present application, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0031] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0032] As used in this application, "and / or" includes any one of the two methods or the simultaneous use of both methods.

[0033] In the description of this application, unless otherwise specified, the meaning of "a plurality" refers to two or more.

[0034] At present, with the booming development of artificial intelligence, as a key field, the intelligent agent construction technology is constantly innovating. However, there are many bottlenecks in the existing intelligent agent construction methods, which seriously restrict its wide application and development. Currently, the intelligent agent construction technology based on large models mainly relies on programming capabilities or predefined frameworks, which makes it extremely difficult for ordinary users to construct intelligent agents. Based on this, the embodiments of the present application provide an intelligent agent generation method, device, storage medium and program product. By responding to the intelligent agent generation information input by the user, an intelligent agent generation instruction is generated, and the intent of the intelligent agent generation instruction is parsed to obtain the atomic tasks of the intelligent agent; and based on the atomic tasks of the intelligent agent and the pre-trained template library, an intelligent agent is generated. Compared with the prior art, the requirements for operators in programming and other fields are lower during the process of generating intelligent agents, and it enables users to generate intelligent agents that meet their needs more simply and conveniently.

[0035] Exemplarily, as Figure 1 shown, it is a schematic diagram of the architecture of an intelligent agent generation system provided by an embodiment of the present disclosure, including: an input layer, a multimodal intent parsing engine, an intelligent agent generator, a measurement and optimization layer.

[0036] Among them, the input layer supports multimodal information input methods. For example, input methods such as voice, text, and pictures. During voice input, based on automatic speech recognition (ASR) technology, the user's voice commands are accurately converted into text format; for text input, the natural language commands input by the user are directly received and processed. For example, for the text of "Develop an intelligent agent that automatically analyzes sales data and generates weekly reports". After receiving the text, the input layer will perform preliminary format conversion and preprocessing on it to ensure that the commands can be smoothly recognized and processed by subsequent system modules.

[0037] The multimodal intent parsing engine: includes hybrid semantic understanding and intent conflict resolution.

[0038] Hybrid semantic understanding, deployed within the hybrid semantic understanding module of the multimodal intent parsing engine, integrates adaptive semantic understanding technologies, such as transfer learning and data cleaning, with a rich domain knowledge graph. This fusion allows the module to perform in-depth semantic analysis of natural language input from users. For example, for the complex requirement of "automatically analyzing sales data and generating weekly reports," it can precisely decompose the task into multiple atomic tasks, such as data capture, data analysis, and report generation, providing clear task planning for subsequent intelligent agent construction.

[0039] Intent conflict resolution can be deployed within the multimodal intent parsing engine's intent conflict resolution module. This module comes into play when user descriptions are ambiguous, such as "automatically replying to emails and analyzing sentiment." Leveraging a large model, the intent conflict resolution module generates multiple possible candidate solutions and presents them to the user for selection. This approach accurately identifies the user's true intent and avoids agent-building errors caused by misunderstandings.

[0040] The agent generator includes dynamic code generation and data flow orchestration.

[0041] Dynamic code generation can be deployed in the Agent Generator's code generation module to quickly call appropriate templates from a pre-trained template library based on intent analysis results. This template library includes a rich variety of templates, including Python functions and RPA scripts, to meet code generation needs in various scenarios. After calling a template, advanced static analysis tools perform comprehensive syntax and logic verification on the generated code to ensure code accuracy, reliability, and stability.

[0042] Data flow orchestration, deployed in the Data Flow Orchestration module of the Agent Builder, uses Directed Acyclic Graph (DAG) technology to clearly define the dependencies between the various modules of the Agent. For example, when processing email-related tasks, the process sequence of "email input → sentiment analysis → reply generation" can be clearly defined. This technology not only intuitively displays the relationships between modules but also supports parallel task scheduling, significantly improving system efficiency and resource utilization.

[0043] Testing and optimization layer, including simulation sandbox environment and feedback loop.

[0044] The simulated sandbox environment closely simulates the real-world scenarios in which intelligent agents operate. When running an intelligent agent in this environment, the system detects various anomalies in real time, such as API call failures and data transmission errors. Once detected, the system quickly generates targeted remediation suggestions, such as replacing alternate interfaces and adjusting data transmission strategies, to ensure the stability of the intelligent agent during actual operation.

[0045] The feedback closed-loop can be deployed on the feedback closed-loop module, providing strong support for the continuous optimization of the intelligent agent. During the process of using the intelligent agent, the user can annotate and provide feedback on the running results, such as "the emotion analysis is inaccurate". After receiving the user feedback, the system will automatically start the optimization program to adjust the model parameters or code logic, such as adjusting the weight of the emotion dictionary and optimizing the analysis algorithm, so as to continuously improve the performance of the intelligent agent.

[0046] Exemplarily, the intelligent agent generation device that executes the intelligent agent generation method provided in the embodiments of the present application can be an electronic device. The electronic device can be a mobile phone, a tablet computer, a desktop computer, a laptop computer, a handheld computer, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, as well as a cellular phone, a personal digital assistant (PDA), an augmented reality (AR) / virtual reality (VR) device. The embodiments of the present application do not impose special restrictions on the specific form of the electronic device. It can perform human-computer interaction with the user through one or more of a keyboard, a touchpad, a touch screen, a remote control, voice interaction, or a handwriting device.

[0047] The intelligent agent generation method will be specifically introduced below.

[0048] As Figure 2 shown, the intelligent agent generation method provided in the embodiments of the present application may include S201 - S203.

[0049] S201. Generate an intelligent agent generation instruction in response to the intelligent agent generation information input by the user.

[0050] In a possible implementation manner, the intelligent agent generation information can be used to represent information such as the functions that the user expects the generated intelligent agent to have or the user's requirements.

[0051] In a possible implementation manner, the intelligent agent generation information is one or more of text format, voice format, and picture format.

[0052] In a possible implementation, in S201, in response to the agent generation information input by the user, an agent generation instruction is generated, which is implemented as follows: a conversion operation is performed on the agent generation information in a non-text format input by the user to obtain the agent generation information in a text format; based on the converted agent generation information in a text format, an agent generation instruction is generated; or, based on the converted agent generation information in a text format and the agent generation information in a text format input by the user, an agent generation instruction is generated. In this way, multi-modal input of agent generation information can be achieved, which can facilitate the user to perform the agent generation process based on various formats of input, improving the user experience. In addition, by converting the agent generation information in a non-text format into the agent generation information in a text format to generate an agent generation instruction, it is simpler and more efficient compared to directly obtaining the agent generation instruction through the agent generation information in a non-text format.

[0053] S202. Perform intent parsing on the agent generation instruction to obtain the atomic tasks of the agent.

[0054] Among them, the atomic tasks of the agent are used to indicate the functions that the user expects the agent to have.

[0055] In a possible implementation, in S202, perform intent parsing on the agent generation instruction to obtain the atomic tasks of the agent, which is implemented as follows: perform semantic parsing on the agent generation instruction to obtain clear semantic information, and based on the clear semantic information, generate atomic tasks.

[0056] In a possible implementation, perform semantic parsing on the agent generation instruction to obtain clear semantic information, and based on the clear semantic information, generate atomic tasks, including: perform semantic parsing on the agent generation instruction to obtain clear semantic information and fuzzy semantic information; perform intent conflict resolution processing on the fuzzy semantic information to obtain candidate semantic information; display the candidate semantic information, and determine the target semantic information in response to the user's selection operation; based on the target semantic information and the clear semantic information, generate atomic tasks.

[0057] It should be noted that in some cases, the agent generation instruction input by the user may not be clear enough and there may be fuzzy or ambiguous situations. At this time, intent conflict resolution processing can be performed on the fuzzy semantic information to obtain clear candidate semantic information that the user may want to express. By displaying the candidate semantic information, the user can perform the selection of the candidate semantic information. In this way, it can assist the user to perform agent generation and improve the user experience. In addition, by generating atomic tasks based on the target semantic information selected by the user and the clear semantic information, it can ensure that the atomic tasks can more reliably and accurately reflect the functions expected by the user.

[0058] S203. Generate an agent based on the atomic tasks of the agent and the pre-trained template library.

[0059] In a possible implementation, S203, generating an agent based on the atomic tasks of the agent and the pre-trained template library includes: selecting, based on the atomic tasks, the templates corresponding to the atomic tasks from the pre-trained template library; generating, based on the templates corresponding to the atomic tasks, the modules corresponding to the atomic tasks; the modules corresponding to the atomic tasks being a set of codes corresponding to the atomic tasks; determining the dependency relationships of the modules corresponding to the atomic tasks based on the dependency relationships between the atomic tasks; determining the execution order of the modules corresponding to the atomic tasks based on the sequential constraints between the atomic tasks; and generating an agent based on the dependency relationships of the modules corresponding to the atomic tasks and the execution order of the modules corresponding to the atomic tasks.

[0060] It should be understood that selecting the templates corresponding to the atomic tasks from the pre-trained template library and generating the modules corresponding to the atomic tasks based on the templates can quickly and accurately obtain the code modules that can implement the functions corresponding to the atomic tasks. After determining the dependency relationships of the modules corresponding to the atomic tasks based on the dependency relationships between the atomic tasks, the reliability, performance, and maintainability of the generated agent can be ensured. Determining the execution order of the modules corresponding to the atomic tasks based on the sequential constraints between the atomic tasks can ensure that the modules corresponding to the atomic tasks can be executed in a reasonable order, ensuring the reliability of the generated agent.

[0061] In a possible implementation, the codes in the modules corresponding to the atomic tasks can be verified for syntax logic through a static analysis tool to ensure the accuracy, reliability, and stability of the code operation.

[0062] It should be noted that by generating an agent generation instruction in response to the agent generation information input by the user, parsing the execution intention of the agent generation instruction to obtain the atomic tasks of the agent; and generating an agent based on the atomic tasks of the agent and the pre-trained template library. Compared with the prior art, the requirements for operators in programming and other fields are lower during the process of generating an agent, enabling users to generate agents that meet their needs more simply and conveniently.

[0063] In some embodiments, the method further includes: testing the agent in a simulated sandbox environment to obtain a test result; and in the case where an abnormal situation exists in the test result, updating the agent based on the abnormal situation. In this way, by testing and updating the agent, the possibility of the agent having abnormal situations can be reduced, improving the user experience.

[0064] In some embodiments, the method further includes: during the user's use of the agent, in response to the user's evaluation operation on the operation result of the agent, collecting feedback information; and updating the agent based on the feedback information. In this way, targeted adjustment of the agent can be achieved based on the user's actual evaluation, improving the user experience.

[0065] The following is an exemplary description of the agent generation method provided in the embodiments of the present application:

[0066] Example 1: An example of general agent generation steps is shown, including:

[0067] User input: The user can input instructions for constructing an agent to the system in a voice or text manner according to their own needs. For example, input "Create an agent that monitors the prices of competing products daily and issues warnings" to clearly express their needs.

[0068] Intent parsing: The instructions input by the user first enter the multimodal intent parsing engine. In this engine, the hybrid semantic understanding module and the intent conflict resolution module work together. The hybrid semantic understanding module combines the domain knowledge graph to deeply parse the user's instructions; the intent conflict resolution module processes possible ambiguous descriptions. Through the joint action of these two modules, the atomic tasks required by the agent are determined, such as a crawler module (responsible for scraping competing product price data), a data analysis module (analyzing and processing price data), and a warning module (issuing an alarm when the price shows abnormal fluctuations).

[0069] Code generation and orchestration: The agent generator selects appropriate templates from the pre-trained template library according to the results of intent parsing and generates the code for each module. For example, generate code with anti-antispider strategies for the crawler module, generate code for calling the statistical library for the data analysis module, and generate code integrating email / sms APIs for the warning module. Then, use DAG technology to perform data flow orchestration on the generated code modules to determine the execution order and dependency relationships of each module, ensuring that the various modules of the agent can work together.

[0070] Testing and running: The generated agent will be tested and run in a simulated sandbox environment. For example, in the scenario of monitoring competing product prices, the simulated sandbox environment will simulate the real environment of an e-commerce platform to verify the stability of the crawler module. If an abnormality is found during the test, such as the crawler being restricted by the anti-spider mechanism, the simulated sandbox environment will capture the abnormality in a timely manner and generate repair suggestions according to preset rules, such as automatically switching the proxy IP, to ensure that the agent can run normally.

[0071] Optimization Feedback: During the process of using the agent, users can provide feedback on its operation results. For example, if the user discovers that "the warning delay is too high", the system will locate the problem bottleneck through log analysis, such as slow API response. Then, based on the analysis results, the system will replace it with a high-performance interface and update the code library in a timely manner, thereby optimizing the performance of the agent and enhancing the user experience.

[0072] Example 2: Illustrates the process of generating an agent for e-commerce competitor price monitoring, including:

[0073] Requirement Submission: The user inputs "Create an agent that monitors competitor prices daily and issues warnings" to the system, clearly expressing the need for e-commerce competitor price monitoring and warning.

[0074] System Processing: After the input layer receives the user's requirement, it passes it to the multi-modal intent parsing engine. This engine disassembles the requirement into tasks for crawlers, data analysis, and warning modules. The agent generator calls the corresponding templates in the pre-trained template library according to the task requirements to generate code, and uses DAG technology to orchestrate the data flow. The generated agent is tested in the test layer by simulating an e-commerce environment. If there are problems with the crawler, such as being restricted by an anti-crawler mechanism, the system will automatically switch the proxy IP to ensure the stable operation of the agent.

[0075] Agent Process Presentation, the agent process generated after voice or text input supports custom modification. Some agent processes with inaccurate understanding or incorrect configuration can be manually modified.

[0076] Result Presentation: The agent after testing and optimization can stably monitor competitor prices daily. When the competitor price fluctuates beyond the set threshold, the agent can send warnings to the user in a timely manner via email or text message, helping the user promptly grasp market dynamics and make corresponding decisions.

[0077] The embodiments of the present application can divide the functional modules of the agent generation device, electronic device, etc. according to the above method examples. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiments of the present application is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0078] In the case of dividing each functional module corresponding to each function, Figure 3 Shows a possible structural schematic diagram of the agent generation device involved in the above embodiments, as Figure 3 shown, the agent generation device may include: a generation module 301, a processing module 302.

[0079] A generation module 301 is configured to generate an agent generation instruction in response to agent generation information input by a user; a processing module 302 is configured to perform intent parsing on the agent generation instruction to obtain atomic tasks of the agent; the atomic tasks of the agent are used to indicate functions that the user expects the agent to have; the generation module 301 is further configured to generate an agent based on the atomic tasks of the agent and a pre-trained template library.

[0080] In one implementation, the processing module 302 is specifically configured to perform semantic parsing on the agent generation instruction to obtain clear semantic information, and generate atomic tasks based on the clear semantic information.

[0081] In one implementation, the processing module 302 is further specifically configured to perform semantic parsing on the agent generation instruction to obtain clear semantic information and fuzzy semantic information; the processing module 302 is further specifically configured to perform intent conflict resolution processing on the fuzzy semantic information to obtain candidate semantic information; the processing module 302 is further specifically configured to display the candidate semantic information and determine target semantic information in response to a user selection operation; the generation module 301 is specifically configured to generate atomic tasks based on the target semantic information and the clear semantic information.

[0082] In one implementation, the processing module 302 is further configured to select a template corresponding to the atomic task from the pre-trained template library based on the atomic task; the generation module 301 is further configured to generate a module corresponding to the atomic task based on the template corresponding to the atomic task; the module corresponding to the atomic task is a code set corresponding to the atomic task; the processing module 302 is further configured to determine the dependency relationship of the module corresponding to the atomic task based on the dependency relationship between the atomic tasks; the processing module 302 is further configured to determine the execution order of the module corresponding to the atomic task based on the sequential constraint between the atomic tasks; the generation module 301 is further configured to generate an agent based on the dependency relationship of the module corresponding to the atomic task and the execution order of the module corresponding to the atomic task.

[0083] In one implementation, the processing module 302 is further configured to test the agent in a simulated sandbox environment to obtain a test result; the processing module 302 is further configured to update the agent based on the abnormal situation in the case where an abnormal situation exists in the test result.

[0084] In one implementation, the processing module 302 is further configured to collect feedback information in response to an evaluation operation of the user on the running result of the agent during the user's use of the agent; the processing module 302 is further configured to update the agent based on the feedback information.

[0085] In one implementation, the information generated by the agent is in one or more of the text format, voice format, and picture format; the processing module 302 is further configured to perform a conversion operation on the agent-generated information in a non-text format input by the user to obtain the agent-generated information in the text format; the generating module 301 is further configured to generate an agent generation instruction based on the agent-generated information in the text format obtained by the conversion; or, the generating module 301 is further configured to generate an agent generation instruction based on the agent-generated information in the text format obtained by the conversion and the agent-generated information in the text format input by the user.

[0086] In the case of adopting an integrated unit, Figure 4 FIG. shows a possible structural diagram of the agent generation device involved in the above embodiment. As Figure 4 shown, the agent generation device may include: a processing module 401 and a communication module 402. The processing module 401 may be configured to control and manage the operations of the agent generation device. The communication module 402 may be configured to support the communication of the agent generation device with other entities. Optionally, as Figure 4 shown, the agent generation device may further include a storage module 403 for storing the program code and data of the agent generation device.

[0087] Among them, the processing module 401 may be a processor or a controller. The communication module 402 may be a transceiver, a transceiver circuit, or a communication interface, etc. The storage module 403 may be a memory.

[0088] Among them, when the processing module 401 is a processor, the communication module 402 is a transceiver, and the storage module 403 is a memory, the processor, the transceiver, and the memory may be connected by a bus. The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc.

[0089] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0090] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0091] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0092] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more integrated media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0094] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An agent generation method, characterized in that, The method includes: Generating an agent generation instruction in response to agent generation information input by a user; Performing intent parsing on the agent generation instruction to obtain atomic tasks of the agent; the atomic tasks of the agent are used to indicate functions that the user expects the agent to have; Generating the agent based on the atomic tasks of the agent and a pre-trained template library.

2. The method according to claim 1, wherein The performing intent parsing on the agent generation instruction to obtain atomic tasks of the agent includes: Performing semantic parsing on the agent generation instruction to obtain clear semantic information, and generating the atomic tasks based on the clear semantic information.

3. The method according to claim 2, wherein The performing semantic parsing on the agent generation instruction to obtain clear semantic information, and generating the atomic tasks based on the clear semantic information includes: Performing semantic parsing on the agent generation instruction to obtain clear semantic information and fuzzy semantic information; performing intent conflict resolution processing on the fuzzy semantic information to obtain candidate semantic information; displaying the candidate semantic information, and determining target semantic information in response to a user selection operation; generating the atomic tasks based on the target semantic information and the clear semantic information.

4. The method according to claim 1, characterized in that The generating the agent based on the atomic tasks of the agent and a pre-trained template library includes: Selecting a template corresponding to the atomic task from the pre-trained template library based on the atomic task; Generating a module corresponding to the atomic task based on the template corresponding to the atomic task; the module corresponding to the atomic task is a code set corresponding to the atomic task; Determining a dependency relationship of the module corresponding to the atomic task based on a dependency relationship between the atomic tasks; Determining an execution order of the module corresponding to the atomic task based on an order constraint between the atomic tasks; Generating the agent based on the dependency relationship of the module corresponding to the atomic task and the execution order of the module corresponding to the atomic task.

5. The method according to claim 1, wherein The method further includes: Testing the agent in a simulated sandbox environment to obtain a test result; In the case where an abnormal situation exists in the test result, performing an update on the agent based on the abnormal situation.

6. The method according to claim 1, wherein The method further includes: During the process of the user using the agent, collecting feedback information in response to an evaluation operation of the user on the running result of the agent; Updating the agent based on the feedback information.

7. The method according to claim 1, characterized in that, The agent generation information is one or more of text format, voice format, and picture format; The generating an agent generation instruction in response to agent generation information input by a user includes: Performing a conversion operation on the agent generation information in a non-text format input by the user to obtain agent generation information in text format; Generating an agent generation instruction based on the agent generation information in text format obtained by conversion; or generating an agent generation instruction based on the agent generation information in text format obtained by conversion and the agent generation information in text format input by the user.

8. An agent generation device, characterized in that, It includes: A generation module and a processing module; The generation module is used to generate an agent generation instruction in response to agent generation information input by a user; A processing module, configured to parse the execution intention of the instructions generated by the agent to obtain the atomic tasks of the agent; the atomic tasks of the agent are used to indicate the functions that the user expects the agent to possess; A generation module, further configured to generate the agent based on the atomic tasks of the agent and a pre-trained template library.

9. An agent generation device, characterized in that, The agent generation device includes: A processor; A memory configured to store executable instructions of the processor; Wherein, the processor is configured to execute the instructions to implement the method according to any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions in the computer-readable storage medium are executed by the agent generation device, the agent generation device is enabled to execute the method according to any one of claims 1-7.

11. A computer program product, characterized in that, The computer program product includes computing technology program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1-7 is implemented.

Images